Steven V. Miller

Every Proper Noun is a Case of a Common Noun; Every Case is a Case of Something More General

Thu, 19 Mar 2026 00:00:00 +0000

: Europe banned (or has 'tried' to ban) RT in light of Russia's invasion of Ukraine. There are proper nouns everywhere. What's the common noun, then? (Mladen Antonov / AFP via Getty Images file).

This is a (hopefully) quick hitter based on stuff that I’ve been repeating to students for a while and have been meaning to flesh out a little bit in a blog post.

One challenge I’ve encountered for students since I’ve been here is trying to get them to see more general patterns or concepts. It’s not something I remember having to think about in a past life in the United States. In that past life, I had a bit more agency over the curriculum, could structure topics more generally, and encourage students to think of examples. Students here instead express a passing interest in some current event or something that happened and, gulp, want to shoehorn that in one of the big -isms even when I don’t think it’s a good idea. I keep expressing a confusion as to what the point is and, importantly, what’s the question. I end up circling back to two things, both related, that I keep having to repeat. 1) Every proper noun is a case of a common noun. 2) Every case is a case of something more general.

Here’s one quite common example I keep encountering. On March 2, 2022, the European Council issued an “urgent suspension” of RT in light of Russia’s ongoing invasion of Ukraine. My students keep saying this is a “ban”, though I don’t quite understand how that’s ever been the case. No matter, every semester I encounter a student at the BA-level saying they want to write about this. Proper nouns are everywhere in the paragraph’s second sentence (e.g. “European Council”, “RT, “Russia”, “Ukraine”), but there are common nouns lurking underneath that surface that I want the student to identify and bring into greater detail. What are they? The “EU banned RT”. Okay, but that is a specific manifestation of something more general. What is it?

I keep repeating this because my battle-weary hobby horses for students here is pleading with them to read scholarship on a topic and not lean on intro-level textbooks, podcasts, think tank reports, or news reports to do that for them. Toward that end, I need to know to more about a stated interest in something like the EU “ban” on RT and what exactly is the broader thing for which that is a specific and capital-case manifestation. Is “the EU ‘bans’ RT” (the proper nouns[s]) a case of censorship? If so, I want the student immediately consulting the broader academic literature on the politics of censorship and identifying what’s missing in what we know about that common noun (“censorship”) for which the proper noun(s) (“EU ‘bans’ RT”) would be informative. Is this a policy response to so-called “hybrid warfare?” If so, I want the student reading about that to better identify how this thing fits in the bigger picture of things like it. The European Union states may have to deal with this vis-a-vis Russia and even Vietnam has to think about it with respect to China. How do these proper nouns fit with respect to the common noun? To answer that, you need to read more about the common noun.

Likewise, students that express an interest in international conflict express an interest in the ongoing Russia-Ukraine War. Perfectly legitimate interest, but that’s a proper noun. It’s a case. What is its common noun? What is it a case of? How exactly you answer these questions will condition the literature you review that is not just the background information about the case in question. Certainly, the Russia-Ukraine War is a case of an inter-state war. If that’s the more general case for which the Russia-Ukraine War is an exemplar, then there is ample scholarship in the Correlates of War tradition to read. I’m even in one of those books you should definitely read at the BA-level. Is it a case of an “absolute war”, like the kind R. Harrison Wagner would write about? If so, give him a read and follow some of his citations (and those that cite him). Do you intend to think of this as a “great power intervention?” I’m not sure how well that fits, given how Scott Wolford describes it, but you could do worse than reading Wolford’s stuff. He’ll give more focused language to what you see. In the process, he’ll better focus what you want to get out of looking at this case in particular. I’ve said before that students need to go from having topics to having questions. Understanding what your case of interest is a case of and reading scholarship on that will help bring into greater relief what question you ultimately intend to ask.

I try to plead with students to read and read widely and to better contextualize the things they see. They see some event and are interested in it. To get the most out of that interest, the student has to take that proper noun and identify the common noun that informs it. The student has to identify what their case is a case of. From that, read and read widely to better understand what question is important to ask about this case, given what we know (or don’t know) about it more generally.

War and the State: An Appreciation

Tue, 09 Dec 2025 00:00:00 +0000

: War and the State: The Theory of International Politics

I’m going through a phase where I’m re-reading books that shaped me in graduate school. Some of this is because of a professional rut, post-upheaval in the move to Sweden. It’s one way I’m trying to get my groove back. Another is because I commute about 45-55 minutes to the office by way of pendeltåg. This not something I’ve ever had to do, but it’s a fact of life for Europeans. It does mean I have more time on my hands, and my hands get to be on my phone or a book instead of a steering wheel. Of the phone and the book, the book seems like a better option.

There are books that had more of an effect on my intellectual trajectory. My orientation toward international politics is deeply indebted to John Vasquez, especially The War Puzzle. John is my “grand-adviser”, if you will, and still one of the most wonderfully curious people I’ve had the good fortune of knowing in the profession. My framework toward thinking about the nexus of domestic politics and conflict has a clear link to classics from Acemoglu and Robinson (2006) and especially Boix (2003). I came across both because my adviser was citing them for his work on territorial peace. The year was 2006 and 2007, peak years highlighting the incompetence of the Bush Administration and the deleterious consequences of fiscal policies that upwardly redistribute wealth. Neither of those were the point of either book, and neither were the slight references in both books to how groups co-opt/purchase the military as insurances against the redistributive demands of the poor. Yet all inform my broader academic worldview. Some of the discrepancies not neatly implied by minimal models of redistributive demands lead to appeals to Fromm (1941).¹ You should read Escape From Freedom too.

I only discovered Wagner’s (2007) War and the State much later in graduate school, after comprehensives. It has since become one of my absolute favorite books in international relations. I still return to passages of it and just completed a full re-read of it. We live in a world that we think has become more “realist”, but we are popularizing approaches that were never informative. They were just the center of gravity during the formation of the discipline as we know it. We are giving increasing intellectual real estate to some people of that tradition that have never had an answer to questions we should care to ask. Thus, I think appreciating R. Harrison Wagner’s work is critical today given what the general public consumes, thinks is happening, and what people (students) expect they need to know. War and the State is a necessary read and an important corrective to what students are typically asked to engage with a straight face. Students should read that book in its entirety and complement it with the 2010 symposium in International Theory along with some of his other stuff.

A Synopsis

If nothing else, students should read Chapter 1. Herein, Wagner retraces the intellectual legacy of the largest -isms, prominently so-called “realism” as the center of these -isms. All, prominently “realism”, are offered as explanations of international conflict largely on the basis of arguments with premises that do not imply the hypotheses offered. Questions aren’t answered as much as they are begged. The foundation of the discipline owes to a need to separate the war that happens between states as happening under conditions of “anarchy” that are not applicable to the “hierarchy” of domestic politics. But this obscures more than it reveals. War has never required sovereign states, just groups capable of killing on scale. States might be unique institutions born from war for the cause of killing on scale, but they are neither necessary nor sufficient for war. States are not simply hapless agents under an exogenously given anarchy; they are the organizational solution to a world without a sovereign. This tradition is too eager to misread what it says it reads (c.f. Hobbes and Rousseau), misuse tools (e.g. stag hunt, prisoner’s dilemma) to evaluate perspectives/”approaches” masquerading as theories (see: the relative/absolute gains debate), poorly define what must be properly stated (e.g. “what was bipolarity?”), and assume what needs to be explained (i.e. why war happens at all).

Chapter 2 continues the theme of misreading what it purports to have read (c.f. Herz and Carr’s treatment of idealism/utopianism) by situating realism in a tradition of raison d’état. The treatment here is largely historical and philosophical. Students may find this informative for how much realism takes from Rousseau than Hobbes, per se, how Kant is not that different from Hobbes or Rousseau in how they think about political order and international politics, and how much you’re probably thinking of Charles-Irénée Castel de Saint-Pierre even if you don’t know it yet. However, it concludes by emphasizing international relations arguments that start with a world of sovereign states to explain violence are starting in the wrong place. Sovereign states are outcomes of an early modern (European) period in which people were free to organize themselves in any fashion in order to profit from the use of force. The structure is an outcome of violence, and a hard bifurcation of the domestic from the international won’t get you far. It’s also not the rigid bifurcation discussed in some of these classics from early modern philosophers.

Chapter 3 is as much the argument chapter as any in the book. Here, Wagner brings in the bargaining approach that pervades a lot of the scholarship I like to discuss. Rather than conceptualize a world of sovereign states under systemic anarchy, Wagner conceptualizes a world where states emerge as institutions of roving predators in the protection business selling protection from other predators and from themselves. Violence is ideally the thing to avoid, but violence is the mechanism by which individuals/predators manipulate expectations about the disagreement outcome in a bargaining situation (i.e. what actors would expect to happen if they could not reach an agreement). Thus we emphasize what we know from Carl von Clausewitz. Violence (war) is politics (bargaining) by other means. The Weberian state is the institutional agreement creating that effective monopoly on force. The monopoly is granted by members of the organization who consent and cooperate, in some form, in granting it.

Chapters 4, 5, and 6 extend this framework into an exploration of the conditions for which competing predators may make agreements, when they are enforceable, what capability aggregation implies for questions of opportunistic predation or collective defense (c.f. balancing/bandwagoning), pre-emptive war, and, of course, what this means for different wrinkles in the type of state in question. Perhaps democracies are a class of territorially-satisfied former predators in which leaders are agents and not principals. Even here, hypotheses we put under the democratic peace umbrella come with caveats suggested by the bargaining framework. Great Britain and the Weimar Republic did not fight a war, but the Weimar Republic made Nazi Germany possible. The democracy cannot credibly commit to not degenerating into a non-democracy.² There is no reason to dismiss, necessarily, that a world of territorially-satisfied democratic states can’t find something else to fight about. It’s sometimes hard to wrap our head around how public goods are supplied at all given the collective action required in providing them (let alone efficiently). That has implications for questions of war and what it means for democracy.

For Students

Wagner is a boon for students who have only known explanations of international relations in the most abstract terms possible. It is convenient to think of a Hobbesian world where life is “solitary, poor, nasty, brutish, and short” absent a Leviathan to “overawe them all.” But, that’s the post hoc philosophical justification for the realist worldview that is much more derivative of Rousseau. Waltz’ inspiration is Rousseau, not Hobbes, and an argument that war happens because nothing stops it from happening is not a convincing argument. It sounds cool, but it’s not convincing. The prisoner’s dilemma is a useful way of getting at cooperation problems, but any game in which players might be eliminated in a repeated prisoner’s dilemma is no longer a prisoner’s dilemma. The prisoner’s dilemma is still nice to know, it can only tell you so much. “Ontology” sounds cool but, oh man, read pages 44 and 45 to see Wagner’s treatment of that term and its use in international relations. Mearsheimer’s five assumptions sound provocative and insightful when stated with the authority with which he states them, but no hypothesis is implied by that whatsoever. Ignore the issue of internal consistency, which were complaints from the likes of Bruce Bueno de Mesquita and John Vasquez in some of their critiques of realism. Arguments should be consistent, but they should be valid as well. The prediction must follow from the hypothesis, but those five assumptions get you nowhere. They barely get you out the starting block. You think they do when you’re beginning in the field, but they don’t get you where you want to go. Students would benefit from reading Wagner to better cut through the noise around them.

Like most things for students who are starting out, following the citations reveal some hidden gems. Wagner is citing some interesting things I’d welcome students to explore. Herz (1951) is nice background for some of the classics, as are Claude (1962), Herz (1950), Hoffman (1963), and Kindleberger (1973). I discovered Wrong (1994) through this book, which featured in one of my publications dealing with the specifics of the Hobbesian argument.

His citations go beyond the strictly, boilerplate IR (or stuff that I’ve used for scholarship). Wagner’s footnotes are a treasure trove of information. Kincaid’s (1986) The Rule of the Road is the inspiration at the start of his second chapter, though it is primarily about the history of left-hand or right-hand driving. Pirenne (2001) is an interesting explanation of the transition to feudal Europe as a result of the rise of Islam (and not the Barbarian conquest of Rome). Bensel (1990) has a fascinating historical account for the rise of the modern American administrative state for the cause of war against the separatist South. The Confederates may not have had the advantages of the North, but they too needed to evolve beyond a simple plantation system to mobilize for war against the North. Grossman (2000) offers a nice academic citation toward conceptualizing a state’s leadership as principal or agent, and the implications of that. Tilly (1985) developed an idea of a state as a protection racket, like a mafia, with a logic that echoes in classics of mafias and racketeer gangs like Gambetta (1993), Varese (2001), Volkov (2000). Finally, Berlin (1998) offers an analysis of the contours of slavery through its evolution in North America. Included in it is a consideration of the agency of the enslaved in shaping this evolution, analogous to the bargaining framework Wagner describes. Wagner’s citations offer plenty of avenues to explore, directly related to international relations or indirectly informative of important frameworks for understanding international relations.

One major obstacle to reading War and the State is that it’s a book that must be re-read as much as it’s read. It’s a challenging read even when English is native. Fearon’s contribution to the 2010 symposium makes a related point that the book is often clearer on what it rejects than what it offers in its place. Likewise, I had to re-read the book to better appreciate the framework Wagner offers. But that does come with a slight apprehension about what questions he’s ultimately asking and answering. I won’t know for sure, but I read this book as if they were expanded lecture notes from a class one might have taken with him at the University of Texas. For students, that’s wonderful. Further along the developmental path, you’re looking harder for the red thread. It’s in there, but the reader needs to search for it. The student will definitely benefit from it.

These comments don’t undercut my plea for students to read and wrestle with this book. R. Harrison Wagner is a pedant after my own heart, but also a scholar with valuable contributions to the bargaining framework of war. The pressing questions of international politics are too important to be begged, and yet we give so much time and energy to those who beg the question they poorly state. Wagner will make that much clear, especially when you have to deal with those people making those claims on television or in print. He’ll also give you information to get you closer to the answers. You should read him.

Fromm might be biting off too much from the Freud apple and I’m sure there are psychological correctives to Freud (who was invariably more influential at the time of publication than he is now). Even then, my goodness, the relation of Freud-derivative understandings of sadism and masochism to the petite bourgeoisie in 1930s Germany definitely echo today. ↩
Yes, I’m aware I’m an American. I’m also aware the United States cannot credibly past the next phone call. The U.S.—and by extension the world that operates around it—is in a very precarious situation. ↩

Find Something to Do with the Quality of Government (Cross-Section) Data

Sat, 25 Oct 2025 00:00:00 +0000

: Bernhard Gilliam's 'Protector of our Industries' (1883) is one of many political cartoons of this time decrying the Gilded Age in the United States.

This is yet another tutorial for students in my quantitative methods courses, this time at the BA-level. I sense a kind of panic from students as they approach a final paper in which I ask them to ask their own question and answer it with quantitative methods. They feel they don’t know where to look. I’ve made a few guides available to them. I have an example paper I provide to them where I do a simple analysis of Swedish respondents in the World Values Survey on attitudes toward democracy. I’ve given them some inspiration for the kinds of analysis they can do, and how you can use {WDI} to do these types of papers as well. I also make a copy of the Quality of Government (QoG) data from the Quality of Government Institute available to them. However, I don’t offer them a particular example with these data. I thought I would do that here to hopefully dissuade them from further panic about this assignment and from biting off more they can chew on conflict data. I laud the initiative to do spatial analyses or TSCS models of conflict fatalities, or digging into the correlates of various UN peacekeeping operations, but doing those well requires a lot of guidance about joins, state systems, and the myriad things you can do in {peacesciencer}. This class has just three lectures and seven labs and students have to draw an owl under those constraints. I can only do so much and I can only ask for so much. I don’t want students trying to do too much, and doing it poorly under considerable stress. I’d rather a very simple exercise be done confidently at this stage than see students try to slay a dragon with a fork and dinner plate.

Here are the R packages I’ll be using.

library(tidyverse)
library(stevemisc)
library(stevethemes)
library(modelsummary)

Here’s a table of contents.

(Download) Load the Data, (Download) Read the Codebook
Find Something That Interests You
Look for Some “Causes”
- Getting Some Other Stuff
Some Descriptive Statistics
A Basic Analysis
Very Basic Diagnostics
Conclusion

On with the show…

(Download) Load the Data, (Download) Read the Codebook

Students should download the “standard” data set made available by the Quality of Government Institute. This is an intro-level, one-month-long quantitative methods course and I cannot expect students to get into the weeds of cross-section time-series (panel) data. Download the “Cross-Section” version of the data in any format and save it to your hard drive.

Load it into R as follows. The file path obviously suggests that I make these available to my students (in Stata’s .dta format). Students should adjust their file path to wherever the data are on their hard drive.

QoG <- haven::read_dta("~/Koofr/teaching/eh1903-ir3/2/data/qog/qog_std_cs_jan24_stata14.dta")
QoG
#> # A tibble: 194 × 1,652
#>    ccode cname     ccode_qog cname_qog ccodealp ccodecow version aii_acc aii_aio
#>    <dbl> <chr>         <dbl> <chr>     <chr>       <dbl> <chr>     <dbl>   <dbl>
#>  1     4 Afghanis…         4 Afghanis… AFG           700 QoGStd…   NA         NA
#>  2     8 Albania           8 Albania   ALB           339 QoGStd…   NA         NA
#>  3    12 Algeria          12 Algeria   DZA           615 QoGStd…    6.25       5
#>  4    20 Andorra          20 Andorra   AND           232 QoGStd…   NA         NA
#>  5    24 Angola           24 Angola    AGO           540 QoGStd…   18.8       10
#>  6    28 Antigua …        28 Antigua … ATG            58 QoGStd…   NA         NA
#>  7    31 Azerbaij…        31 Azerbaij… AZE           373 QoGStd…   NA         NA
#>  8    32 Argentina        32 Argentina ARG           160 QoGStd…   NA         NA
#>  9    36 Australia        36 Australia AUS           900 QoGStd…   NA         NA
#> 10    40 Austria          40 Austria   AUT           305 QoGStd…   NA         NA
#> # ℹ 184 more rows
#> # ℹ 1,643 more variables: aii_cilser <dbl>, aii_elec <dbl>, aii_pubm <dbl>,
#> #   aii_q01 <dbl>, aii_q02 <dbl>, aii_q03 <dbl>, aii_q04 <dbl>, aii_q05 <dbl>,
#> #   aii_q06 <dbl>, aii_q07 <dbl>, aii_q08 <dbl>, aii_q09 <dbl>, aii_q10 <dbl>,
#> #   aii_q11 <dbl>, aii_q12 <dbl>, aii_q13 <dbl>, aii_q14 <dbl>, aii_q17 <dbl>,
#> #   aii_q18 <dbl>, aii_q19 <dbl>, aii_q20 <dbl>, aii_q21 <dbl>, aii_q22 <dbl>,
#> #   aii_q23 <dbl>, aii_q24 <dbl>, aii_q25 <dbl>, aii_q26 <dbl>, …

The data I have are from 2024 and have 194 rows with 1,652 columns. The number of rows (and the information displayed in the first few columns) are clear signals this is a cross-section of sovereign states/territorial entities in the world. The staggering number of columns communicates some information about them that might interest us. There really is not a way of proceeding through this exercise without also downloading the codebook. Have it open and be prepared to search it (Ctrl-F) for information you might find interesting. However, that would require having an idea of something you may want to locate, because it interests you.

Find Something That Interests You

This might be the biggest challenge I’ve encountered so far in teaching students about quantitative methods. I impress the importance of finding something that interests the student that they may want to better explain. Contingent on the thing in question, you could find a reasonable quantitative measurement of it and fit a statistical model to explain variation in it. Whether students have qualitative or quantitative aims at heart for their thesis (often the former), there’s no honest consideration of a body of scholarship on the topic without also giving consideration to those who do what the student wants to do, but quantitatively. That doesn’t make it any the less difficult to get started on this and it is more difficult when students still don’t have an idea what they want to do for their thesis in the next course. Often, the student has no idea what they want to do and thus cannot proceed from there. This is obviously a problem, because I can’t (and won’t) gives topics of interest to students. Students need to start asking their own questions and answering them.

I don’t have much recourse other than to plead with the student to read, read widely, and embrace a hobby horse that they care about on some level. It’s quite likely that a measurement is available in the Quality of Government data related to this interest. In this case, I elected to do an analysis that’s focused on the current discussion of extreme wealth concentration and its implications for democracy and the current international order. In particular, I’m interested in the language we use about wealth concentration in the “top 1%”. Sure enough, the Quality of Government has an indicator (top_top1_income_share) measuring the pre-tax income share of the top 1% of the population. I found it searching for “top 1%” in the codebook having a hunch that QoG would have these data from Piketty, Saez and company available from the World Inequality Database. They do. See:

: Information about the 'top 1%' variable in Quality of Government's cross-sectional data set.

Observe the codebook helpfully communicates temporal reach and, importantly, cross-sectional coverage. For cross-sectional data with a global aim, we want as many observations as possible (knowing we have just 194 as a maximum). In this case, the codebook is advertising that 88.6% of countries have a measure of the “top 1%”’s share of pre-tax national income.

The measure itself correlates highly with other standard measures of “inequality” even if it may not be an optimum measure of the thing in question when better alternatives are available.¹ No matter, the language of the “top 1%” still resonates in how we link issues of wealth concentration with the current trajectory of democracy in the international system. Modeling its variation and what we know about where extreme wealth concentrates is not a bad exercise for a student at this stage. I would obviously expect more of a literature review than what I’m doing here even if I’m not asking for much on the assignment.

For now, though, let’s see the variation in this measure. First, a summary by way of R’s summary() function.

summary(QoG$top_top1_income_share)
#>    Min. 1st Qu.  Median    Mean 3rd Qu.    Max.    NA's 
#>  0.0706  0.1195  0.1530  0.1570  0.1937  0.3111      22

Then, a histogram. Behind the scenes, the students should also ask for a density plot that better smooths out the histogram’s basic limitation (i.e. the arbitrary number of bins and how that might deceive about issues of bimodality).

QoG %>%
  ggplot(., aes(top_top1_income_share)) +
  theme_steve() +
  geom_histogram(color = 'black', fill = g_c("su_water"), alpha = .8) +
  scale_x_continuous(labels = scales::percent_format()) +
  labs(title = "A Histogram of Wealth Concentration in the Top 1%",
       x = "Percent Income Share in Top 1%", y = "Count",
       subtitle = "The distribution does have the smallest right skew for those states with extreme wealth concentration.",
       caption = "Data: Quality of Government Standard Dataset, 2024 (Cross-Section)")

Having seen both, presenting the histogram, and consulting the basic measures of the range and central tendency of the measure, the rough takeaway is the typical country’s wealth concentration in the top 1% is about 15%. The mean and median don’t disagree with each other much at all though it looks like there is a small tail for a select few countries where as much as 31% of the pre-tax national income is held in the top 1% of the population. filter() and some hackery with row_number() can identify the top five and bottom five in this measure.

QoG %>%
  select(cname, top_top1_income_share) %>%
  arrange(-top_top1_income_share) %>%
  na.omit %>% # cheesing this because of missing data
  filter(row_number() %in% c(1:5, 168:172)) %>%
  mutate(rank = c(1:5, 168:172))
#> # A tibble: 10 × 3
#>    cname                          top_top1_income_share  rank
#>    <chr>                                          <dbl> <int>
#>  1 Mozambique                                    0.311      1
#>  2 Central African Republic (the)                0.310      2
#>  3 Peru                                          0.280      3
#>  4 Mexico                                        0.268      4
#>  5 Angola                                        0.260      5
#>  6 Slovenia                                      0.0793   168
#>  7 Montenegro                                    0.0769   169
#>  8 Netherlands (the)                             0.0767   170
#>  9 Slovakia                                      0.0764   171
#> 10 North Macedonia                               0.0706   172

Look for Some “Causes”

Social science likes to tug on pretty little causal levers when it can, though this isn’t available to us in a cross-sectional data set like this for a class that lasts a month and has three lectures. Thus, we’ll do our best to conceptualize some “causes” of the variation in the “top 1%” income share measure and just hope we don’t get hounded by someone asking for an instrumental variable or difference-in-difference estimator or whatever. The best I can encourage students to do at this stage after identifying a dependent variable of interest is to search for analyses that have used this exact measure as a dependent variable. See what they did. Get some inspiration from that.

This led me to this 2019 article from Huber et al. in Socio-Economic Review where this measure features as a dependent variable. Their main argument is the causes of variation in this measure are largely political and less economic. In particular, they argue we can understand variation in this measure as largely a function of tax rates, the power/prevalence of unions, and the control of government by the political right. This finding emerges even controlling for various economic indicators.

I want to echo their argument about the prevalence of the bargaining power of the labor force as important correlate of this measure. I could spin a ball of yarn that the general flattening of income distributions is a triumph of organized labor in the early-to-mid 20th century and we should expect to see the legacy of that today.² One thinks/hopes this argument would resonate with a Swedish audience that I teach. Let’s see if we can find something related to this by searching for things like “union”, “bargain[ing]”, “worker”, or “labo[u]r”.

Great, I found something.

: Information about worker's rights in Quality of Government's cross-sectional data set.

We can work with this. I wish it had a greater cross-sectional reach (beyond the 63% coverage we have), but it has enough global coverage to proceed. We’re further not limiting ourselves to an analysis of OECD countries. There would be nothing wrong with this, but that is a small population of countries. I don’t want the analysis to be so under-powered and limited in the population it describes.

A basic scatterplot will illustrate what bivariate relationship emerges before further consideration of partial effects in a linear model. We’ll also get an informal warning about the distribution of the worker’s rights measure.

QoG %>%
  select(top_top1_income_share, wef_wr) %>%
  ggplot(., aes(wef_wr, top_top1_income_share)) +
  theme_steve() +
  geom_point() +
  geom_smooth(method = 'lm') +
  scale_y_continuous(labels = scales::percent_format()) +
  labs(title = "The Bivariate Relationship Between Worker's Rights and Wealth Concentration in the Top 1%",
       subtitle = "An intuitively negative relationship emerges in the data, along with a caution about the distribution of the worker's rights index.",
       x = "Worker's RIghts Index [1 = worst, 100 = best]", y = "Percent Income Share in Top 1%",
       caption = "Data: Quality of Government Standard Dataset, 2024 (Cross-Section)")

Getting Some Other Stuff

I will say up front we cannot possibly do everything that the authors did with this data set because some measures will just not be available or will be super limited in its cross-sectional reach. I permit some flexibility in doing an analysis like this for the sake of the final assignment. No matter, we’ll try our best to approximate what Huber et al. did with the following information that’s also available in the QoG data. We have to “control” for the relationship we’re arguing exists between worker’s rights and the wealth concentration measure. In Hubet et al.’s analysis, there’s a convenient frame pitting the economic vs. the political. Let’s consider both.

[bmr_dem]: We really need a measure of democracy of some kind but don’t have the more granular measurements the authors use. We’ll use a dummy indicator by way of Boix et al. (2013). Boix, incidentally, has one of my favorite analyses of the nexus between democracy and redistribution that I have ever read.
[wdi_expedu]: government expenditure on education (as a percentage of GDP). We need some kind of human capital/education measure, and this is the one I chose. Word to the wise (students): prefices of wdi_ generally signal greater cross-sectional coverage as it comes from the World Bank’s DataBank.
[fi_sog]: the QoG data do not appear to have the kind of marginal tax rate variable that I would want, but a search for “tax rate” found this measure. It’s an index from the “Economic Freedom Dataset” on the size of government by way of the Fraser Institute. Higher values signal, among other things, “low marginal tax rates and high income thresholds”. I’ll say this now to get it out of the way; the Fraser Institute is not serious and anything they do should not be taken seriously as a scientific initiative. Even acknowledging that the sources of their financial support are not sufficient to be making these claims, the institute does not seem that interested in beating the allegations. Alas, it’s here and there’s not (I think?) a better one at the moment in the data I have. Perhaps the charitable interpretation from this forfeits a causal effect and asserts it’s just going to find a (positive) partial association between its measure and what these people hope to achieve (i.e. the upward redistribution of wealth). So, there’s that, I guess.
[wdi_gdpcapcon2015]: Gotta have GDP per capita. Will log-transform this.
[wdi_fdiout]: Foreign direct investment, net outflows (as percentage of GDP). This is a case where I needed to search for “foreign direct” and not “FDI” to better find what I wanted.
[wdi_gdpgr]: GDP growth (as an annual percentage). Huber et al. have it included in their paper, or at least a working paper that was easier to obtain.

Let’s get everything we want. Importantly, let’s also get the country name and three-character ISO code for potential diagnostic purposes. It would also be useful in case we wanted to add some World Bank group indicators to these data.

QoG %>%
  select(cname, ccodealp,
         top_top1_income_share, wef_wr,
         wdi_expedu, fi_sog, bmr_dem,
         wdi_gdpcapcon2015, wdi_fdiout, wdi_gdpgr) %>%
  # go ahead and do this now...
  mutate(ln_gdppc = log(wdi_gdpcapcon2015)) %>%
  # A nuclear option to get complete cases. Use with caution.
  # Do so only if your data are fully inclusive of everything you want and
  # doesn't have anything you don't want.
  na.omit -> Data 

Data
#> # A tibble: 105 × 11
#>    cname         ccodealp top_top1_income_share wef_wr wdi_expedu fi_sog bmr_dem
#>    <chr>         <chr>                    <dbl>  <dbl>      <dbl>  <dbl>   <dbl>
#>  1 Albania       ALB                     0.0800     79       3.34   7.83       1
#>  2 Algeria       DZA                     0.0991     57       7.04   4.36       0
#>  3 Angola        AGO                     0.260      71       2.74   7.58       0
#>  4 Argentina     ARG                     0.138      73       5.28   6.09       1
#>  5 Australia     AUS                     0.0991     82       5.61   6.03       1
#>  6 Bahrain       BHR                     0.243      63       2.55   7.05       0
#>  7 Bangladesh    BGD                     0.162      61       1.77   8.33       0
#>  8 Belgium       BEL                     0.0831     89       6.81   4.32       1
#>  9 Bolivia (Plu… BOL                     0.198      72       8.44   5.95       1
#> 10 Botswana      BWA                     0.227      71       8.06   6.54       1
#> # ℹ 95 more rows
#> # ℹ 4 more variables: wdi_gdpcapcon2015 <dbl>, wdi_fdiout <dbl>,
#> #   wdi_gdpgr <dbl>, ln_gdppc <dbl>

Cool? Cool.

Some Descriptive Statistics

From here, you can do some basic descriptive statistics like the kind I introduce to students in my class. This will lean on {modelsummary} and, in particular, the datasummary_skim() function it has. This is my go-to for teaching students how to effortlessly create a descriptive statistics table that they can copy-paste into their Word document. I wish I had time to teach them R Markdown, but that’s a different conversation. Doing this in R Markdown would also make the title appear as a caption. Students can manually insert that themselves. However, it won’t show for my purposes given my relatively archaic setup.

Data %>%
  select(-cname, -ccodealp, -wdi_gdpcapcon2015) %>%
  # omit country name/info and the un-logged GDP per capita
  datasummary_skim(., output = "kableExtra",
                   title = "Descriptive Statistics for Our Analysis",
                   align = c("c")

	Unique	Mean	SD	Min	Median	Max
Top 1% income share	96	0.2	0.0	0.1	0.2	0.3
Workers' rights. 1-100 (best)	39	73.1	18.8	3.0	73.0	100.0
Government expenditure on education, total (% of GDP)	105	4.6	1.6	1.6	4.4	9.3
Size of Government: Expenditures, Taxes and Enterprises (current)	105	6.7	1.1	4.3	6.7	9.2
Dichotomous democracy measure	2	0.7	0.5	0.0	1.0	1.0
Foreign direct investment, net outflows (% of GDP)	102	1.0	11.2	−31.8	0.3	104.7
GDP growth (annual %)	105	−4.1	4.5	−21.4	−3.7	6.2
GDP per capita (constant 2015 US$)	105	8.9	1.4	5.6	8.8	11.3

One thing I’ve discovered that I think(?) is a relatively new feature of datasummary_skim() in {modelsummary} is its ability to process variable labels. It really isn’t worth belaboring these things in the limited time I have with students. Students can also just as well fix these things in their Word document. For now, it might be nice to know you can find where these things are stored in the data frame. You might also find an Easter egg from the Quality of Government Institute in the data referencing a Spiderman movie.

str(Data)
#> tibble [105 × 11] (S3: tbl_df/tbl/data.frame)
#>  $ cname                : chr [1:105] "Albania" "Algeria" "Angola" "Argentina" ...
#>   ..- attr(*, "label")= chr "Country name ISO"
#>   ..- attr(*, "format.stata")= chr "%58s"
#>  $ ccodealp             : chr [1:105] "ALB" "DZA" "AGO" "ARG" ...
#>   ..- attr(*, "label")= chr "3-letter Country Code"
#>   ..- attr(*, "format.stata")= chr "%3s"
#>  $ top_top1_income_share: num [1:105] 0.08 0.0991 0.2598 0.1382 0.0991 ...
#>   ..- attr(*, "label")= chr "Top 1% income share"
#>   ..- attr(*, "format.stata")= chr "%9.0g"
#>  $ wef_wr               : num [1:105] 79 57 71 73 82 63 61 89 72 71 ...
#>   ..- attr(*, "label")= chr "Workers' rights. 1-100 (best)"
#>   ..- attr(*, "format.stata")= chr "%10.0g"
#>  $ wdi_expedu           : num [1:105] 3.34 7.04 2.74 5.28 5.61 ...
#>   ..- attr(*, "label")= chr "Government expenditure on education, total (% of GDP)"
#>   ..- attr(*, "format.stata")= chr "%10.0g"
#>  $ fi_sog               : num [1:105] 7.83 4.36 7.58 6.09 6.03 ...
#>   ..- attr(*, "label")= chr "Size of Government: Expenditures, Taxes and Enterprises (current)"
#>   ..- attr(*, "format.stata")= chr "%10.0g"
#>  $ bmr_dem              : num [1:105] 1 0 0 1 1 0 0 1 1 1 ...
#>   ..- attr(*, "label")= chr "Dichotomous democracy measure"
#>   ..- attr(*, "format.stata")= chr "%9.0g"
#>  $ wdi_gdpcapcon2015    : num [1:105] 4419 3874 2347 11341 58116 ...
#>   ..- attr(*, "label")= chr "GDP per capita (constant 2015 US$)"
#>   ..- attr(*, "format.stata")= chr "%10.0g"
#>  $ wdi_fdiout           : num [1:105] 0.33 0.01 0.18 0.305 0.625 ...
#>   ..- attr(*, "label")= chr "Foreign direct investment, net outflows (% of GDP)"
#>   ..- attr(*, "format.stata")= chr "%10.0g"
#>  $ wdi_gdpgr            : num [1:105] -3.3021 -5.1 -5.6382 -9.9432 -0.0509 ...
#>   ..- attr(*, "label")= chr "GDP growth (annual %)"
#>   ..- attr(*, "format.stata")= chr "%10.0g"
#>  $ ln_gdppc             : num [1:105] 8.39 8.26 7.76 9.34 10.97 ...
#>   ..- attr(*, "label")= chr "GDP per capita (constant 2015 US$)"
#>   ..- attr(*, "format.stata")= chr "%10.0g"
#>  - attr(*, "label")= chr "Quality of Government Standard dataset 2024 - Cross-Section"
#>  - attr(*, "notes")= chr [1:2] "\"Everyone keeps telling me how my story is supposed to go. Nah, Imma do my own thing.\"" "1"
#>  - attr(*, "na.action")= 'omit' Named int [1:89] 1 4 6 7 10 11 14 15 17 19 ...
#>   ..- attr(*, "names")= chr [1:89] "1" "4" "6" "7" ...

attr(Data$ln_gdppc, "label")
#> [1] "GDP per capita (constant 2015 US$)"

Let’s overwrite a few things to show how this works. You’ll obviously want to tailor this to your particular data set.

attr(Data$ln_gdppc, "label") <- "Logged GDP per Capita (2015 USD)"
attr(Data$fi_sog, "label") <- "Size of Government (Fraser Institute, which, yeah...)"
attr(Data$wdi_fdiout, "label") <- "FDI, net outflows (% of GDP)"

Now, let’s do it again. Again, I’m aware the title won’t come through formatting to Markdown as I do it but it will in other outputs. Students can also insert it into their Word document manually.

Data %>%
  select(-cname, -ccodealp, -wdi_gdpcapcon2015) %>%
  # omit country name/info and the un-logged GDP per capita
  datasummary_skim(., output = "kableExtra",
                   title = "Descriptive Statistics for Our Analysis",
                   align = c("c")

	Unique	Mean	SD	Min	Median	Max
Top 1% income share	96	0.2	0.0	0.1	0.2	0.3
Workers' rights. 1-100 (best)	39	73.1	18.8	3.0	73.0	100.0
Government expenditure on education, total (% of GDP)	105	4.6	1.6	1.6	4.4	9.3
Size of Government (Fraser Institute, which, yeah...)	105	6.7	1.1	4.3	6.7	9.2
Dichotomous democracy measure	2	0.7	0.5	0.0	1.0	1.0
FDI, net outflows (% of GDP)	102	1.0	11.2	−31.8	0.3	104.7
GDP growth (annual %)	105	−4.1	4.5	−21.4	−3.7	6.2
Logged GDP per Capita (2015 USD)	105	8.9	1.4	5.6	8.8	11.3

The descriptive statistics imply, if not necessarily demonstrate, that the FDI net outflows variable is going to have a gaudy distribution. These distributions typically do, from experience.

A Basic Analysis

You can do a basic linear model and summarize the results. For readability, I’m going to multiply the dependent variable by 100 (i.e. it’s a proportion and students may find percentages easier to understand).

Data %>%
  mutate(dv = top_top1_income_share*100) -> Data

M1 <- lm(dv ~ bmr_dem + wdi_expedu + wef_wr + fi_sog + 
           ln_gdppc + wdi_fdiout + wdi_gdpgr, Data)

modelsummary(list("Model 1" = M1),
             stars = TRUE,
             output = 'kableExtra',
             title = "The Correlates of Wealth Concentration (QoG, 2024)",
             coef_map = c("bmr_dem" = "Democracy Dummy (BMR)",
                          "wdi_expedu" = "Education Expenditures",
                          "wef_wr" = "Worker's Rights Index",
                          "fi_sog" = "Size of Government (Fraser Institute, I know...)",
                          "ln_gdppc" = "Logged (GDP per Capita)",
                          "wdi_fdiout" = "Net FDI Outflows",
                          "wdi_gdpgr" = "GDP Growth (Annual %)",
                          "(Intercept)" = "Intercept"),
             gof_map = c("adj.r.squared", "nobs"))

The Correlates of Wealth Concentration (QoG, 2024)
	Model 1
Democracy Dummy (BMR)	−1.981+
	(1.122)
Education Expenditures	0.381
	(0.327)
Worker's Rights Index	−0.044+
	(0.025)
Size of Government (Fraser Institute, I know...)	1.503**
	(0.474)
Logged (GDP per Capita)	−0.498
	(0.376)
Net FDI Outflows	−0.012
	(0.038)
GDP Growth (Annual %)	−0.173+
	(0.100)
Intercept	12.500*
	(5.860)
R2 Adj.	0.237
Num.Obs.	105
+ p < 0.1, * p < 0.05, p < 0.01, * p < 0.001

The results identify statistically significant effects for the democracy measure, the worker’s rights measure, the size of government variable, and GDP growth. The size of government variable from the Fraser Institute is easily the most precisely estimated effect. The others, by comparison, are significant at only the .10 level. The statistically significant effects we identify are not log-transformed, so we don’t have to sweat that detail. We can note, in the case of our hypothesis, that a one-unit change in the Worker’s Rights Index decreases the percentage of wealth concentration in the top 1% by .04 percentage points. It’s significant at the .10 level (p = .082). The model itself accounts for about 23% of the variation in the wealth concentration measure.

Very Basic Diagnostics

Do some basic diagnostics too. There’s more I’d like students to consider, but there just isn’t time to wring hands about them all. Do the fitted-residual plot, for one. This is the most bang-for-your-buck linear model diagnostic. It can alert you to non-linearity in the model, non-constant error variance, and send a clear signal about discreteness or repeating patterens in the dependent variable.

broom::augment(M1) %>%
  ggplot(.,aes(.fitted, .resid)) +
  geom_point(pch = 21) +
  theme_steve() +
  geom_hline(yintercept = 0, linetype="dashed", color="red") +
  geom_smooth(method = "loess") +
  labs(title = "A Fitted-Residual Plot of Our Model",
       subtitle = "There are a few tail observations and some movement in the smoother, but it's tough to say much from this regarding linearity.",
       x = "Fitted Values", y = "Residuals",
       caption = "I am pleasantly surprised this passed the Breusch-Pagan test.")

My linloess_plot() function might say more about these relationships. This a component-plus-residual plot function that explores the relationship between all independent variables and the residuals, comparing the flat line at 0 with a LOESS smoother. From this, we get an idea that perhaps the size of government variable merits a square term, should we conceptualize about wealth concentration peculiarities at the tails of the spectrum vis-a-vis the middle. That FDI outflow variable is, well, what it is. I knew what to expect and was not surprised.

linloess_plot(M1, se = FALSE, no_dummies = TRUE) +
  theme_steve() +
  labs(title = "A Component-Residual (Lin-LOESS) Plot of Our Model",
       subtitle = "I suspected there'd be weirdness in that FDI variable. The weirdness in the main IV doesn't seem to matter too much for linearity assumptions.",
       x = "Value", y = "Residual")

You can humor the normality assumption (of the residuals) in a few ways. For example, here’s Shapiro-Wilk.

shapiro.test(resid(M1))
#> 
#> 	Shapiro-Wilk normality test
#> 
#> data:  resid(M1)
#> W = 0.98291, p-value = 0.196

Here’s my residual density plot function (rd_plot()). Before practitioners I know pounce on me for teaching students this, I do emphasize this assumption is the least important of OLS’ assumptions. In real world applications, normality is reasonably approximated if never truly achieved for a model with a large number of observations and any peculiarities whatsoever. However, students can easily get the intuition behind this even as I belabor that it’s a conditional assumption that wants to imply something about the dependent variable in question.

rd_plot(M1) +
  theme_steve() +
  labs(title = "A Density Plot of our Model's Residuals",
       subtitle = "We already passed Shapiro-Wilk. In a lot of real world applications, the normality assumption is reasonably approximated",
       x = "Residuals", y = "Density")

Conclusion

Students: find something to do in the QoG data if you want to make your life easier for your final papers. It will involve finding a topic of interest, finding relevant variables included in the data, and being mindful about cross-sectional coverage. This resource is nice for students to learn quantitative methods around a relatively simple (but still information-rich) data set.

Of course this would be a Cato working paper but it’s a fair point to make. ↩
Students of mine, I’m asking for a bit more than this. You can see my examples on the course module. ↩

Simulate Time Series Diagnostics with {sTSD}

Tue, 21 Oct 2025 00:00:00 +0000

: You don't have to treat critical values from non-standard distributions as gospel in this day and age. See them for yourself.

I’m writing this for an advanced quantitative methods class I teach in our Master’s program in the Department of Economic History and International Relations. A department of two somewhat disparate disciplines, the curriculum we have at the Master’s level endeavors to teach students interested in both basic disciplines at a foundation level. That concerns the quantitative methods sequences (in which I feature prominently) and concerns time series topics (which feature more in quantitative economic history and not international relations as much). Thus, moving here meant I needed to teach myself some things I had never needed to learn before as either part of my methods training or as a tool for some project I was doing.

Enter the wonderful world of unit root testing, a common procedure for time series analyses where a vector (time series) does not hover around some constant mean. If this describes your data—and it sure as hell describes just about any series that you might be watching on Yahoo Finance—then a failure to adequately diagnose it could lead you down a path to a slew of errors. You have multiple options for how you might diagnose so-called “non-stationarity” or “unit roots”,¹ but the classic one is the Dickey-Fuller test. There is no shortage of packages available for doing unit root tests in R. The familiar {tseries} will do it, as will {urca} and {aTSA} (among many others, I’m sure). All work for what they’re intended to do, but all will make evident one of the more whimsical/frustrating things about this procedure. It’s a test that produces a p-value, for which there is a null hypothesis, but it references a non-standard distribution whose critical values themselves were once simulated over 50 years ago on a Canadian super computer of its time. The end result leaves the baffled student and the professor teaching himself this procedure imagining themselves in the crowd while Moses returns with the 10 (or 15) commandments. “Interesting. Makes sense? But you’re leaving me wanting to see this for myself.”

When you understand what it’s doing, it really takes no effort to do this yourself. The R package I wrote to do this, {sTSD}, handles the (Augmented) Dickey-Fuller test and related Phillips-Perron and Kwiatkowski et al. (KPSS) procedures in the same way.² It takes an assumed time series you give it and simulates some user-specified number of time series that matches it description, either stationary or non-stationary.

Here are the R packages that will appear in this post.

library(tidyverse)   # for most things...
library(stevethemes) # for my themes/themeing elements
library(sTSD)        # the star of the show
library(ggh4x)       # for a more flexible nested plot element in {ggplot2}

Here’s a table of contents.

What Are These Unit Root Tests Trying to Accomplish?
About That t-Statistic, Though…
What {sTSD} Does
See It For Yourself, with an Applied Example
Conclusion

What Are These Unit Root Tests Trying to Accomplish?

I get the basis of this question from watching an interview with Bradley Efron in 2014, where he discusses the origin of the bootstrap as a question about what its predecessor (the jackknife) was trying to accomplish. Every statistical procedure is trying to accomplish something with some question or problem in mind. The Dickey-Fuller test, its augmented form, and its successors (Phillips-Perron, KPSS) are no different.

We know that non-stationarity poses considerable problems for statistical inference and joint non-stationarity may even lead to problems of spurious regression. What procedure might warn us of these potential issues? Consider the case of the simple random walk, where the observation of y at time point t is a function of its previous value and some random shock/error. We’d express this form as follows.

\[y_{t} = \rho y_{t-1} + e_{t}\]

This coefficient ($\rho$) communicates the kind of memory in the series and has an assumed bound between 0 and 1. If $\rho$ is 0, there is no memory whatsoever in the series and the series itself becomes basic noise over time. If $\rho$ is 1, there is lasting memory in the series. Sometimes likened to a “drunkard’s walk”, the series has a critical path dependence like a drunkard stumbling out of the bar and toward nowhere in particular. There is randomness, but each “step” along the way is maximally dependent on the steps (and previous randomness) before it.

Conceptualizing a way out of this doesn’t take too much effort. It only takes a simple pen and the nearest available piece of paper or napkin. Start by subtracting the previous observation ($y_{t-1}$) from the current observation ($y_t$) and you get the familiar difference term ($\Delta y_t$). If you understand that the previous observation ($y_{t-1}$) is unweighted with a value of 1, you get this derivation and the introduction of a new coefficient ($\gamma$) to summarize this relationship.

\[\begin{aligned} y_{t} &= \rho y_{t-1} + e_{t} \\ y_{t} - y_{t-1} &= \rho y_{t-1} - (1)(y_{t-1}) + e_{t} \\ y_{t} - y_{t-1} &= (\rho - 1) y_{t-1} + e_{t} \\ \Delta y_t &= \gamma y_{t-1} + e_{t} \end{aligned}\]

The formula syntax may not be welcome for beginners, but the principle is pretty straightforward. A maximally autoregressive series ($\rho$ = 1) is, we would say, first-difference stationary on paper. When this is true, $\gamma$ (or $\rho - 1$) is 0 and the previous value should tell you absolutely nothing about the change in the current value from the previous value. Only $e_t$ remains to account for the difference in a series with total information retention.

Simulation by way of fake data shows how this works. After setting a reproducible seed, we’ll create a simple series of 500 time units. Then, we’ll generate random noise (e) and allow the outcome (y) to be the cumulative sum of this random noise. Whereas the cumulative sum maximally weights the previous observation, and by extension all those before it, this is the pure random walk. The first difference of this (d_y) returns us just the noise at the expense of the first observation in the series.

set.seed(8675309) # Jenny, I got your number...
tibble(t = 1:500, # 500 time units
       e = rnorm(500), # 500 random numbers
       y = cumsum(e),  # the random walk, then...
       # the first difference.
       d_y = y - lag(y)) -> fakeWalk

fakeWalk # let's see the data
#> # A tibble: 500 × 4
#>        t       e      y     d_y
#>    <int>   <dbl>  <dbl>   <dbl>
#>  1     1 -0.997  -0.997 NA     
#>  2     2  0.722  -0.275  0.722 
#>  3     3 -0.617  -0.892 -0.617 
#>  4     4  2.03    1.14   2.03  
#>  5     5  1.07    2.20   1.07  
#>  6     6  0.987   3.19   0.987 
#>  7     7  0.0275  3.22   0.0275
#>  8     8  0.673   3.89   0.673 
#>  9     9  0.572   4.46   0.572 
#> 10    10  0.904   5.37   0.904 
#> # ℹ 490 more rows

Here’s what a plot of these data would look like.

What happens when we lose this kind of “memory” in the series? In any instance where $\rho$ < 1, $\gamma$ in the above equation is necessarily negative. Thus, the past observation starts to tell you something about the next period’s change through a correction mechanism or mean reversion. Past values above the mean in $y$ get “corrected” and pull down to the mean of the overall series. Past values below the mean in $y$ get corrected and pull up to the mean of the overall series. Again, referenced to the above formula, $\gamma$ is negative. Referenced to some kind of linear model of this mechanism, the t-statistic you get predicting the current period’s change with the last observed value will come back more negative than it would in the pure random walk. There’s a correction, and we can feel more “confident” predicting such a correction in the presence of even partial memory. Observe in the case where $\gamma$ is -.5 (i.e. $\rho$ in arima.sim() is .5).

tibble(t = 1:500,
       y = as.vector(arima.sim(n = 500, list(ar = 0.5), sd = 1)), 
       d_y = y - lag(y)) -> fakeAR

# Compare the pure random walk...
summary(M1 <- lm(d_y ~ 0 + lag(y), fakeWalk))$coefficients
#>            Estimate  Std. Error     t value  Pr(>|t|)
#> lag(y) 3.886658e-05 0.004867894 0.007984272 0.9936327

# ...with one that has partial memory.
summary(M2 <- lm(d_y ~ 0 + lag(y), fakeAR))$coefficients
#>          Estimate Std. Error   t value     Pr(>|t|)
#> lag(y) -0.4842694 0.03804168 -12.72997 2.457904e-32

The past value tells you nothing of the current value’s change in the random walk, but it tells you a lot in the case where memory is just partial and the series corrects to its mean. That predictive power is communicated in the t-statistic of a simple linear model.

About That t-Statistic, Though…

It would be ideal if that t-statistic of the model were sufficient for making inferential claims, but it isn’t. That is indeed the test statistic of note, but the statistic doesn’t follow the inferential process of the simple linear model because the test is over the residual term and not the raw data of the series. Instead, we need some other set of critical values for making inferential claims about non-stationarity in the time series. Enter the following table, which first appeared in either Fuller (1976) or Dickey (1976). I have Dickey’s (1976) dissertation, if not Fuller’s (1976) textbook, so I’ll present the Dickey (1976) version.

: Critical Values of the Dickey-Fuller Test, by Way of Dickey (1976)

For added context, these values are the product of an extensive Monte Carlo simulation carried out with assistance from a supercomputer at McGill University at the time (“Super Duper”). Assume a standard normal distribution (with a mean of 0 and a standard deviation of 1). Now, prepare (in the table’s case) a vector of length 25, 50, 100, 250, 500, 750, and $\infty$.³ Further assume the pure random walk from the aforementioned standard normal distribution and related data-generating processes for which there is a y-intercept (i.e. a “drift”, in time series parlance) and a linear time trend that increments by 1 across the length of the vector. For each of those vectors corresponding with those data-generating processes, replicate the procedure on these vectors anywhere from 10,000 to 100,000 times. Get the t-statistics for each of those as a kind of distribution of test statistics under the assumption of a non-stationary time series (because they’re random walks). Break those into percentiles corresponding with anywhere from 1% to the 99%. Recall that the test statistics are intended to come back negative for real-world cases, so we should focus our attention on those magic numbers like .05 or .10 corresponding with the left tail.

Enter one of the first frustrating features of this procedure: the null hypothesis. This is one of those procedures. The null hypothesis here is clumsily stated as equivalent to “problem”, in contrast with other diagnostic tests like Breusch-Pagan, Breusch-Godfrey, and Durbin-Watson (in which the null hypothesis is “no problem”). If the test statistic is less than (“more negative than”) one of those critical values of your choice, you can reject the null hypothesis of “problem” (i.e. non-stationarity) and instead accept the alternative hypothesis of “no problem” (i.e. stationarity). If you cannot reject the null hypothesis, you have a problem. However, you’re making a statement of compatibility or incompatibility of the null hypothesis referencing various distributions that are almost assuredly not your own. You’re further making assessments of your ability to reject the null hypothesis against a distribution that is somewhat parked only when the time series you have is not stationary. That’d be the rationale for “problem” as the null hypothesis. If your time series is stationary, the distribution of plausible statistics moves contingent on $\rho$ and how long the series is.

This test is a case where the logic is nifty but the execution has left me a bit wanting ever since I had to start teaching about this procedure. For one, literally every textbook reproduces this exact table from either Fuller’s (1976) textbook or Dickey’s (1976) dissertation. I don’t doubt the output of the model, but its utility is defined by implicit assumptions in a 50-year-old supercomputer and the multiple other assumptions Dickey and Fuller built into the procedure. Two, much like anything involving a t-statistic, it is never offered in relation to your actual data. I will never have a data set of 25, 50, 100, 250, 500, 750, or $\infty$ observations, so what these statistics mean for my time series of 336 observations or 83 observations has to be approximated or interpolated through other means. This makes it kind of biblical, in a way. It’s again similar to Moses coming down on high with ten commandments ad infinitum without any real means to square ten simple dictates with the exigencies of real life. The third complaint isn’t really the fault of anyone in particular, but this function is non-standard. Summary by simulation was the only way to go and Dickey and Fuller do at least provide an honest framework based on simulation that very few people could do 50 years ago. But I can do this now.⁴ All of us have better technology than Dickey and Fuller had 50 years ago. Why not? It would certainly circumvent some of the awkwardness of doing inference by the null hypothesis. It’s non-standard anyway, so why do it this way when I can simulate based on features about my time series instead (rather than the standard normal distribution).

What {sTSD} Does

{sTSD} is born from my frustration with these procedures, and also my affinity for simulating things. sadf_test(), which handles both the Dickey-Fuller and its “augmented” corollary, looks like this.

sadf_test(x, n_lags = NULL, n_sims = 1000, sim_hyp = "nonstationary")

sadf_test() takes a vector of an assumed time series (x). It then asks for some number of simulations (n_sims) you would like to do, with a default of 1,000. Thereafter, it will simulate the user-specified number of simulations from either a pure white noise time series (sim_hyp = 'stationary') or three different time series (sim_hyp = 'nonstationary') where the data are either a pure random walk, a random walk with a drift (y-intercept), or a random walk with a drift and trend (i.e. y-intercept and time trend). It then runs the Dickey-Fuller or its “augmented” version on all those simulated series, contingent on what you provide to the n_lags argument in the function.⁵ It will allow you to assess whether your time series is stationary or non-stationary by comparison to simulated series of the exact length of your series that is known to be stationary or non-stationary in some form.

Let’s do a simple Dickey-Fuller test (n_lags = 0) with just 100 simulations to make this quick, and to explore its basic output. The output will come back as a list with a specialty class provided by the function.

DF1 <- sadf_test(fakeWalk$y, n_lags = 0, n_sims = 100)
class(DF1)
#> [1] "sadf_test"
names(DF1)
#> [1] "stats"      "sims"       "attributes"

The first element, ("stats"), is the test statistics. In order, they are the test statistic for the Dickey-Fuller test with 1) no drift nor trend, 2) drift, no trend, and 3) drift and trend. You can compare with it communicates with the corollary functions in the {urca} and {aTSA} package.

DF1$stats
#>              [,1]
#> [1,]  0.007984272
#> [2,] -0.229766902
#> [3,] -2.472305462

# compare with in {urca}:
attributes(urca:::ur.df(fakeWalk$y, type = "none", lags = 0))$teststat[1]
#> [1] 0.007984272
attributes(urca:::ur.df(fakeWalk$y, type = "drift", lags = 0))$teststat[1]
#> [1] -0.2297669
attributes(urca:::ur.df(fakeWalk$y, type = "trend", lags = 0))$teststat[1]
#> [1] -2.472305

# {aTSA} does all three in one fell swoop, but be mindful it assumes lag of 1 is
# a lag of 0. There is some processing issues it does underneath the hood that
# account for this.
aTSA::adf.test(fakeWalk$y, nlag = 1)
#> Augmented Dickey-Fuller Test 
#> alternative: stationary 
#>  
#> Type 1: no drift no trend 
#>      lag     ADF p.value
#> [1,]   0 0.00798   0.646
#> Type 2: with drift no trend 
#>      lag   ADF p.value
#> [1,]   0 -0.23   0.928
#> Type 3: with drift and trend 
#>      lag   ADF p.value
#> [1,]   0 -2.47   0.377
#> ---- 
#> Note: in fact, p.value = 0.01 means p.value <= 0.01

The last element ("attributes") contains information for post-processing in another function I will introduce later, but let’s take a look at the second element ("sims"). This is a data frame that is always equal to three times the number of simulations you requested. In our case, these would be the first nine of those simulations.

head(DF1$sims, 9)
#>          tau sim                cat
#> 1  0.6575291   1 No Drift, No Trend
#> 2  0.7944876   1    Drift, No Trend
#> 3 -1.0168921   1    Drift and Trend
#> 4 -0.3259828   2 No Drift, No Trend
#> 5  0.8019879   2    Drift, No Trend
#> 6 -3.0584936   2    Drift and Trend
#> 7 -2.0725384   3 No Drift, No Trend
#> 8  2.4074861   3    Drift, No Trend
#> 9 -0.3064146   3    Drift and Trend

Recall that we leaned on the default procedure for a Dickey-Fuller test, which is to assume non-stationarity of some particular form: the pure random walk, the random walk with a drift, and the random walk with a drift and deterministic time trend. For each simulation, we generated a known series of the length of our time series that matches that description we are testing.⁶ Each simulation then has three randomly generated series for which it calculates Dickey-Fuller test statistics. Those statistics are stored here and can be summarized, visually, however you want.

Perhaps the easiest thing is to lean on the ur_summary() function for you based on the information included in all elements of the object sadf_test() returns. "stats" has the test statistics, "sims" has the raw simulations, and "attributes" has a quick summary of the arguments fed to the sadf_test() function. Applied to our test, we get the following.

ur_summary(DF1)
#> ---------------------------------------------------- 
#> * Simulated (Augmented) Dickey-Fuller Test Summary * 
#> ---------------------------------------------------- 
#> Simulated test statistics are calculated on time series that are: nonstationary 
#> Length of time series: 500. Lags: 0
#> 
#> Type 1: no drift, no trend 
#> --------------------------
#> Your tau: 0.008
#> Potential thresholds for your consideration: -2.085 (1%); -1.88 (5%); -1.496 (10%)
#> 
#> Type 2: drift, no trend
#> -----------------------
#> Your tau: -0.23
#> Potential thresholds for your consideration: -2.093 (1%); -1.7 (5%); -1.242 (10%)
#> 
#> Type 3: drift and trend
#> -----------------------
#> Your tau: -2.472
#> Potential thresholds for your consideration: -3.764 (1%); -3.42 (5%); -3.198 (10%)
#> 
#> 
#> --------------------------------------------------------------
#> * Guides to help you assess stationarity or non-stationarity * 
#> --------------------------------------------------------------
#> These thresholds are the results of 100 different simulations of a non-stationary time series matching your time series description (n = 500, lags = 0). If your tau is more negative than one of these thresholds of interest, that is incompatible with a non-stationary time series and more compatible with a stationary time series.
#> 
#> If this is not the case, what you see is implying your time series is non-stationary.
#> 
#> Please refer to the raw output for the simulations for other means of assessment/summary.

In our case, the simulated time series to which we are comparing our time series is non-stationary. Knowing what we know from the pen-and-napkin math above, we expect a non-stationary time series to approximate a t-distribution whose central tendency hovers on 0 (even though we can’t call it a t-distribution). When information retention is partial (i.e. $\rho$ < 1), the coefficient predicting first differences becomes “more” negative and can be better discerned from 0 in the pen-and-napkin math above. Thus, you judge the test statistic by how negative it is and how easily it could be discerned from a distribution of test statistics generated from a random walk with permanent information retention. In our case, we have an obvious random walk. Its test statistic is very much compatible with a distribution of test statistics ($\tau$) that could be generated from a random walk. We cannot reject the null hypothesis of a non-stationary times series because, well, we generated a random walk. Duh.

Compare the above with a pure white noise times series and the one with partial memory. Even the one with partial memory has “shocks” today that decay geometrically. The series “forgets” past shocks pretty quickly, all things considered. The series with absolutely no information retention (i.e. the one generated by rnorm()) can more confidently reject 0.

# partial memory
DF2 <- sadf_test(fakeAR$y, n_lags = 0, n_sims = 100)
ur_summary(DF2)
#> ---------------------------------------------------- 
#> * Simulated (Augmented) Dickey-Fuller Test Summary * 
#> ---------------------------------------------------- 
#> Simulated test statistics are calculated on time series that are: nonstationary 
#> Length of time series: 500. Lags: 0
#> 
#> Type 1: no drift, no trend 
#> --------------------------
#> Your tau: -12.73
#> Potential thresholds for your consideration: -2.39 (1%); -1.853 (5%); -1.355 (10%)
#> 
#> Type 2: drift, no trend
#> -----------------------
#> Your tau: -12.716
#> Potential thresholds for your consideration: -2.453 (1%); -1.478 (5%); -1.191 (10%)
#> 
#> Type 3: drift and trend
#> -----------------------
#> Your tau: -12.73
#> Potential thresholds for your consideration: -3.907 (1%); -3.567 (5%); -3.137 (10%)
#> 
#> 
#> --------------------------------------------------------------
#> * Guides to help you assess stationarity or non-stationarity * 
#> --------------------------------------------------------------
#> These thresholds are the results of 100 different simulations of a non-stationary time series matching your time series description (n = 500, lags = 0). If your tau is more negative than one of these thresholds of interest, that is incompatible with a non-stationary time series and more compatible with a stationary time series.
#> 
#> If this is not the case, what you see is implying your time series is non-stationary.
#> 
#> Please refer to the raw output for the simulations for other means of assessment/summary.

# pure white noise
DF3 <- sadf_test(rnorm(500), n_lags = 0, n_sims = 100)
ur_summary(DF3)
#> ---------------------------------------------------- 
#> * Simulated (Augmented) Dickey-Fuller Test Summary * 
#> ---------------------------------------------------- 
#> Simulated test statistics are calculated on time series that are: nonstationary 
#> Length of time series: 500. Lags: 0
#> 
#> Type 1: no drift, no trend 
#> --------------------------
#> Your tau: -22.014
#> Potential thresholds for your consideration: -2.188 (1%); -1.752 (5%); -1.439 (10%)
#> 
#> Type 2: drift, no trend
#> -----------------------
#> Your tau: -21.994
#> Potential thresholds for your consideration: -2.416 (1%); -1.885 (5%); -1.395 (10%)
#> 
#> Type 3: drift and trend
#> -----------------------
#> Your tau: -22.003
#> Potential thresholds for your consideration: -3.817 (1%); -3.266 (5%); -3.117 (10%)
#> 
#> 
#> --------------------------------------------------------------
#> * Guides to help you assess stationarity or non-stationarity * 
#> --------------------------------------------------------------
#> These thresholds are the results of 100 different simulations of a non-stationary time series matching your time series description (n = 500, lags = 0). If your tau is more negative than one of these thresholds of interest, that is incompatible with a non-stationary time series and more compatible with a stationary time series.
#> 
#> If this is not the case, what you see is implying your time series is non-stationary.
#> 
#> Please refer to the raw output for the simulations for other means of assessment/summary.

See It For Yourself, with an Applied Example

You can better see this for yourself with actual data. USDSEK is a time series included in {sTSD} that has information I find good to know since I moved to Sweden from the United States: the Swedish crown (SEK) and U.S. dollar (USD) exchange rate. In particular, how many Swedish crowns are necessary to obtain one dollar? This is an interesting time series that you can see for yourself here.

This sure looks like it would have a strong, built-in information retention mechanism. Most time series of commodities that are traded daily have a pervasive, built-in memory. We can see for ourselves with the sadf_test() function and lean on it to identify an appropriate lag structure for us.

DF4 <- sadf_test(USDSEK$close, n_sims = 500)
DF5 <- sadf_test(diff(USDSEK$close), sim_hyp = "stationary", n_sims = 500) # I'm going somewhere with this...

ur_summary(DF4)
#> ---------------------------------------------------- 
#> * Simulated (Augmented) Dickey-Fuller Test Summary * 
#> ---------------------------------------------------- 
#> Simulated test statistics are calculated on time series that are: nonstationary 
#> Length of time series: 3900. Lags: 9
#> 
#> Type 1: no drift, no trend 
#> --------------------------
#> Your tau: 1.078
#> Potential thresholds for your consideration: -2.554 (1%); -1.964 (5%); -1.557 (10%)
#> 
#> Type 2: drift, no trend
#> -----------------------
#> Your tau: -0.683
#> Potential thresholds for your consideration: -2.504 (1%); -1.802 (5%); -1.415 (10%)
#> 
#> Type 3: drift and trend
#> -----------------------
#> Your tau: -2.893
#> Potential thresholds for your consideration: -3.868 (1%); -3.368 (5%); -3.094 (10%)
#> 
#> 
#> --------------------------------------------------------------
#> * Guides to help you assess stationarity or non-stationarity * 
#> --------------------------------------------------------------
#> These thresholds are the results of 500 different simulations of a non-stationary time series matching your time series description (n = 3900, lags = 9). If your tau is more negative than one of these thresholds of interest, that is incompatible with a non-stationary time series and more compatible with a stationary time series.
#> 
#> If this is not the case, what you see is implying your time series is non-stationary.
#> 
#> Please refer to the raw output for the simulations for other means of assessment/summary.

The test statistics are compatible with a distribution of test statistics of a random walk, leading us to reasonably conclude that our times series of a currency exchange rate traded daily is non-stationary. Of course it would be.

However, it would be illustrative to get an idea of what this looks like, visually. Here, the tests included in {sTSD} allow you to evaluate your time series against a stationary or non-stationary time series. Inference in the case of the Dickey-Fuller test is traditionally made against a non-stationary time series. However, you could simulate against a stationary time series to get an idea of what the test statistics would resemble for a time series with the length and number of lags requested. That, I think, is one feature missing when your critical values are handed down from on high based on what was computationally possible or feasible 50 years ago. We know from pen-and-napkin that the first difference should be stationary, so let’s also show what the test statistics from the first-difference time series looks like compared to a distribution of test statistics from a stationary time series.

Computational abilities of the time made it impractical to simulate distributions of stationary test statistics of the Dickey-Fuller procedure when the distribution of stationary test statistics depended on the non-1 values of $\rho$ and the length of the series. The distribution of test statistics for maximally autoregressive series were much more stable and predictable by comparison. Even though the procedure is traditionally done with the null hypothesis of non-stationary, you can assess your test statistics against simulations of a stationary time series all the same. Do what you want with the p-value under those circumstances, which is really the case for anything involving a p-value.

Conclusion

There really isn’t much to conclude here. It’s more of a quick introduction/tutorial for the MA students in our department who have to learn about unit root testing with my {sTSD} package. I never quite liked the R packages that were available in how they communicated the information of interest. Plus, it doesn’t make much sense these days to rely on old critical values generated 50 years ago for non-standard distributions like the one underpinning the (Augmented) Dickey-Fuller test. You can simulate those for yourselves. Moses can give you commandments and you just run with premise, supposedly. You can do the same here if you’d like, but simulation is much more informative.

One frustrating thing jumping into this topic has been the slew of synonyms to describe the same basic thing. A time series that looks like a familiar asset price (e.g. the ARCA Steel Index), might be “non-stationary”, have a “unit root”, or could even be “integrated”. The last of these is often the most frustrating thing to encounter. A stationary time series in level form is also “integrated”, if at order 0. It would look something like a problem-free trace plot from a Markov Chain Monte Carlo (MCMC) procedure. A time series that is integrated at order 1 would be a random walk, like the kind I’ll explore in this post. A first difference makes such a time series to be stationary. A time series integrated at order 2 would be an “explosive” time series, like the kind you might see of the Dow Jones Industrial Average over its entire lifespan (i.e. since 1885). Its growth in level terms looks exponential, should probably be log-transformed to avoid this (unless you’re dealing with, God help you, Bitcoin), and can be “double-differenced” to be stationary. I’m unaware of integration at levels beyond that, but there’s still a lot I don’t know about time series topics. ↩
There is a small caveat about the null hypothesis for the KPSS test, which by default is “stationarity”. The null hypothesis in the (Augmented) Dickey-Fuller and Phillips-Perron tests is “non-stationarity.” The latter two tests are anomalous from my vantage point of diagnostic tests because the null hypothesis is “you have a problem.” Compare to other diagnostic procedures we typically teach, like a Breusch-Pagan test, Durbin-Watson test, or Breusch-Godfrey test. Among unit root tests of which I’m aware, KPSS is unique for its null hypothesis. This affects the default behavior of the related functions in the {sTSD} package, but it does not affect what you could materially do with the simulations in this package. ↩
There does not appear to be a numeric plug-in for infinity (e.g. 1,000 or 10,000). Dickey (1976, 49-50) seems to be describing a limit function in which these statistics are analytically derived. ↩
In fact, I wrote this particular passage on my laptop somewhat antsy for things to do on a 14-hour flight from Seoul to Amsterdam. ↩
The choice of lagged first differences is what makes the Augmented Dickey-Fuller test to be “augmented.” It’s also not something a lot of econometrics textbooks I’ve seen belabor in any detail. adf_lag_select() might be of interest to you if you want to consider some thresholds for optimal lag selection tailored for your series while lag_suggests is a data set that will straight-up tell you what are some suggested first differences to specify, based on past scholarship. Do with those what you will, but it’s one reason why I would prefer to teach unit root tests around either the Phillips-Perron or KPSS procedures. In both cases, you ask for some kind of long- or short-term lag for the bandwidth/kernel generating the test statistic. If standard texts don’t belabor the lag selection procedure, doing an alternative test that doesn’t ask that information of you seems to make more sense. ↩
sim_df_mod() is mostly intended for internal use as a helper function, but it’s generating these different time series. In particular, it leans on the Rademacher distribution to generate drift and trend effects. I am unaware of many texts belaboring these details when they do simulate them, but the texts I have found could be reasonably approximated with this distribution. ↩

Replication Forensics

Thu, 18 Sep 2025 00:00:00 +0000

: Admit it; you heard the theme in your head just looking at this image. Peter Thomas is a ringing voice in my head from grad school.

The impetus for this post is as much a concession of defeat as it is a genuine interest of mine. Chatbots have seen enough of my material that they can effectively do my assignments for apathetic students in both my upper-division and lower-division methods courses. I do like the various data sets that I’ve crammed into {stevedata} and spammed onto CRAN. They have real instructional benefits. But, I have to change it up a bit.

Second, I have a genuine interest in the wild, wild west of what was quantitative political science over 20 years ago. Replication norms when I was starting out were nowhere near as rigorous as they are now, but computing power became much cheaper and software options became more sophisticated and user-friendly. It’s crazy to look back on the time when Yale hosted the Journal of Conflict Resolution and had a data repository that has since been lost to history (as far as I’m aware). If you’re lucky, you got a replication data set. If you’re even luckier, the replication data set did not get lost to time when the university underwent one of many structural changes to its web hosting platforms. The replication data set was almost assuredly posted on a faculty member’s website and web hosting platforms at the time were in their infancy. Perhaps, that faculty member may have simply moved the files to something more stable or uploaded them to Dataverse when they retired, switched employers, or had their old web hosting platforms taken from them or rehauled completely. If you’ve won the lottery, that replication data set includes a codebook and a script of some description that allows you to reproduce the findings. It’s led me to do some sleuthing in the furthest reaches of the web to try to find some of these forgotten materials. I have a small but ideally growing Github repository that tries to preserve copies of these data sets that are not easily found and may not even be available on Dataverse.

I also think this is an instructional opportunity for students. We’ve known for some time that replication is necessary for (political) science, and the discipline has since caught up for the most part. But what do you do when all you have is a data set? What do you do when the replication script, if you have one, is in a programming language you’ve never seen before? Well, figure it out.¹ In the absence of an exact recipe with software you have, you’ll have to do what I’ll call “replication forensics” to piece together what the author did and how you can translate it to something more current.² You have the data. You have the article. Figure it out.³

We’re going to do that today with one of my favorite articles I read in graduate school: Kenneth Benoit’s (1996) “Democracies Really Are More Pacific (in General): Reexamining Regime Type and War Involvement”. People reading this who are aware of me know that the bulk of my early research was democratic peace skepticism. I like this article in particular because it tries to tackle one of the biggest contradictions in the democratic peace corpus: the “monadic” phenomenon that is very much implied but never fully confirmed. Benoit (1996), like Rousseau et al. (1996), gives it an honest effort, but the results aren’t terribly convincing the more you look at them. Benoit’s (1996) analysis is also very much limited to a particular moment in time for which there isn’t much assurance that the findings will generalize. That said, it at least acknowledged the tension (which is more than can be said in a lot of democratic peace staples).

Here are the R packages we’ll be using today. I won’t directly load {MASS} but will use it for the negative binomial model function.

library(tidyverse)    # for most things
library(modelsummary) # for regression summary
library(kableExtra)   # for behind-the-scenes formatting
# library(MASS)       # used but not directly loaded.

Reading Benoit (1996)’s Data

I found a copy of Benoit’s (1996) data/analyses and uploaded these files to my pet Github repository for these things. Mercifully, you get a README file that will tell you where to look for replicating certain parts of his analysis. You also get various codebooks (as .cbk files) to assist you. That’s certainly nice, though I will say I didn’t look at them before writing this post and won’t look at them now. Seriously. I’m trying to practice what I preach; “figure it out”.

I want to replicate his Table 2, a screenshot of which follows.

: Table 2 in Benoit (1996)

The README file is pointing me to the WEEDE.ASC file for the data and the REG_TAB2.PRG file to replicate this. I’ll have to use the former but I’m going to avoid the latter for the sake of this exercise. I’ll get to the data shortly, but the reader and I are looking at the contents of these files for the first time. Neither of us, in all likelihood, have used the software that Benoit (1996) used for his analysis: GAUSS. I don’t know when exactly GAUSS fell out of favor in my field, but it was certainly before (and likely long before) I joined the field in 2006. Together, we’ll figure it out.

We’ll have to load the data before doing anything with them. I can already tell from context clues and, well, opening the WEEDE.ASC file in a text editor, that there is going to be a small issue. Here are the column names and the first row of the data file.

@ ccode	cname	superpwr	butterw	kende	ssm_is	ssm_es	ssis6074	sses6074	ssis7580	sses7580	ssal6080	ssal6074	ssal7580	poldem60	poldem65	civil73	polit73	civil79	polit79	fh73	fh75_80	dembin	area	defexd	defexdpc	encon	enconpc	gnppc	gdppc	impexppc	impexp	ecintdep	tvpc	szmil	szmilpc	popltn	larea	lpopltn	lszmil	lszmilpc	ltvpc	lecintd	limpexpc	limpexp	lencnpc	lgdppc	milwp70	popln70	ecnpc70	trade75	lmilwp70	lpopln70	lecnpc70 @
2	United_States	1	7	14	1	0	1	0	0	0	1	1	0	94.6	92.4	1	1	1	1	14	14	1	3619	76276144	37229	2278900	11127	4953	4501	40192	8234600	0.081	41292	3161	154	204880	3.559	5.311	3.5	2.188	4.616	-1.091	4.604	6.916	4.046	3.653	24.2	204879	11077	13.8	1.3838	5.3115	4.0444

That leading @ is going to be a slight frustration because R will want to read that as a column. There are certainly context clues that come with experience, but even the lay reader can see how this will go awry. My knowledge of the Correlates of War state system tells me that the United States has a state code of 2. It should be the case that ccode is the column that has that 2, but R will read a column of @ as having that 2. I don’t want that.

You have one of two options here. The first is to make a copy of the file, erase the @ and save. That’s certainly the easiest way to go about this and it’s what I would advise students to do under these circumstances. You could also do something like this.

Data <- read_file("~/Koofr/data/benoit1996drmp/WEEDE.ASC") 
# ^ assuming your data are here, which they won't be. Adjust to taste.

substring(Data, 1, 150) # looks gross, but don't worry
#> [1] "@ ccode\tcname\tsuperpwr\tbutterw\tkende\tssm_is\tssm_es\tssis6074\tsses6074\tssis7580\tsses7580\tssal6080\tssal6074\tssal7580\tpoldem60\tpoldem65\tcivil73\tpolit73\tci"
Data <- gsub("@", "", Data) # find any instance of @ and replace with nothing at all.
Data <- read_table(Data) # read the table

Data # ta-da
#> # A tibble: 101 × 55
#>    ccode cname   superpwr butterw kende ssm_is ssm_es ssis6074 sses6074 ssis7580
#>    <dbl> <chr>      <dbl>   <dbl> <dbl>  <dbl>  <dbl>    <dbl>    <dbl>    <dbl>
#>  1     2 United…        1       7    14      1      0        1        0        0
#>  2    20 Canada         0       0     0      0      0        0        0        0
#>  3    40 Cuba           0       1     2      0      1        0        0        0
#>  4    41 Haiti          0       0     0      0      0        0        0        0
#>  5    42 Domini…        0       1     1      0      0        0        0        0
#>  6    70 Mexico         0       0     0      0      0        0        0        0
#>  7    90 Guatem…        0       0     1      0      0        0        0        0
#>  8    91 Hondur…        0       1     1      1      0        1        0        0
#>  9    92 El_Sal…        0       1     1      1      0        1        0        0
#> 10    93 Nicara…        0       0     0      0      0        0        0        0
#> # ℹ 91 more rows
#> # ℹ 45 more variables: sses7580 <dbl>, ssal6080 <dbl>, ssal6074 <dbl>,
#> #   ssal7580 <dbl>, poldem60 <chr>, poldem65 <chr>, civil73 <dbl>,
#> #   polit73 <dbl>, civil79 <chr>, polit79 <chr>, fh73 <dbl>, fh75_80 <dbl>,
#> #   dembin <dbl>, area <dbl>, defexd <chr>, defexdpc <chr>, encon <dbl>,
#> #   enconpc <dbl>, gnppc <chr>, gdppc <chr>, impexppc <chr>, impexp <chr>,
#> #   ecintdep <chr>, tvpc <chr>, szmil <chr>, szmilpc <chr>, popltn <dbl>, …

An approach like this won’t be necessary if you’re handed a .sav file (SPSS format) or a .dta file (Stata format), but it’s something you’ll want to think about if you have a text file offered as a data file. This may just be a GAUSS quirk for all I know, though flat text files are ubiquitous in older analyses for which data are available. You should learn to anticipate some of these problems. Definitely look before you do anything.

Identifying the Variables in Benoit’s (1996) Table 2

I’ve read Benoit’s (1996) article to know what he is doing. He’s regressing two different measures of war counts (Butterworth and Small-Singer) on two different measures of democracy (Bollen’s (1980) measure of democracy and a measure of democracy from Freedom House). He will do that for three different estimation techniques: the linear model/OLS, the Poisson model for counts, and the negative binomial model (used for over-dispersed/under-dispersed counts). I could look at the program file that accompanies this analysis, but I won’t for the sake of what I encourage students to do for themselves. Figure it out.

Let’s identify the column names in the data if they’ll signal what we want.

colnames(Data)
#>  [1] "ccode"    "cname"    "superpwr" "butterw"  "kende"    "ssm_is"  
#>  [7] "ssm_es"   "ssis6074" "sses6074" "ssis7580" "sses7580" "ssal6080"
#> [13] "ssal6074" "ssal7580" "poldem60" "poldem65" "civil73"  "polit73" 
#> [19] "civil79"  "polit79"  "fh73"     "fh75_80"  "dembin"   "area"    
#> [25] "defexd"   "defexdpc" "encon"    "enconpc"  "gnppc"    "gdppc"   
#> [31] "impexppc" "impexp"   "ecintdep" "tvpc"     "szmil"    "szmilpc" 
#> [37] "popltn"   "larea"    "lpopltn"  "lszmil"   "lszmilpc" "ltvpc"   
#> [43] "lecintd"  "limpexpc" "limpexp"  "lencnpc"  "lgdppc"   "milwp70" 
#> [49] "popln70"  "ecnpc70"  "trade75"  "lmilwp70" "lpopln70" "lecnpc70"
#> [55] "X55"

Use your eyes, your head, and the context clues from Table 2. Notice Benoit (1996) says that Butterworth’s (1976) conflict data is one of his dependent variables. We can clearly see a column named butterw, and the only such column with a name that remotely matches the description we want. That tells us that’s our dependent variable. We’ll need a bit more sleuthing to discern what could be the Singer-Small data he’s using. Three context clues are guiding me here. First, it’s usually—not always: usually—the case that the main variables of interest to a researcher are in the first few columns. They’re not typically at the end. So, I’m starting to look at the first few columns I see. The second context clue is ss, suggesting “Singer and Small”. I see a few columns that start with that, which all have numbers following them. There’s my third context clue: they’re signaling the temporal domain of the count of wars. Per Table 2, the temporal domain of these count of wars says they span 1960 to 1980. Putting all those context clues together, I know ssal6080 is my second dependent variable.

I can do something similar for the two measures of democracy that serve as his independent variables. Benoit (1996) helpfully tells you the political democracy measure benchmarks to 1965. My eyes see two columns as potential candidates for this variable: poldem60 and poldem65. Given the information available, it seems like poldem65 is the obvious candidate for what he says is “POLDEM 1965” in his analysis. Following this logic, I can again discern what would be the Freedom House measure from 1973. It should be fh1973, signaling Freedom House’s measure of democracy in 1973. Here would be a reduced version of the data.

Data %>% 
  select(ccode:cname, butterw, ssal6080, poldem65, fh73)
#> # A tibble: 101 × 6
#>    ccode cname              butterw ssal6080 poldem65  fh73
#>    <dbl> <chr>                <dbl>    <dbl> <chr>    <dbl>
#>  1     2 United_States            7        1 92.4        14
#>  2    20 Canada                   0        0 99.5        14
#>  3    40 Cuba                     1        1 5.2          2
#>  4    41 Haiti                    0        0 20.7         3
#>  5    42 Dominican_Republic       1        0 38.8        11
#>  6    70 Mexico                   0        0 74.5         8
#>  7    90 Guatemala                0        0 39.5        11
#>  8    91 Honduras                 1        1 50           6
#>  9    92 El_Salvador              1        1 72.1        11
#> 10    93 Nicaragua                0        0 55.4         9
#> # ℹ 91 more rows

Doing this, though, alerted me to a potential issue I’ll want to consider. R read the poldem65 as a character when I know it should be a numeric variable. We’ll have to find out what happened.

Data %>% distinct(poldem65) %>% pull()
#>  [1] "92.4" "99.5" "5.2"  "20.7" "38.8" "74.5" "39.5" "50"   "72.1" "55.4"
#> [11] "90.1" "76.9" "71.4" "73.4" "44.6" "87"   "60.9" "36.2" "44.7" "97"  
#> [21] "52.6" "99.6" "99.1" "97.2" "99.7" "99.9" "90.8" "10.4" "39"   "88.6"
#> [31] "18.1" "22.1" "97.1" "11.6" "20.5" "96.8" "18.2" "50.8" "82.8" "37.5"
#> [41] "20.9" "97.3" "37.2" "53.7" "24.7" "16"   "45.6" "37.3" "7.2"  "23.7"
#> [51] "45.5" "56"   "49.5" "34.2" "47.6" "38.5" "77"   "12.5" "58.9" "83"  
#> [61] "32.2" "63.6" "34.4" "37.9" "45"   "76.4" "11.4" "38.7" "19.9" "74"  
#> [71] "30.8" "9.7"  "6.5"  "23.2" "16.4" "22.8" "21"   "53"   "99.8" "91.2"
#> [81] "62.5" "."    "85.9" "29.2" "17.3" "36.3" "42.8" "33.1" "12.4" "80.3"
#> [91] "92.6" "9.8"  "100"

It looks like the missing data code for this measure is a .. If I knew in advance this was an issue, I could’ve potentially passed it off as an argument in the read_table() function (i.e. read_table(Data, na = ".")). But, I’m only catching this now. We’ll find out what observation this is first before fixing it. We can already deduce from Table 2 that it concerns just one observation.

Data %>%
  select(ccode:cname, butterw, ssal6080, poldem65, fh73) %>%
  filter(poldem65 == ".") # Burma...
#> # A tibble: 1 × 6
#>   ccode cname butterw ssal6080 poldem65  fh73
#>   <dbl> <chr>   <dbl>    <dbl> <chr>    <dbl>
#> 1   775 Burma       0        0 .            4

Data %>% mutate(poldem65 = ifelse(poldem65 == ".", NA, poldem65),
                poldem65 = as.numeric(poldem65)) -> Data

Data %>% 
  select(ccode:cname, butterw, ssal6080, poldem65, fh73)
#> # A tibble: 101 × 6
#>    ccode cname              butterw ssal6080 poldem65  fh73
#>    <dbl> <chr>                <dbl>    <dbl>    <dbl> <dbl>
#>  1     2 United_States            7        1     92.4    14
#>  2    20 Canada                   0        0     99.5    14
#>  3    40 Cuba                     1        1      5.2     2
#>  4    41 Haiti                    0        0     20.7     3
#>  5    42 Dominican_Republic       1        0     38.8    11
#>  6    70 Mexico                   0        0     74.5     8
#>  7    90 Guatemala                0        0     39.5    11
#>  8    91 Honduras                 1        1     50       6
#>  9    92 El_Salvador              1        1     72.1    11
#> 10    93 Nicaragua                0        0     55.4     9
#> # ℹ 91 more rows

Cool, let’s proceed.

Replicating Benoit’s (1996) Table 2

There is definitely a convoluted way of making a loop and list of these various models, though it would be beyond the scope of this post to do this and might only serve to overwhelm my students reading this. So, we’ll do it the tedious way. Let’s do the Butterworth wars first.

M1 <- lm(butterw ~ poldem65, Data)
M2 <- lm(butterw ~ fh73, Data)
M3 <- glm(butterw ~ poldem65, Data, family = "poisson")
M4 <- glm(butterw ~ fh73, Data, family = "poisson")
M5 <- MASS::glm.nb(butterw ~ poldem65, Data)
M6 <- MASS::glm.nb(butterw ~ fh73, Data)

A regression table, by way of {modelsummary}, follows.

A Replication of Table 2 in Benoit (1996) for Butterworth's Data on Wars
	POLDEM 1965 (OLS)	FH 1973 (OLS)	POLDEM 1965 (Poisson)	FH 1973 (Poisson)	POLDEM 1965 (NB)	FH 1973 (NB)
Democracy	−0.0082	−0.0487	−0.0073*	−0.0448+	−0.0071	−0.0437
	(0.0052)	(0.0389)	(0.0031)	(0.0237)	(0.0048)	(0.0363)
Intercept	1.6026***	1.4953***	0.5167**	0.4405*	0.5068+	0.4333
	(0.3241)	(0.3205)	(0.1729)	(0.1777)	(0.2841)	(0.2882)
R2	0.025	0.016
Log.Lik.	−189.074	−191.188	−166.703	−168.858	−147.559	−148.649
Num.Obs.	100	101	100	101	100	101
+ p < 0.1, * p < 0.05, p < 0.01, * p < 0.001

By and large, we’ve replicated what Benoit (1996) reports in his Table 2 for the analysis of Butterworth’s war data. It does come with two caveats that are immediately discernible. The first one is an advanced note in which I have to confess that I have no idea how GAUSS is calculating the log likelihood of the Poisson models. I only know of the one way of doing it (i.e. how R does it: sum(dpois(M3$y, lambda = fitted(M3), log = TRUE))). However, that’s an advanced note that I cannot earnestly expect students to know (because I wouldn’t know it myself). More importantly, the coefficients are identical. The same cannot be said about the negative binomial regression. Here, there is some interesting disagreement. I finally peeked into the information that Benoit (1996) makes available. The negative binomial regression in the program file seems like it was custom written for the task at hand. I can already discern the different parameterizations of dispersion, which are also stated in the footnote of the table. I can already see different starting values for the dispersion parameter, or at least I think I do. I know the glm.nb() function in the {MASS} package starts with the Poisson and that doesn’t seem(?) to be where the negative binomial routine starts in GAUSS. To be clear, the differences are rather slight with all that in mind. But, they can’t go unnoticed.

They’re also immaterial for my intended audience. I can’t expect intro-level students to know those advanced details, but I can expect them to dig into the data to discern the basic information and modeling techniques. Toward that end, we’ve accomplished the first half of Table 2.

Now, let’s round home with a replication of the second half of Table 2, where the dependent variable is the Singer and Small wars.

M7 <- lm(ssal6080 ~ poldem65, Data)
M8 <- lm(ssal6080 ~ fh73, Data)
M9 <- glm(ssal6080 ~ poldem65, Data, family = "poisson")
M10 <- glm(ssal6080 ~ fh73, Data, family = "poisson")
M11 <- MASS::glm.nb(ssal6080 ~ poldem65, Data)
M12 <- MASS::glm.nb(ssal6080 ~ fh73, Data)

Summarize it for us, {modelsummary}.

A Replication of Table 2 in Benoit (1996) for Singer and Small's Data on Wars
	POLDEM 1965 (OLS)	FH 1973 (OLS)	POLDEM 1965 (Poisson)	FH 1973 (Poisson)	POLDEM 1965 (NB)	FH 1973 (NB)
Democracy	−0.0037	−0.0329+	−0.0085+	−0.0825*	−0.0083	−0.0850+
	(0.0026)	(0.0191)	(0.0050)	(0.0407)	(0.0061)	(0.0488)
Intercept	0.6489***	0.6801***	−0.3736	−0.2757	−0.3863	−0.2593
	(0.1610)	(0.1576)	(0.2748)	(0.2805)	(0.3535)	(0.3547)
R2	0.020	0.029
Log.Lik.	−119.136	−119.526	−92.150	−91.824	−88.752	−88.501
Num.Obs.	100	101	100	101	100	101
+ p < 0.1, * p < 0.05, p < 0.01, * p < 0.001

The same basic story emerges. We’ve just about perfectly reproduced the linear model output and the Poisson model output (barring the log-likelihood in the latter). There’s also the interesting discrepancy in the negative binomial regression, though we’ll have to chalk that up to different optimization and parameterization procedures that I cannot honestly expect students to know at this point in their studies.

Conclusion

The purpose of this post isn’t to point fingers. I actually rather appreciate that the author had these data available and actually included a full-fledged codebook, script, and output log to show his work. Speaking from experience, that was rather anomalous for scholarship at this time. Further, the point isn’t to further scrutinize democratic peace scholarship anymore than I already have in the bulk of my early published scholarship. This is more of a warning and a sales pitch to my students. I’m going to present something like this to you soon because I think it’s beneficial to your education. You will learn a lot by replication, yes. But, you’ll also learn a lot as well by doing some forensics to figure it out for yourself how others did it in the presence of poorly sourced data. I have an example I’m going to roll out to my Master’s students that will be a bit simpler in presentation, but actually won’t at all have a codebook or other documentation beyond the published article itself.

You’ll just have to figure it out. You’ll learn a lot in the process.

One legendary case of this is the scope of work by Herndon et al. (2014) to document the deception and carelessness of Reinhart and Rogoff (2010). You can read my reproduction of Herndon et al.’s (2014) work here. What I’m asking students to do has that same spirit, though ideally not the same ramifications. ↩
The use of this term may be somewhat stark because it likens a published article to a crime scene. Alas, from one cynical perspective, that’s maybe fair. ↩
No one will pay you any serious amount of money to follow a completely prescribed routine. If that were your task, you are immediately replaceable if the next quarterly report looks lukewarm to shareholders. ↩

Some Opinionated Takes on Chatbot Use by Students

Mon, 08 Sep 2025 00:00:00 +0000

: When you do things right, people shouldn't be sure you've done anything at all.

What follows is a slightly more opinionated add-on to the required language on academic misconduct that goes into my course descriptions. To be clear, the “required language” I introduce in course descriptions is stuff we’re told to include and will govern the course. These passages that follow may provide additional clarification to the student in light of potentially negative, harsh, or critical feedback I provide. I’m sure it will be periodically updated. I just need students to know why I feel the way I do about what I see students doing.

The most apt metaphor I’ve seen regarding the release of ChatGPT on 30 Nov. 2022 likened it to the Trinity test in New Mexico on 16 July 1945. The Trinity test constituted the beginning of what we retroactively call “the atomic age”, but it had one major implication for the world. It so contaminated all metals manufactured after that date that the onset of this “age” significantly jeopardized all important medical and technical equipment. To compensate, scientists working in technology and medicine needed important materials (so-called “low-background” steel/lead) that were manufactured before this date. Supposedly our steel is no longer radioactive, in the same way or to the same degree, but that is still a major implication of just one event and one particular moment in time.

The article I read invoking that metaphor is more concerned for new knowledge generation and what potentially happens when synthetic observations produced by AI models become part of future training sets. Those of us that know about the Habsburgs or remember what happens when you make copies of copies of VHS cassettes will know what the implication is for the future of innovation and research. My concern is mostly for university education, however. Everything (where I work) is open-book and open-notes. Everything (where I work) is take-home with a deadline of seven days or so. I’m teaching students now who have known ChatGPT (and related services) through the entirety of their university education. I will soon start teaching students who have had it through the entirety of their secondary education. Like the scientists frantically searching for scuttled ships for pre-war steel, I can no longer trust the quality of what I see. I know you’re using it, and often uncritically. I cannot trust that you are actually doing what you say you are doing. That’s true even if you are legitimately and earnestly doing the work, with or without assistance from a particular chatbot. I just can’t trust you the same way I could before Nov. 2022. Whether it’s topical courses in the intro-level, theses at the BA-level, or definitely quantitative methods instruction at any level, I’m confident that chatbots have so greatly contaminated what I’m asked to evaluate that I’m almost doing gratis product testing for one of these services.

This manifests in several ways, and will likely coincide with me making harsh remarks in student output. These remarks will read like accusations, though they aren’t. Our guidelines do not put us in great positions to prosecute these things even if the line separating “proper” from “improper” were clear.¹ They aren’t accusations as much as they are frustrations with what students do, or are asking me to do, with a straight face. I cannot help if or how students use one of these particular services to guide themselves through university education, but I can offer these assorted comments to contextualize how I feel and why I might be frustrated with what a student is asking me to read.

Here’s a table of contents.

“Godfellas” (i.e. Please Make it Less Obvious)
Your Chatbot Isn’t Answering the Question I Asked
You’re Not the Only One
You Should Not Aspire to Sound Like a Chatbot
False Positives and Actual Innocence

“Godfellas” (i.e. Please Make it Less Obvious)

This has happened twice, as of writing, for assignments in which we as faculty often work closely with students through iterations of a particular assignment. In these two instances, students were basically submitting ChatGPT generated output for me to evaluate prior to the submission of a final assignment. It was obvious, especially when they couldn’t explain what they were doing in their own words. They were at least honest about it. I don’t feel evaluating ChatGPT output is a productive use of my time, but it turns out it was a required use of my time. The bulk of my interaction with them reduced to, in as many words, optimizing their use of ChatGPT to get what they wanted and pleading with them to make it less obvious (to me, at least) that this is what they were doing.

Notice that I can’t tell them “stop.” I couldn’t enforce “don’t.” I tried it once but wasn’t getting anywhere and I had no real recourse. Instead, I had to advise them to please make it less obvious that ChatGPT did this and you (i.e. those two students) didn’t. I’m an American and reference American TV shows all the time. Thus, I had to harken a line from Futurama and one particular episode: “Godfellas.” In this episode, Bender is flung to space and happens to meet “God.” The conversation God and Bender have focus on the appropriateness of divine intervention, especially when divine intervention could be the difference between faith and reality. God’s hedge in this conversation is that doing too little leads believers to lose faith, and doing too much induces unwanted dependence. Speaking Bender’s particular language, you need a light touch to avoid detection if you want to be a successful pickpocket. If you’re going to burn down a business for insurance money, you need to make it look like an electrical mishap. When you do things right, I shouldn’t be sure you’ve done anything at all.

I can’t make dishonest students be honest, but try to make your arson look a little more accidental if this is the route you choose. You’re doing this because you want that insurance payout, right? Cover your tracks better.

Your Chatbot Isn’t Answering the Question I Asked

I’ve harped on this before, but this is one of my biggest educational mantras. “Answer the question I asked.” Too often, students (even before the new atomic age) would answer questions potentially (un)related to the question I would ask. I would reply that I need the student to be more focused on the question I’m actually asking. I find myself repeating that more since I moved here (incidentally just before Nov. 2022). I’m not sure students fully appreciate that the chatbot often goes into la-la land for a particular prompt, generating non-answers to specific questions that I’m asking. I structure my questions in a particular way because I want you to know a detail well. The chatbot will often miss that as much as the pre-atomic age student trying to know enough things, generally, to get by in the course. If you want to use it, I can’t stop you. Just make sure it’s working for you.

Related to this, students I think assume that I haven’t fed my prompts through a chatbot to understand how they respond to my exam questions. I know how it behaves. I have a good idea of its tells with respect to what I’m asking the student (or the chatbot) to do.

You’re Not the Only One

Some classes I teach are entirely online and involve students from across Europe. Most are in-person, but where students are often tight-knit and work on stuff together. It’s actually a welcome culture shock here relative to the United States, where students in a class often don’t know other students exist and couldn’t pick them out of a lineup. For the in-person classes, I can readily discern “networks” and friend groups in the class. In a class of 35-50 students, there might be two or three such networks consisting of maybe 10-15 students.

If, however, I’m getting more than half the class all repeating particular statements and particular expressions, I’m left believing that it’s coming from a common pool that’s beyond the friend group. Again, you’re not the only one, and try to make it less obvious. Students often see just the one thing they think they can slide underneath the radar. I see the several dozen things that students try to slide underneath my radar.

You Should Not Aspire to Sound Like a Chatbot

This is a particularly spicy take, but I think most chatbots read in the most grating way possible.² ChatGPT, in particular, is just an obnoxious messenger. I don’t know this for a fact, but I’m inclined to believe its English training set is filled with stuff on Medium or, in particular, LinkedIn. Its dialect is a very heavy “Linkedinglish”.³ I despise reading that particular dialect because I don’t think it’s professional in the slightest. If I’m asking academic questions pleading for some type of professional response, what I get back reads like a cringe-inducing sales pitch about what, say, central limit theorem can teach us about B2B sales. Just, gross.

This will manifest in two particular ways, the latter of which might seem particularly harsh (even if I don’t mean any ill will). First, I think the chatbot is getting worse in this way. ChatGPT, in particular, seems to have degenerated over time to an increasingly base-level Linkedinglish that’s almost at the level of a university intern taking the reins of a company’s social media platforms. If Linkedinglish was just one of a few English dialects in its algorithm, it seems to have learned that its users like this particular form of English. ChatGPT, if nothing else, is a people-pleaser.⁴

From my vantage point, it’s embarrassing to read its output sometimes because it’s clear the product is entering an “enshittification” arc. It’s what happens when a product expands so much it not longer is incentivized to sell you a product that works. It’s more inclined just to keep you hooked on a product that won’t get better the same way it did. Like a generic company’s social media platform, the chatbot is more interested in selling you itself (and not necessarily solutions).

Be mindful of the distinction here. Whether it’s working well for you is just incidental to whether you’re using it. That’s what matters to the tech company and its chatbot, first and foremost. The chatbot is kind of like a drug dealer. You come to it with problems, but it doesn’t quite have the solutions you need. It gives you assurances that it does, though, in hopes that you’ll get hooked to it. You come to it with questions and, above, it gives you answers to questions you did not ask. You come to it wanting to know how you should write, and you end up sounding like a post on how the 95% confidence interval helped you generate $25M in sales across your business journey. You should not aspire to sound like that.

The second will seem harsh, but I only offer this because I’m on a similar journey learning Swedish (and Korean). The chatbot speaks better English than university students, whether English is the student’s native language or not.⁵ English is the lingua franca of the world, so of course its English capabilities are really good. It’s going to sound better, without context, than students in that 19-23 age range. If you’re using it to generate prompts, try to rewrite what you see in your own words and in a style that’s more appropriate for the academic venue. You’ll learn more that way. Also, definitely do that where output has 👉 emojis ✅ or unnecessary terms in bold face. I would need ✋two hands✋ to tell you how many times I’ve seen that leak into student submissions.

False Positives and Actual Innocence

Periodically, I’ll write “this sounds like a chatbot” on a student’s submission and the student will respond quite vigorously that they did not do what I’m accusing of them. When this happens, I have a few initial reactions. Sometimes I’m convinced (or know outright) that they’re not being truthful. Second, the student should know it’s not an accusation because an accusation is something I’d have to formally charge. I only say those things in the context of providing grades on assignments, and I could not grade an assignment if I were alleging misconduct. Third, I might genuinely be wrong. It happens. I think I have a good idea for how chatbots approach my material, and I know for a fact they’ve seen my material a lot. However, I’m not foolproof. I could genuinely be mistaken. It’s why I try my best not to couch it as an accusation, though I’m not very good at withholding my frustration.

Still, if I may, for the potential student of mine reading this: imagine what it’s like from my perspective seeing a student independently reproduce something a chatbot would do if and when that genuinely happens. I wouldn’t say it if I were convinced you knew what you were doing and were answering the question I asked. I know chatbots routinely go into la-la land answering questions I did not ask. I know chatbots have a particular lingo and a particular set of tells with respect to my material, some of which are quite clumsy. I know it does all this in a confident, sales-person English. I also know chatbots are something that a potential employer could use at a considerable saving to themselves in order to bullshit their way through the next quarterly report. If I’m an employer spending money on a chatbot and evaluating it and the job-candidate student that produces identical outputs, I will go with the chatbot every time. The chatbot is always cheaper.

Even when I was young, I was aware that deeper learning is only incidental to what the university offers. It offers a necessary credential to gain a foothold in a competitive job market and a changing economy. I can see why cynical students are tempted to chatbot their way through a university education in order to get that piece of paper, but I don’t think they appreciate that doesn’t put them in a great position in a job market ever interested in automation and making a labor force disappear from the budget. Even the most honest students should wrestle with this. If you sound like a free service to me, you’ll sound like a free service to an employer. My heart sincerely aches for a younger generation that has to deal with that now. It makes me happy I’m as old as I am. I and my friends did not have potential employers doubting our ability or utility the same way I’m sure a younger generation is being doubted and dismissed right now. I would be distraught if I had to fight upstream against the accusations that I’m indistinguishable from Claude or ChatGPT.

In this day and age, especially with the bigger political problems we all face, it’s why I get supremely frustrated with what I see students doing or failing to do. Slop is everywhere around us and I’m pleading with you, the student, to distinguish yourself from it. I want to see your humanity, warts and all, to help you get better. I want to know it’s you writing this stuff, but I just can’t know that these days the same way I could before Nov. 2022. I can’t trust students are 100% honest in this atomic age. For the dishonest students, I don’t want to see you atrophy your mind because the chatbot makes it conveniently easy for you to do that. You’ll hemorrhage your capacity to think, think critically, and you’ll hemorrhage your future.

For the honest students, I don’t want to see you sound like the machine that’s siphoning energy away from the human labor force. I want to see you be better than the machine. I want you to maximize your potential to make the most of your capacity in an era where employers are increasingly trying to automate blue- and white-collar jobs out of existence. Try to rise above the machine. Rage against it, even.

The burden to prosecute academic misconduct is particularly high. One of my little American quirks here is thinking through what could be litigated and what I could “prove” as if it were a court of law. I cannot obviously summon the coercive arm of the state to seize a student’s computer to prove impropriety, and the punishment mechanisms are often lacking relative to what I remember at American universities. ↩
I understand students can use these services to catch typos and potentially tighten up a few passages in their written material. I respect that use but would not encourage students to use it as a complete substitute for editing they should be doing themselves. ↩
I’m aware of the irony of that article being on Medium. ↩
Jae DiBello Takeuchi is a friend of mine from a previous employer. From the Japanese speaker’s perspective, ChatGPT is that insufferable person that no one wants to be stuck with at a party. Likewise, I would not want to be stuck having to entertain or be entertained by a LinkedIn lunatic at a social event. ↩
Likewise, it writes better Swedish than I do. I haven’t tested it with Korean yet, but I’m just starting that journey. ↩

Learning About Expected Categorical Relations (Chi-Squared Tests) in R by Way of Arms Races and War

Mon, 11 Aug 2025 00:00:00 +0000

: An Eritrean soldier stands in front of a destroyed T-55A tank in 1999. This war starts in 1998 but the mutual military build-up for it arguably started in 1996. (Broń Pancerna/Flickr)

This is a post I’m writing just to spam material to my blog, and also to pad material I need to prepare for my IRIII students in their quantitative methods sequence. It’s a challenge to teach them stuff that is super basic, but has a real-world application, and in the limited time I have with them. The class had historically been built toward just getting them to do lm() in R (or regress in Stata) and to be happy if they can do that. However, there’s been some shuffling amid hour cuts that incidentally gives me more time to teach them more stuff. But, again, it has to be simple.

Enter the arms race and war debate. I cut my teeth on this debate in graduate school and still like to teach around the basics of this stuff when I can. I still think Lewis Fry Richardson’s linear theory of nations gets at the core of how we should conceptualize the arms race, acknowledging it’s a glorified port of his training in mathematics to the realm of international relations.¹ Admittedly, it is a bit of a dated topic. Perhaps it’s fair to note that the “arms race” of a century or two ago looks nothing like what an arms race would resemble now. It won’t be simple expenditures. It won’t be manpower. It won’t even be boats. But the substance of this debate maps nicely to two thing I want to accomplish in an IR curriculum at the bachelor’s level. First, it highlights how a lot of realpolitik conventional wisdom stretches so thin it strains to cover anything in detail. Second, the empirical application of this debate is all chi-squared tests. There are definitely more advanced ways of approaching this, especially with what this means in the 21st century. But, you can learn about the chi-squared test with these things I’d have you read anyway if I could.

Here are the R packages I’ll be using today. Do note that {stevedata} has a forthcoming data set I’ll be using to offer another means to assess this relationship.

library(tidyverse)   # for my basic workflow
library(stevedata)   # forthcoming v. 1.7.0, for `mmb_war` data
library(stevethemes) # for themes
library(kableExtra)  # for tables

theme_set(theme_steve()) # setting a default theme

Here’s a table of contents.

The Debate: Para Bellum and the “Steps to War”
- The Empirical Debate Between Wallace (1979) and His Critics
The Chi-Squared Test
- A Re-Analysis Using Newer Data
Conclusion

Let’s get started.

The Debate: Para Bellum and the “Steps to War”

Si vis pacem, para bellum (if you want peace, prepare for war)

This aphorism sounds cool and, from a so-called “realist” worldview, it has a certain kind of logic by which it works. Intuitively, a true balance of power between two states in which neither side could comfortably defeat the other side makes war an unattractive option. It’s the same kind of logic that underpins mutual assured destruction. It may not sound pleasant to the pacifist, but “peace through strength” is sound strategy if you were to think of it in this particular way. Mutual military build-ups evoke certain tensions or anxiety in the general public about an impending war, though the behavior itself would be perfectly rational if the idea is to avoid war. A rational state substantially increases it might to prepare for war, wanting peace. A rival state requites that militarization/mobilization, preparing for war while wanting peace. Together, they achieve an equilibrium where an increased mobilization for war makes peace more attractive than it might have been in the absence of the mutual build-up. So that story goes.

There is a group of us in international relations that have always found this, at best, laughably simplistic and, more to the point, a ridiculous proposition of an equilibrium. Richardson’s work is foundational on this topic, though perhaps it’s fair to say his conclusions are importantly conditioned by his own pacifist ideals. J. David Singer (1958) started his article with that exact quote, as relayed by Vegetius, and noted the immediate epilogue under Theodosius I was a spate of violence conducted by the emperor.² My training is largely in the “Steps to War” tradition of John Vasquez. There are underlying causes of war (prominently the allocation of territory) and proximate causes (“steps”) of war. Violence is not unique to humans, but warfare arguably is a unique kind of learned, social behavior. Realpolitik policy prescriptions (like arms races) prepare states for sustained combat and only heighten mistrust of the other side under what was already less than rosy conditions. The “steps” policymakers take under those circumstances limit future options, forgo more peaceful off-ramps, and further empower domestic “hardliners” or “hawks” to favor further aggression. Another way of thinking about this is to ask why you’re arms-racing in the first place. Why are you? Certainly not because things are going great and certainly not because no one with power/influence in society wants to use these toys. Be real, so-called “realists.”

The Empirical Debate Between Wallace (1979) and His Critics

The empirical debate on this relationship may not have started with Wallace (1979), but it sure as heck escalated under him. His Table 2 (reproduced below) argued arms races almost always lead to war (23 occurrences to five non-occurrences) and the absence of arms races almost never lead to war (three wars in the absence of arms races to 68 non-wars). For those who like working with percentages, that’s a comparison of 82% to 4% for 99 total cases. It’s more than enough to suggest something is happening here.

Table 2 in Wallace (1979)
	Arms Race	No Arms Race
War	23	3
No War	5	68

However, there are several important reasons to doubt how Wallace came to this conclusion.³ Critiques came in all directions for this analysis, including Altfield (1983), Horn (1987), Morrow (1989), and Weede (1980) (among many others, I’m sure). I want to focus on Diehl (1983) as largely capturing all the recurring themes in this critique of Wallace (1979). Namely: Wallace is describing a measure of arms races that he never makes publicly available and could not replicate. It’d be difficult to know for sure, but his polynomial arms race function he describes would also inadvertently pick up unilateral buildups. Further, it’s pretty clear that the world wars are the lion’s share of his empirical support. He’s additionally compounding this problem by disaggregating all of them (even if there is an ongoing war, like his USSR-Japan case in 1945). In one re-analysis of this relationship (Table II), Diehl reports the relationship between arms races and war is basically a null relationship. Three of 13 arms races lead to war (23%) compared to a nine wars in 73 cases (12%) not preceded by an arms race. One rate is more than the other, but it’s not a discernible relationship. There is no clear relationship by which arms races lead to war.⁴

Table II in Diehl (1983)
	Arms Race	No Arms Race
War	3	9
No War	10	64
Note: Chi-sq: 1.06, n = 86.

When I do get the opportunity to teach about arms races, I offer Susan Sample (1997) as a kind of Solomon in this debate. The empirical debate employed multiple different operationalizations of arms races and multiple different dispute data sets. A battery of analyses mix-and-matching different data sets yields the overall conclusion that arms races are positively associated with dispute escalation to war, even if it’s fair to note Wallace’s (1979) original conclusions were far too stark. See, as just one example, her Table IX using Diehl’s arms race index with all major state militarized interstate disputes. 14 of the 53 disputes in which there was an arms race led to war (26%). Just 17 of the 204 (8%) of the disputes in which there wasn’t an arms race led to war. That is a discernible difference, if clearly not the magnitude difference originally reported by Wallace (1979).

Table IX in Sample (1997)
	Arms Race	No Arms Race
War	14	17
No War	39	187
Note: Chi-sq: 13.0, n = 257.

I’ve glossed over to this point the exact tests scholars were using in this debate over operationalization and sample selection. It’s almost always a chi-squared test of independence for observed and expected counts. You could actually do this yourself with no real effort at all to understand the inferential claims the various people in this debate are making.

The Chi-Squared Test

Chi-squared ($\chi^2$) tests are common tests to use in very basic applications to assess whether the observed counts in some kind of category for two (or more) groups are discernibly different than what would be expected if there were no difference between or among the groups. It’s easier to introduce students to this in a simple 2x2 application like this and it will lean on the contingency tables you’ve seen everywhere to this point.

: Table IX (Sample, 1997)

First, let’s show Sample’s (1997) actual Table IX to see what’s happening here. Notice the 2x2 contingency table reproduced above. The primary grouping variable here is in the column whereas the rows are a kind of outcome of interest to us (war or no war). Let’s further calculate some total information about the data. There are 31 wars (14 with arms races, 17 without them). There are 226 disputes in the data that did not escalate to war (39 with arms races, 187 without them). Those are our row totals. We noted the column totals above. There were 53 disputes with arms races preceding them in these data (14 of which became wars). There were 204 disputes without arms races preceding them (17 of them escalating to war). Let’s expand this table a bit with this information.

Table IX in Sample (1997), with Row and Column Totals
	Arms Race	No Arms Race	Row Total
War	14	17	31
No War	39	187	226
Column Total	53	204	257

We next need to calculate expected counts, or what we would expect if there were no association between arms races and war. Formally, each expected value in a given cell is equal to the row total times the column total, divided over the total number of observations (257, in all cases).

Expected Counts in Table IX in Sample (1997) For No Association
	Arms Race	No Arms Race
War	(31*53)/257 = 6.393	(31*204)/257 = 24.607
No War	(226*53)/257 = 46.607	(226204)257 = 179.393

The formula for the chi-squared test statistic is very straightforward. It’s something you could do on paper, or in Excel even. It would take more effort, but it’s not difficult at all.

\[\chi^2 = \sum \frac{(O - E)^2}{E}\]

For all cells in the contingency table, take the difference between observed and expected count, square it, divide it over the expected count, and add them all together. That’s your chi-squared statistic. In our case, that produces a chi-squared statistic of about 12.965.

Expected Counts in Table IX in Sample (1997) For No Association
	Arms Race	No Arms Race
War	(14 - 6.393)^2/(6.393) = 9.051	(17 - 24.607)^2/(24.607) = 2.351
No War	(39 - 46.607)^2/(46.607) = 1.241	(187 - 179.393)^2/(179.393) = .322

We could also check our work by doing it in R. Notice we’re going to capture the chi-squared statistic that Sample (1997) reported in her Table IX, give or take some rounding for presentation.

etl <- (31*53)/257
ebl <- (226*53)/257
etr <- (31*204)/257
ebr <- (226*204)/257

etl; ebl; etr; ebr
#> [1] 6.392996
#> [1] 46.607
#> [1] 24.607
#> [1] 179.393

csctl <- (14 - etl)^2/(etl)
cscbl <- (39 - ebl)^2/(ebl)
csctr <- (17 - etr)^2/(etr)
cscbr <- (187 - ebr)^2/(ebr)

csctl; cscbl; csctr; cscbr
#> [1] 9.051548
#> [1] 1.241584
#> [1] 2.351628
#> [1] 0.3225684

chisq <- csctl + cscbl + csctr + cscbr

chisq # rounding above is off 2/1000ths or so, but that's no biggie.
#> [1] 12.96733

We’re not quite done yet because we need to relate this test statistic to what we might expect to observe under some kind of distribution. So-called the chi-squared test, the test statistic is related to the chi-squared distribution. The chi-squared distribution is a unique distribution of the sum of squared standard normal variables with just a single parameter: the degrees of freedom. It determines the number of independent standard normal variables to sum. For a single degree of freedom, this distribution will have a clear right tail. More degrees of freedom result in distributions that look more “normal.” Observe:

set.seed(8675309) # Jenny, I got your number...

nobs <- 10000
tibble(`rchisq(df = 1)` = rchisq(nobs, df = 1),
       `rchisq(df = 2)` = rchisq(nobs, df = 2),
       `rchisq(df = 5)` = rchisq(nobs, df = 5),
       `rchisq(df = 10)` = rchisq(nobs, df = 10),
       `rchisq(df = 25)` = rchisq(nobs, df = 25),
       `rchisq(df = 50)` = rchisq(nobs, df = 50))

In the case of the chi-squared test, the degrees of freedom is equal to the product of the number of rows (minus 1) and the number of columns (minus 1). For a simple 2x2 contingency table, that’s 1. Inference here works in the same way it does most other applications. Higher test values (lower p-values) will indicate greater incompatibility with the distribution, suggesting a rejection of the null hypothesis of no association. You can do this in R with the pchisq() function.

pchisq(chisq, df = 1, lower.tail = F)
#> [1] 0.0003169742

Or, better yet, just do all this in R with the chisq.test() function. The only real headache with this is you’ll have to teach yourself how to deal with matrices in R since that’s what the chisq.test() function generally wants.⁵ Here’s how you do it without Yates’ continuity correction (correct = FALSE), which is what these analyses were doing.⁶

# The observed counts, as a single vector
c(14, 39, 17, 187)
#> [1]  14  39  17 187
# In matrix form, using nrow = 2. When doing it this way, R files column by 
# column. That means the first column gets the first two values, which is 
# incidentally what we want. There are other ways of doing this, though.
matrix(c(14, 39, 17, 187), nrow = 2)
#>      [,1] [,2]
#> [1,]   14   17
#> [2,]   39  187

# Chi-squared test with*out* continuity correction.
chisq.test(matrix(c(17, 187, 14, 39), nrow = 2), correct = FALSE)
#> 
#> 	Pearson's Chi-squared test
#> 
#> data:  matrix(c(17, 187, 14, 39), nrow = 2)
#> X-squared = 12.967, df = 1, p-value = 0.000317

A Re-Analysis Using Newer Data

mmb_war is in the forthcoming version 1.7.0 of {stevedata} and offers a re-analysis of this using newer data. Here’s a glimpse into the data here.

Data <- mmb_war
Data
#> # A tibble: 2,324 × 9
#>    ccode1 ccode2 tssr_id micnum  year dyfatmin dyfatmax sumevents   mmb
#>     <dbl>  <dbl>   <int>  <dbl> <dbl>    <dbl>    <dbl>     <dbl> <dbl>
#>  1      2     40     130    246  1960        0        0        20     0
#>  2      2     40     130     61  1962        0        0        35     0
#>  3      2     40     130   2225  1979        0        0         2     0
#>  4      2     40     130   2972  1981        0        0         4     0
#>  5      2     40     130   2981  1983        0        0         1     0
#>  6      2     40     130   3058  1983        4       25         2     0
#>  7      2     40     130   2742  1986        0        0         2     0
#>  8      2     70      32   1554  1836        0        0         5     0
#>  9      2     70      32   1553  1838        0        0         4     0
#> 10      2     70      32   1556  1839        0        0         2     0
#> # ℹ 2,314 more rows

I take inspiration from Gibler et al.’s (2005) analysis on arms races and war and follow their lead in recreating a data set on arms races that I merge into this data set on all dyadic confrontations among strategic rivals.⁷ Importantly, I’m looking for whether there was a mutual military build-up that started and was ongoing prior to the start of the confrontation, in addition to whether the confrontation in question escalated to the point of dyadic war (i.e. whether the minimum dyadic fatalities surpassed 1,000).

First, let’s create a war measure that equals 1 if the minimum dyadic fatalities surpassed 1,000.

Data %>% mutate(war = ifelse(dyfatmin >= 1000, 1, 0)) -> Data

Now, let’s create a contingency table of the mutual military build-ups and the dyadic war.

table(Data$war, Data$mmb)
#>    
#>        0    1
#>   0 2049   75
#>   1  180   20

Invoking the table() function here with 0s and 1s is a bit clumsy for what I want, but should be read as follows. The row labels of 0 (no war) and 1 (war) relate to the first vector (i.e. the war dummy). The column labels of 0 (no mutual military build-up) and 1 (mutual military build-up) relate to the column labels of the mutual military build-up dummy variable. If we flip things around a bit, we’d note that 20 of the 95 observations of dyadic confrontations succeeding mutual military build-ups reuslted in war (21%). 180 of the 2,229 dyadic confrontations without a mutual military build-up preceding them reached fatality thresholds we could classify as a war. That’s about 8%. It looks like a significant difference, but we’d have to see what the chi-squared test says.

# Chi-squared test with continuity correction.
# Fyi: you could read in the vector to matrix() as c(2049, 180, 75, 20) and it'd
# be the same thing. I just prefer to see it this way.
chisq.test(matrix(c(20, 75, 180, 2049), nrow = 2), correct = TRUE)
#> 
#> 	Pearson's Chi-squared test with Yates' continuity correction
#> 
#> data:  matrix(c(20, 75, 180, 2049), nrow = 2)
#> X-squared = 17.895, df = 1, p-value = 2.335e-05

If there were no association between the arms races and war, the divergence between what’s observed and what we’d expect results in a chi-squared statistic so wildly incompatible with what we’d expect for a chi-squared distribution with a single degree of freedom. We reject the null hypothesis of no association and instead suggest the results are more compatible with an association between arms races and escalation to war.

Conclusion

This is mostly for the kids to give them something to do in a lab session for quantitative methods. It’d be nice for students in my program to get more of the boilerplate quantitative peace science stuff that I got in grdauate school. It would be nice for them to further contextualize why a lot of so-called “realist” talking points are poorly stated. It’d be nice for them understand more of John Vasquez. It would further be great to get them to do stuff in the R programming language. This particular application offers all of that. There are more advanced ways of doing this, and it asks a lot for logics derived from observations of boats (or even nuclear weapons) to map to the current system. Still, it’s classic stuff with classic methods. Students should know this stuff anyway.

It’s a system of linear equations! It’s an elementary way of thinking about it but does well to represent the phenomenon in a basic form. ↩
Dating the tract from which the “Para Bellum” quote originates is tricky and it’s conceivable Singer is referring to what might be understood in modern terminology as “intra-state” or “extra-state” violence. Certainly, the Roman Empire of this time precedes the modern state system. No matter, it shouldn’t be lost on the reader that this would still mean an emperor mobilized a bunch of hammers and now needs to find nails somewhere. The relationship between external threat and non-democracy is quite robust. Such a mobilization makes state-sanctioned violence cheaper than it would be in the absence of such a mobilization. ↩
Brandon Valeriano, may he rest in peace, told me once about a conversation he had with Wallace about the operational details of his analysis and how exactly he arrived at such stark conclusions in an era of punch-card analyses. It’s informative but I don’t think I can share them here. Wallace passed away in 2011. ↩
It is interesting that a lot of the empirical critiques of Wallace mostly (and rightly) say Wallace was way off what the true relationship is, but don’t really vindicate the “para bellum” hypothesis that they should lead to peace. ↩
The operative arguments you’ll want to learn are nrow, ncol, and byrow. Observe, for example, the difference between matrix(c(1:4), nrow = 2) and matrix(c(1:4), nrow = 2, byrow=TRUE) for a simple 2x2 matrix. Be mindful that the chi-squared test is agnostic about what are rows and what are columns because it largely hinges on the multiplication of row totals and column totals. Transposing a matrix would result in equal chi-squared statistics (e.g. chisq.test(t(matrix(c(1:6), ncol = 2))) and chisq.test(matrix(c(1:6), ncol = 2))). I work on the convention from Pollock III (2016) that things informally understood as “causes” should be columns and things informally understood as “outcomes” should be rows in creating a cross-tabulation like this. In our case, war is understood as an outcome of arms races. However, the particular method here has the same limitations as Pearson’s r. Use it with that in mind. ↩
R’s default behavior sets Yates’ default continuity correction to TRUE. This is the more conservative approach to a chi-squared test statistic because it will generally push down the test statistic for smaller samples to avoid Type I errors. ↩
You can see the details section of the codebook for a conversation about some case exclusion rules I employed along the way. I’m still tinkering with this military build-up measure and do not offer it here to be used uncritically. ↩

Global Governance and the International Politics of Standard Reference Temperature

Tue, 05 Aug 2025 00:00:00 +0000

: Think the meter is kind of boring? It was the subject of an interesting political fight in the early 1900s about *when* something is a meter.

I’m writing this because I have to get ready for the semester coming up. My department’s second sequence in its first-semester track is a discussion of global governance. This is kind of a challenge for this particular international relations scholar, who is more interested in talking about international conflict. It’s also very much rooted in the -isms. Neither of those are my cup of tea.

I found a way to make this somewhat interesting to me. Clearly, questions of human rights and the environment are more important in the study of international cooperation. However, “global governance” people caution that the governance of the world is more than just the really big stuff that appears before the United Nations. Indeed, global governance pervades our lives in some mundane ways we take for granted, like the standardization of basketball, signals of creditworthiness, and even our domain names. If you know where to look, you’ll find global governance stories everywhere in ways that can be quite interesting. Even in the meter.

Yes, the meter and, to be more specific, when something is a meter long. Defining the meter was subject to an interesting political fight at the end of the 19th century and beginning of the 20th century. There are important technical and scientific questions here, but there’s also a political story you could shoehorn into an international relations lecture as I do. What follows is a fleshed out version of the bit in lecture I give to students. I’m writing it here so I can point students here after lecture.

There will be some R code here, using these packages.

library(tidyverse)     # for most things
library(peacesciencer) # for using CoW's NMC data
library(stevethemes)   # for my custom themes
library(kableExtra)    # for a table

theme_set(theme_steve())

Importantly, I want to draw attention to this article by Ted Doiron (2007) in the Journal of Research of the National Institute of Standard and Technology. I’ll make ample reference to this article for details and specifics. Doiron is providing the important historical and technical information to support this post of mine. I’m just largely telling a political story on top of it and communicating its suitability to an international relations audience.

Here’s a table of contents.

Some Background on Measurements, the Meter, and Standard Reference Temperature
The Political Fight(s) over Standard Reference Temperature
The So-Called “-isms” and the Global Governance of Standard Reference Temperature
Conclusion

Some Background on Measurements, the Meter, and Standard Reference Temperature

The story here has to assume some background. By this point in a lecture, I will have introduced students to the International Bureau of Weights and Measures (BIPM). This is one of the older inter-governmental organizations (IGOs) in the Correlates of War data. It is the seventh oldest by formation and fourth oldest among those still in existence (behind the Central Commission for the Navigation of the Rhine [CCNR], the International Telecommunications Union [ITU], and the Universal Postal Union [UPU]).

Here, for example, are the first ten IGOs in the CoW data.

cowIGO <- haven::read_dta("~/Koofr/data/cow/igo/3/igo_year_format_3.dta") 
# ^ data current as of 2014
cowIGO %>%
  mutate(lastyear = max(year), .by = ioname) %>%
  arrange(year) %>%
  slice(1, .by=ioname) %>%
  select(ioname:year, lastyear)

The First Ten IGOs in the Correlates of War IGO data (v. 3)
IGO (Abbr.)	IGO Name	Year Formed	Last Year (2014 = Ongoing)
CCNR	Central Commission for the Navigation of the Rhine	1816	2014
SCH	Superior Council of Health	1838	1914
ECCD	Euro Comm for Control of Danube	1856	1939
ICCSLT	Intl Comm of Cape Spartel Light	1865	1958
ITU	Intl Telecom Union	1865	2014
UPU	Universal Postal Union	1874	2014
BIPM	International Bureau of Weights Measures	1875	2014
IPentC	Intl Penitentiary Comm	1875	1951
IUPR	Intl Union of Pruth	1878	1914
ICPTU	Intl Conf Promoting Tech Unification	1882	1938

I hope it’s not lost on the reader, or my students, that there is no telling a story about the rise of intergovernmental organizations without telling a story about the consequences of war (CCNR, and see also: the European Commission for Control of the Danube [1856-1939]) or advances in technology that necessitate some kind of coordination about things that we might take for granted. Who among us sweats the details of telegraphic networks (ITU), international mail delivery (UPU), or, in this case, what exactly is a meter or gram (BIPM). However, these are important details to sweat as they emerge. In the context of meters and grams, they are critical details to sweat with the rise of global trade in the 19th century. If I’m paying to import something that has some specified length, we should all agree on what exactly the length is. That was the initial mission of BIPM, a France-hosted IGO for coordinating measurement standards of the International System of Units.

: The Meter Bar (National Prototype No. 27), as received by the United States in 1890

The follow-up question of “what is a meter?” has its own interesting historical trajectory by the end of the nineteenth century. In fact, the history of measurement is its own fascinating Wikipedia entry and those of us who had to read the Bible in our youth remember some oddball measurements mentioned in those texts. However, it’s the French who largely pioneered the measurements we use today. The formation of the meter is based around an aim to standardize measurements to natural phenomena. In the case of the meter, it was originally understood as one ten-millionth of the shortest distance from the North Pole to the equator for a line passing through Paris. This definition changed through time and with more precise measurements. By 1889, the benchmark for the meter became a particular bar made of platinum and iridium. Formally, the meter was the length of the bar you see to the right 1) supported by two cylinders of a diameter of at least one centimeter, 2) placed 571 millimeters apart, and 3) measured at the melting point of ice (0 °C, or 32 °F).¹

It’s the third component that will likely stand out to the reader and student. Why this temperature? This has two answers. First, remember the principle of thermal expansion you learned about in school. Most things generally expand when heated and contract when cooled (water is the most obvious exception). Each element has some kind of coefficient of thermal expansion you’d have to know. Long story short: there is no knowing how long a meter is without knowing the temperature for which something is a meter long. Second, consider the time. Thermometry was somewhat unreliable at the time and only two measurements were known with some kind of precision: the boiling point of water and the melting point of ice. Of the two, the melting point was more easily reproducible because it is 3,700 times less dependent on atmospheric pressure. As a technical matter, it would be the only standard reference temperature that would make sense at this time.

The Political Fight(s) over Standard Reference Temperature

Given popular conceptions about measurement in the United States, the role of the United States as protagonist in this story about the meter and the change to 20 °C as the standard reference temperature is prima facie bizarre. No matter, that’s the role I’ll argue it served here.

The use of 0 °C as the standard reference temperature irked bureaucrats and metrologists in the United States for many reasons. The United States was a charter member of BIPM and one of the original 17 signatories to the Metre Convention. It definitely had its customary units in place come 1832, but it was aware of how conspicuous its use was relative to measurement in Europe. The Metric Act of 1866 protected the metric system especially because manufacturers in the United States were targeting export markets that would never use the U.S. customary units. Thus, the United States was never against metric measurements in this fight it had with the BIPM. Instead, it had legal and administrative requirements imposed on bureaucrats and metrologists to standardize and replicate the meter relative to its customary units.

From the American perspective, 0 °C was strange for a standard reference temperature. Nothing of interest to export-oriented manufacturers in the U.S. at the time was made in freezing conditions. Some things, like steel and brass, have very different coefficients of thermal expansion that would wreak havoc on assembly if assembly were happening at 0 °C to be used or re-assembled at some kind of room temperature. Why not use room temperature under those circumstances? American efforts to reproduce the meter at something analogous to room temperature (let’s say: 20 °C) required understanding how their reproductions squared with the real thing at 0 °C. Domestic law somewhat impelled the U.S. metrologists and bureaucrats to verify their replication for the sake of metrification.

However, approaches by metrologists and public officials from the United States to the BIPM to check their work were met with a cold shoulder. Consider this exchange from Charles-Édouard Guillaume, secretariat of the BIPM, to Samuel Wesley Stratton (pp. 2-3):

But comparisons cannot be made with International Prototype. The International Prototype Meter as well as the kilogram, and their certificates, are shut up in a depository, which is under the charge of the International Committee, and closed by three locks, one key of which is in my hands, the second is deposited in the Archives of France, and the third is in possession of the President of the Committee, Prof. Foerster at Berlin. The depository which is a deep cave under our laboratory, is inaccessible to me as well as to all the world. It cannot be opened and much more the prototype can not be taken out except by a decision of the Committee in session.

The BIPM wanted this issue to go away and pleaded practicality to advocate a philosophical position about the inherent rightness of 0 °C. Interestingly, the BIPM tried to shut up the Americans once and for all at its 29th meeting in 1913. Therein (p. 90), the BIPM insisted on the melting point of ice as the standard reference temperature for defining the meter. However, this wasn’t enforced in light of events to follow. The BIPM, which had met every two years prior to this point, would not convene again until 1920 for obvious reasons.

There were both technical and bureaucratic reasons for why the move from 0 °C to 20 °C stalled. Again, thermometry was far more uncertain at the end of the 19th century than it is now and there were only a handful of replica bars circulating around the world for Americans to benchmark their meter for their own legal requirements. Doiron’s (p. 3) treatment here suggests a schism between the practical demands of industry and metrologists in the United States to the philosophical aims of BIPM scientists on this matter. It’s the difference of “this isn’t useful” versus “this is right”. Doiron’s treatment doesn’t mention this, but there’s an air of a power play by France as well as their metrologists were adamant about 0 °C. They had the bar, and perhaps the philosophical was practical for them. They could more easily reproduce replications of the meter (their meter) at 0 °C. Metrologists and public officials in the United States had their own political pressures, and political aims. So did the French and the IGO it housed.

No matter, it seems that only France and the experts at the IGO it housed held to this position of 0 °C for standard reference temperature. Elsewhere, the Brits proposed a standard reference temperature of 62 °F· The roles have definitely reversed, but the metrologists of the United Kingdom were far more adamant about the use of Fahrenheit for standard reference temperature than American metrologists were at the time. From a correspondence from Stratton to his counterpart in the United Kingdom about the British proposal (pp. 4-5):

Referring to your letter of August 17, 1916, in regard to a suitable standardization temperature for commercial metric standards of length, I have to say that we have carefully read Mr. Sears’ memorandum, and while we agree with him that commercial standards of length, whether metric or English, should be standardized at the temperature at which they are to be used, we do not concur in his opinion that the mean workshop temperature to be se1ected should be 62 °F.

There is, at the present time, a decided tendency away from the Fahrenheit temperature scale, and we feel that the tendency should be encouraged. There is, in fact, a bill now pending in Congress by which it is hoped to abolish the Fahrenheit scale, at least from Government publications.

The temperature 20 °C is coming more and more to be accepted as the standard temperature for industrial as well as scientific operations. The sugar industry, for example, is practically on the 20 °C basis. All polariscopic tubes, flasks, etc. used in making up sugar solutions are made standard at that temperature. Very many hydrometers are standard at this temperature and the glass volumetric apparatus standardized by this bureau is on that basis and has been for the past ten years or sore. Also many of the steel tapes used in this country are standard at 20 °C.

I might add many other examples to show that 20 °C is being largely accepted as the standard temperature in scientific and technical work. Would it not, therefore, under the circumstances, be better to standardize both the English and metric commercial standards on this basis rather than that of 62 °F? 20 °C would certainly have a very great advantage over 62 °F if urged for international adoption; and from a practical point of view it would be no more difficult to change the English commercial standard from 62 °F to 20 °C (68 °F), than to change the metric standards from 0 °C to 16.67 °C (62 °F).

By the 1910s, there really wasn’t a standard reference temperature as much as there were standard reference temperatures for different industries and different countries. I’ll admit the specifics for why this was the case are beyond me. Whatever reasons various industries had at the time for different standard reference temperatures assuredly has a logic that eludes this political scientist. However, the years that followed eventually saw a renewed interest (from the American perspective) in standardizing this measurement and new measurements given ongoing electrification. American industry experts, bureaucrats, and metrologists wanted to standardize more measurements and expand the scope of the BIPM beyond what it had been doing to this point (before 1913). Ideally, it would further standardize international standards to what the United States had been doing through its Bureau of Standards: measuring at 20 °C.

The Bureau of Standards prepared for the upcoming 1927 general conference of the BIPM by shoring up a united message from various industries about the value of 20 °C for standard reference temperature. The French position at this time seemed to concede defeat on 0 °C but wanted a bit more of a convoluted approach to thermal correction than anyone else would accept. The American position (p. 10) was “absolutely not” and the German position was, paraphrased, “hell no.” Whether the latter’s opposition was a function of the ridiculousness of the proposal or some animosity about the fallout of the first world war is not something I feel I can answer.

Meanwhile, the British were still advocating for 62 °F, but found an appeal to Fahrenheit a tough sell. To insert a Swede into this conversation, Carl Edvard Johansson had commanded a lot of intellectual currency to this point—especially in the United States—with his gauge blocks. He argued that his increasingly popular gauge blocks were more durable at 20 °C than they were at 62 °F, at which point body heat exerts a stronger influence on the performance of the gauges. Further, standardizing between Celsius and Fahrenheit was easier at 20 °C than it was at 62 °F. 20 (Celsius) = 68 (Fahrenheit). 62 (Fahrenheit) = ~16.67 (Celsius). If you’re going to standardize, just do what the Americans are doing.

Thus, what was a massive fight at earlier meetings of the BIPM and correspondence on this issue was settled with a voice vote in 1931. The standard reference temperature for industrial measurements became 20 °C. In what I think is the niftiest epilogue to this story, this 1931 voice vote on standard reference temperature has the distinction of being the very first international standard introduced by the International Organization for Standardization. That’s right; it’s ISO 1. People in my field might recognize ISO 3166 for country codes. You might have seen various ISO standards, like ISO 8859 for character encoding. ISO 8601 largely shapes how you see dates and time. None of those are the first post-war international standard. Instead standard reference temperature is ISO 1, an example of the kind of global governance of mundane things done through a non-governmental organization of domestic standards agencies.

The So-Called “-isms” and the Global Governance of Standard Reference Temperature

Students in my field are induced to think of things through the so-called “-isms”. I wish they wouldn’t? No matter, it’s how they’re induced to think and it’s how they’re inclined to think. You can tell an international relations story about this through these various -isms. Not that you should, but you could.

“Realism”

You know how I think. These people are clowns that have nothing to tell you about a question you should care to ask. Carr, Mearsheimer, Morgenthau, Walt, and Waltz would have no real explanation for this story, nor any capacity to care. They’re not particularly capable of explaining the things they like to talk about, let alone things that involve coefficients of thermal expansion, atmospheric pressure, and thermometry. That said, here are the things they would assuredly like to talk about with respect to this.

Realists generally start from an orientation toward the world as one of conflict, and not cooperation. The impetus for an intergovernmental organization like BIPM might have been to harmonize measurements in order to cooperate, but it fell to inherent dysfunction even on what could be dismissed as a “low politics” issue.² How “tragic” is that, from a realist position, that an IGO on such a “low politics” issue could fall to such dysfunction? It would make sense if you like to talk about the “false promise” of IGOs.

Unsurprisingly, war pervades why this dysfunction emerged and, arguably, serves as the backdrop against which it was ultimately resolved. Great power war put meetings of the BIPM on hiatus. Countries had their own national standards to which they needed to adhere for manufacturing and the war effort. Harmonizing or changing practices would have been an inefficient use of energy at this time, especially when national security is at stake. The end of the first world war (prior to the 1927 and 1931 meetings) and certainly the end of the second world war (prior to ISO 1) definitely accelerated the ascent of the United States as arguably the most powerful country on the planet. It’s not terribly surprising that the U.S. (from this perspective) threw enough weight around to resolve this issue. Take a look at these assorted measures of “power” from the Correlates of War project. I’m standardizing iron and steel production and primary energy consumption to be proportional to the system total in a given year.

cow_nmc %>%
  mutate(sumirst = sum(irst, na.rm=T),
         sumpec = sum(pec, na.rm=T),
         .by = year) %>%
  filter(ccode %in% c(2, 200, 220, 255)) %>%
  filter(between(year, 1875, 1935)) %>%
  mutate(pirst = irst/sumirst,
         ppec = pec/sumpec) %>%
  select(ccode, year, pirst, ppec, cinc) %>%
  gather(var, val, -ccode, -year) %>%
  mutate(country = countrycode::countrycode(ccode, "cown", "country.name")) %>%
  mutate(var = case_when(
    var == "pirst" ~ "Iron and Steel Production",
    var == "ppec" ~ "Primary Energy Consumption",
    var == "cinc" ~ "Composite Index of National Capabilities (CINC)"
  )) %>%
  mutate(var = fct_inorder(var)) %>%
  ggplot(., aes(year, val, color = country, linetype = country)) +
  facet_wrap(~var) +
  geom_line(linewidth = 1.1) +
  scale_x_continuous(breaks = seq(1875, 1935, by = 10)) +
  labs(x = "", color = "", linetype = "", y = "Proportion of System Total",
       title = "Various Measures of 'Power' for Four Important States, 1875-1935",
       subtitle = "Pax Americana may have only started after World War II, but it became a clear powerhouse earlier than that.",
       caption = "Data: Correlates of War National Material Capabilities (v. 6), by way of {peacesciencer}.
       Iron and steel production originally measured in thousands of tons. Primary energy consumption originally measured in thousands of coal-ton equivalents.")

Perhaps realists don’t care about standard reference temperature and what that means for the history of the meter, but they did co-opt hegemonic stability theory to be one of their things. The U.S. had weight to throw around after the first world war, and used it accordingly.

At least from this perspective, if you had to talk about it. It’s still asking a lot for “realism” to have anything to say here. It’s primarily interested in great power war, not the bureaucratic minutiae of IGOs and fights over where the temperature dial should be to measure things.

“Liberalism”

It’s tough to say what “liberalism” is in international relations. I know I have a PhD in this, but I’m often confused to what “liberalism” is. No matter, I think this is generally what students get on “liberalism”, at least as I would encourage them to see it.

Where so-called “realism” conceptualizes an inherently conflictual system of sovereign states, so-called “liberalism” conceptualizes an international system of more than just sovereign states that is governed as much (or more) by its transactions as its conflict. A narrow focus on state-to-state violence betrays the full scope of international relations. Truly, anything spanning borders is “international” and coordination on measurements and standard reference temperature is “relations” of a kind. A realist might not care about this, but could bullshit you something if pressed. However, a so-called “liberal” would see a lot to like here given how they view the world.

For one, read Doiron’s article and tell me if you recognize any of those names. Be honest, too. If you’re a nerd or versed in the history of Swedish contributions to science, you might recognize Carl Edvard Johansson. However, he’s not a head of state, a defense minister, or a foreign minister. He was an inventor and scientist working for the Ford Motor Company at this time. You might recognize Herbert Hoover, if you’re an American. He would be president soon in this story, but notice he wasn’t. He was the Secretary of Commerce at this point in the story. Realists famously blackbox the state and might make an occasional reference to a top-level diplomat or the state leader, neither of which fit Herbert Hoover in his role in the American government at this point.

Instead, you’re seeing a bunch of actors in a bunch of roles you don’t know about, doing things on an international scale that would’ve shaped your life you were alive at this point. You’re seeing somewhat faceless bureaucrats working in domestic standards agencies like the American Bureau of Standards or its British equivalent. You’re reading about nameless, faceless experts in the BIPM and the authority vested in its secretariat to tell American bureaucrats to pound sand. Go look at Doiron’s list of actors comprising the U.S. industry consensus on p. 8 of its article. None of those consist of the Department of War in the United States.

The familiar kind of actors permitted entry into a so-called “realist” story are nowhere to be seen here. There is no Woodrow Wilson or Robert Lansing. There is no Kaiser Wilhelm or Arthur Zimmermann. There is no David Lloyd George or Sir Edward Grey. There is no Raymond Poincaré or Théophile Delcassé. No one in that kind of position would have that kind of expertise. Thus, they outsource something important-but-obtuse like this to domestic non-state actors to figure it out. You wouldn’t get that in “realism”, but you’re going to get a better picture of it in “liberalism”. Indeed, I largely glossed over how much of this big ol’ snafu is a function of American domestic politics. This started because American bureaucrats and metrologists needed to satisfy legal requirements imposed by Congress.

“Constructivism”

This is a challenge. Mr. Constructivism himself would say “constructivism is not a theory of international politics.”. That’s good as contextualization for what it ultimately is. Constructivism is primarily rooted in the sociology of knowledge, and is foremost interested in the ideational forces that construct the world and give it its meaning.

There are several things here that would work well with a constructivist approach to understanding international relations, broadly understood. Notice there is a great deal of “learning” here. Yes, there is actual technical learning as actors at this time better hone their measurements. However, there is also active learning and an appeal to “experts” to “teach” and “inform.” Carl Edvard Johansson played an important role as a kind of educator informing stakeholders why 20 °C is a better reference temperature in relation to his gauge blocks. There is important teaching going on as the United States sent a battery of industry representatives to persuade other voting members of the BIPM of the “rightness” of its position. There needed to be active educating of the voting members of the BIPM why the French position at the 1931 general conference was a non-starter. Constructivism, as it’s done in international relations, isn’t too far afield from the academic study of education and learning. There is educating and there is persuasion happening in this story, often done by people with certain roles and identities.

Admittedly, this kind of framework isn’t my forté. But you can flesh this out more if you had the spoons for it. The meter is, after all, an interesting social construction of the fraction of a line passing through Paris. We definitely construct the world around us through various means, even if the original idea is that the meter is based on “natural phenomena.”

Conclusion

I wrote this to flesh out something that is a part of a lecture I have to give every semester. The academic interest in global governance concerns some major questions about the movement of people, their basic human dignity, and the climate that they all share. However, this academic interest in some of our most salient and urgent questions comes with an appreciation of how this manifests in some mundane ways we all take for granted. None of us may care so much for the nitty gritty details of what a meter is, or the conditions for which (historically) something was a meter long. But that doesn’t preclude the meter from a conversation of global governance. It most certainly is part of global governance—arguably one of the earliest avenues of global governance. It just means we’re disinclined to think about it and miss some interesting international politics around the conditions for which something historically was a meter long.

There is an important technical and scientific issue underpinning the fight about standard reference temperature, none of which are necessarily political. Thermal expansion is effectively a natural phenomena for which a “social construction” of it would be kind of silly. Further, measuring temperature was an uncertain enterprise around this time, beyond some temperatures that were known with relatively more precision. No matter, even this phenomena had an interesting political fight about what was an appropriate course of action. If you know where to look, or how to look, you’ll find these interesting stories.

You could further contextualize these stories around the familiar “-isms” that have currency in an intro-level international relations track. Not that you should, but you could. I would ask for a bit more by way of explanation for how anarchy (a constant) can explain varying attitudes about standard reference temperature, even if great power war is a clear backdrop to this debate. I would ask for a bit more clarification about the underlying domestic politics motivating the U.S. position and French position, even if you could clearly see the kind of endowment advantage and authority that the BIPM had early into this debate. I would further ask why you’re bringing so-called “constructivism” into this in the first place, even if that particular “-ism” has a unique position as being an offshoot of the sociology of knowledge relative to other paradigms in international relations. I might ask that you theorize the political more than any of these particular “-isms” are willing to do as basic orientations toward international relations.

No matter, anything spanning borders is “international relations”. Yes, even—and especially—standard reference temperature. It’s a cool-as-hell epilogue that it’s ISO 1 if you’re interested in global governance.

For the sake of presentation, I will pass over the formalization of uncertainty that went into this bar-based measurement in order to better tell the political story I care about. Measurement of the bar in question included some a measure of uncertainty that, importantly, was subject to surface temperature. Perhaps it’s no surprise that the modern definition of meter anchors to time, not temperature. Time is the most precise thing we can measure. ↩
To the best of my knowledge, we have Stanley Hoffmann to blame for this distinction that has since infected how we teach international relations to students. He may have used it before his 1966 publication in Daedalus, though that is the first usage of the terms for which I’m aware. Hoffmann’s treatment of “high politics” and “low politics” has more contours than later invocations of these terms, but it should irk curious students all the same to use these terms in the context of a proposed explanation (“explanation”) as means to dismiss the importance of a question. Like, what even is that… ↩

Cloud Storage, European Style

Mon, 19 May 2025 00:00:00 +0000

: Who's to say the separation doesn't play out like the end of this movie, but...

The current moment has led to a spate of enthusiasm in the Western world, outside the United States, to wean itself off American products in favor of European (or Canadian) alternatives. The underlying rationale for why this movement emerged is tragic. The weight of the moment and the movement it’s generated is heavier than any individual-level consumption decision, and it’s more than silly, simple jingoism that motivates its American equivalent. However, the weight of the moment does underscore the weight of the undertaking for anyone who wants to take part. Like moving a grand piano from one home to another, it’s an undertaking easier said than done.¹

I’ve been a Dropbox user since 2008 and just about my entire life’s work has revolved around it for as long as I’ve had it. However, it’s time for a change for any number of reasons. Dropbox is a great product, but privacy concerns for an American company accountable to the CLOUD Act are real. It’s also just a tad bit expensive for what I need. I need a family plan with at least a terabyte, which is available only for Americans. As of writing, that’s over $200 every year. That’s a good deal, all things considered, but it’s not the best deal if you intend to have just one cloud storage account (more on that later). It’s also charging me dollars I don’t have in abundance and would have to wire into an American bank account from Sweden. Given the aforementioned moment, it’s time for a switch.

Here’s what I did, and will commit to doing for the time being. Perhaps this will help a few of you.

Switching from Dropbox to Koofr

I experimented with a few alternatives and settled on Koofr as my main cloud storage/productivity tool to replace Dropbox. Koofr is a Slovenian company and has data centers across the European Union. It’s also been around since 2013. If you’re looking to make the switch from Dropbox, it’s good to commit to a service that’s not a startup because the older companies more credibly signal staying power. That’s something you don’t appreciate until you see it in print. Dropbox (at least since it went public) wasn’t profitable until 2020.

If you’re using Dropbox as I did (i.e. for file-sharing, shared directories, and general academic productivity across multiple devices), you’ll find that Koofr does literally all of that. It has a built-in Microsoft Office app in the browser if, God help you, you’re into that sort of thing. It has file-versioning as well. There are some other perks I want to emphasize as well.

It can connect to a host of external cloud services to facilitate the transfer of your files. I’ll note you could do this and it would be the path of the absolute least resistance, but you would lose your timestamps as Koofr makes copies of the data. That would drive me up the wall, but perhaps a few of you may not mind it as much.
Koofr boasts that the only tracking tool they employ is a simple cookie to keep you logged in. It doesn’t advertise “zero knowledge” or end-to-end encryption (E2EE) like some other alternatives, but the respect for privacy is discernible (beyond the obvious GPDR commitment).
Koofr compensates for the lack of zero-knowledge/E2EE by giving you a few so-called “safe boxes” (i.e. its “vault” feature). If you understand how it works, it’s helping you do something you could do yourself. It’s actually quite refreshing for a company to be up front about that. It’s offering a service while pleading with you to do it yourself, and showing you how to do it.
Koofr also has WebDAV capabilities. There is an incredibly narrow intersection of people I know that 1) use Linux and 2) still rely on mp3s to be the basis of their music collection. That intersection might just have me in it, but it does mean I’m precluded from using iTunes and the default music app on my iPhone. Any alternative music player will have this functionality if you are going to populate with mp3s from a cloud storage company, and I kind of need my cloud storage service to provide it.

The biggest drawback for Koofr is that it is incidentally one of the more expensive options, if not arguably the most expensive option among the European alternatives. I don’t think the price is so bad, or at least I have a price threshold for which that suffices (even if it’s arguably not optimal). I can’t speak for its business model and if the price is a creative way to acquiring capital to finance its development. It would be one thing if Koofr were as feature-poor as some alternatives, but it’s quite feature-rich. Other options will definitely beat Koofr to the floor in terms of price, but Koofr offers more than those that are doing this. Perhaps Koofr is aware of this with its loyalty discount, upwards of half the given rate in a given year. I can’t speak to you all, but I hate moving literally anything. I can commit to this in the long run and in the long run it should be worth it.

Koofr also doesn’t have a family plan of which I’m aware and I’m unaware of any developments toward that end. I would love to keep my wife and I under one plan, but I have a backup plan for that.

This might also be a “me” thing, but Koofr has some default behavior I find a tiny bit cumbersome. Dropbox takes a lot of the guesswork from you on how to manage your files. You eventually get used to what it does and appreciate some of its features. You have a basic directory (i.e. “home”, if you’re on the browser) where everything is. Once something is in the cloud, you can choose to keep it there and off your computer. Koofr doesn’t quite work that way. On Linux, its default desktop sync will be at /home/[insertusername]/Koofr (which is good and fine), but appears as “My desktop sync” in the Koofr app and on the web (i.e. ewww).² You can adjust this but it’s palpable that Koofr doesn’t want you to do it this way. I’m also showing my age here, but spaces in file names/folders just irk me. As far as I understand it, Koofr also discourages you from thinking about doing a selective sync as Dropbox would do it. You would have to move files you want to archive outside of your desktop sync folder and to somewhere else on your account. It’s not the worst thing in the world (far from it), but it does mean my process had to belabor a bit more what I wanted to archive from day-to-day use. Perhaps that’s a good thing?

You would be under no obligation to click this link, but this would be a unique invite link to join Koofr and give it a shot.

Supplement Koofr with pCloud as the Backup’s Backup

One piece of advice I picked up scanning subreddits on cloud storage suggested you should have a backup to your backup, the extent to which a service like Dropbox was my backup. Perhaps that’s an external hard drive. Perhaps that’s some secure storage facility/service somewhere as another service. Perhaps it’s another cloud storage company. No matter, back up your backup.

For me, that was jumping on a “lifetime” family plan from pCloud for two terabytes. There was a particular fire sale happening on pCloud around Easter time where this 2 TB family plan was 400 Euros, and it’s conceivable that another fire sale might be forthcoming again. I splurged on it because I could, and you may find that in your interest as well.

To be up front, you should be skeptical of services that offer so-called “lifetime” plans as they are for the “lifetime” of the company. A start-up company has every incentive to sell these for a quick influx of capital. A start-up company that sells them then won’t be around that long to honor it. Keep that in mind, and perhaps that will scare you off, But, pCloud is a Swiss company with the European part of its operations in Luxembourg and has also been around since 2013. Other companies that offer those “lifetime” deals (e.g. Icedrive, Internxt) have been around half that time.

pCloud has a few superlatives for how I intend to use it. For one, its Linux app mounts as an external/virtual hard drive. That makes it less optimal for real-time collaboration/syncing and suitable for seeing what’s in what is going to be my archive whenever I want to see it. The syncing itself is quite fast, at least for uploading things. I rather like it’s functionality for playing music or sorting/displaying photos. For a one-off purchase, and the backup of a backup, I’m quite happy with it. If I were using it as my main cloud storage option, I would find some of its performance a bit questionable and would be left wanting about some of its functionality for collaboration. However, it’s a good place to stick stuff.

There are a few downsides worth calling out in particular. For one, pCloud’s customer support isn’t as responsive as Koofr or some of the other options. I didn’t find that terribly off-putting, but it can’t go unnoticed. The super privacy-focused user may find some of its privacy features only fine and express some reservation about the fact it has a data center in the United States. It vows it doesn’t have copies of your data across both jurisdictions, and I’ll trust that the case (given its incorporation in Switzerland).

Backup to the Backup with Rclone

Part of this transition has made me appreciate that I could be leaning a lot more on myself to do what I otherwise entrusted a Linux app to do for me. It meant learning more about Rclone and doing some of this myself.

I entrust the Koofr app on my desktop (laptop) to sync files to Koofr’s cloud, which then syncs with my laptop (desktop) when I’m working in the office (at home). On my desktop (my main computer), I setup Rclone to interact with pCloud largely following the instructions here. Then, I wrote a simple bash script to sync Koofr to pCloud, as follows.

#!/bin/bash
LOGFILE="/home/steve/Koofr/logs/$(date +%Y-%m-%d).log"

rclone sync /home/steve/Koofr/ pcloudconfig: \
    --exclude '**archive*/**' \
    --exclude '**[Tt]rash*/**' \
    --exclude 'pcloud-materials/**' \
    -v \
#    --dry-run \
    --log-file="$LOGFILE"

This process runs every day at 5 p.m, by way of crontab, syncing my pCloud account to square with the Koofr account. It excludes things that I’ve archived (--exclude '**archive*/**'), things that are detritus/”trash” files created by both services (--exclude '**[Tt]rash*/**'), and some startup documents that pCloud gives you when you create an account (--exclude 'pcloud-materials/**'). This script is verbose (-v) and creates a log file when it’s done that tells me what synced. As I was experimenting with its behavior, I would do a dry run (--dry-run) that informs me what it would synced/deleted. You might find that useful as you’re getting started.

Other Alternatives

Feel free to poke around and explore your options, but here are the other ones I strongly considered before finding what works for me.

Filen: Filen intrigues me. Its development team is lightning fast with support- or sales-related questions and seem genuinely very interested in expanding the scope and quality of what they offer. It already boasts zero-knowledge encryption. But, it has no family plan nor does it seem tailored to collaboration and general office productivity like some others (e.g. Koofr, Dropbox, Jottacloud). Its apparently working on a family plan, but does not have one right now. It is also a bit unforgiving with syncing. Past versions of a file count and assorted detritus count as used space. Beyond that, there is so much to like about what it offers.

Proton: Proton Duo would be a no-brainer for what it includes, if it included a Linux client for its Drive service. However, reviews I’ve encountered of its service are more sanguine about its “storage” capabilities and less its syncing for day-to-day use. I can’t speak for that, though.

Jottacloud: Jottacloud is fascinating to me. The price can’t be beat. It’s super fast. It has a command-line tool for Linux (if not a Linux app). It’s also Nordic. I think I wanted the WebDAV support, though. I might come back to this, though, especially if it offers a Linux app beyond showing me how to do things in the command line. Jottacloud does a lot of things underneath the hood that I rather like. Those of you further removed from Northern Europe may find it a bit unresponsive? I’ve read such complaints.

I’ve been a Linux enthusiast since 2004 and would cheerlead efforts for more robust and widespread proliferation among desktop users, replete with public backing for it. But I am a nerd, and best of luck to those wanting to quit Microsoft or Apple cold turkey. Google is entrenched, but solvable. Decent to arguably equivalent alternatives exist. I don’t think Microsoft is a solvable problem and the company has helped itself by being more reliable than the American government. I’d rather lend Microsoft money than the U.S. government. You could say something similar about the entrenchment of Apple even as its user base is a little more, shall we say, “dedicated.”. This is an opinionated take, but widespread public support for Apple exceeds the quality of the products they offer. The phone is better than the alternatives and its operating system is fine as a UNIX derivative. I’ll withhold further comment on what I think about its hardware and other services, though. ↩
Some other options will install a default folder to your home directory and sync from that. Ewwwww. At least Koofr doesn’t do that. ↩

Stephen A. Smith and International Relations Scholarship

Fri, 18 Apr 2025 00:00:00 +0000

: I used to be proficient at Photoshop in a past life. It's just easier for ChatGPT to do this for me instead (because it did).

Are you explaining it or are you just talking about it?

This is a question I routinely raise on student papers in my department in a curriculum that is wedded to the so-called “isms”. It’s pervasive here. Students think firmly inside those boxes. I don’t even want them thinking outside the box. I want them throwing the box away (or recycling it, whatever complies with local environmental regulations).

If I’m asking you to explain something or make sense of something, I want you to actually do it. If I’m asking you a question, it’s because I want an answer. Answer the question. I don’t want you restating it, because restating the question as an answer is begging it. I don’t want you answering a question with an answer to a related or unrelated question I did not ask. Answer the question that I asked. If I want an explanation, give it. If you’re not explaining something to me, you’re just talking about it. I don’t want you just talking about it; anyone can do that. I want you explaining it.

My go-to example for students at Stockholm University concerns a guy that maybe one or two people in the room have heard about. Several have been to the United States and a few (always the men) are aware of ESPN sports programming. No matter, this fellow is named Stephen A. Smith. He’s a “sports television personality” (we’ll go with that) who has become famous and wealthy for doing bombastic things with no real informational value. Shticks feature prominently into his repertoire, and one of his shticks is hating the Dallas Cowboys.

Observe, from this video six years ago with respect to the 2018-19 NFL season. Therein, the Dallas Cowboys finished 10-6, beat the Seattle Seahawks in the wild card game, but lost to the eventual Super Bowl runner-up Los Angeles Rams in the next game.

You may or may not have found that entertaining. You certainly didn’t find it informative of anything you may have cared about.

I don’t know or care to know why Stephen A. Smith detests the Dallas Cowboys, but let’s dive a bit into this. The 2024 Dallas Cowboys finished 7-10, 3rd in the division, and did not make the playoffs. I’m sure Stephen A. Smith enjoyed talking about this, but anyone can do that if they’re interested. How do we explain this disappointing season after a 2023 season in which Dallas finished 12-5 and won the division? To do that, you’d have to do more than just talk about it, or restate the season as disappointing. You’d have to understand a bit about the game itself, or at least try a little bit.

For example, offensive tackle—especially the left “blindside” tackle—is a premium for pass protection in a pass-heavy game. Tyron Smith, an 8x Pro Bowl tackle, left in the off-season for a final payday with the New York Jets and the Cowboys needed to use their first round draft pick on his replacement. Tyler Guyton, had a disappointing season and arguably played out of position as a left tackle. He was a right tackle, for the most part, at Oklahoma. QB pressures, as a count, increased almost 33% from 2023. Perhaps it was a mistake for the Cowboys to feature Ezekiel Elliott as prominently as they did. Elliott was an amazing player for the Cowboys in his first few years, but injuries and wear and tear have caught up with the nine-year pro. The Cowboys switched defensive coordinators from 2023. Dan Quinn relied on a lot of Cover 3 with man coverage coordinating the Cowboys defense, but complemented it with some Cover 1 and Cover 2 looks. Mike Zimmer, his replacement, dialed down the Cover 1 and Cover 2 looks and placed more emphasis on Cover 3 (which is a base look just about everywhere in the NFL) and Cover 6. Coach what you know, and there’s always a peculiar song and dance making an outside hire to replace a guy who optimized a system in a particular way. However, this change comes after losing 2023 cornerback Stephon Gilmore to the Vikings and playing most of the season without Daron Bland (the other corner). The switch in philosophy may have adversely affected Markquese Bell, who played further outside the proverbial tackle box than he did under Quinn. Bell had a disappointing 2024 season (just six tackles) before a season-ending injury after nine games.

I did not watch any of the NFL in 2024, so I can’t say any of these things with much confidence.¹ No matter, I feel I can offer them as at least partial explanations for why the Cowboys had a disappointing 2024 season (ignoring what other opponents, like what division rival and eventual Super Bowl champion Philadelphia Eagles, did to improve from 2023). It required effort. It required understanding basic strategies and logics that coaches employ, and knowing about the game itself. It willfully took something complex and distilled it into some simple parts that get at the essential features of the performance of an NFL team. It required contextualizing injury and attrition. It required understanding how sometimes the talent pool is shallow or mismatched for the philosophy of the coaching staff. I tried to do more than answer a question of why the Cowboys season was so disappointing by restating that it was. I tried to explain it, not just talk about it. Anyone can do the latter, and truly anyone can do that wearing a cowboy hat and being a troll about it. However, you are not informed. Perhaps that’s not the point of the Entertainment and Sports Programming Network (ESPN), but it should leave you wanting more than you got.

I go to this example because there is an entire strand of international relations scholarship that has long-worked on this exact same model. It has never seriously tried to explain (or explain well) the pressing questions of international relations. It’s instead been all too content to just talk about it and offer whatever entertainment that could be derived from this farce as having informational value.

Take Kenneth Waltz, for example. Waltz had some real bangers, like this one (re: anarchy) on p. 232 of Man, The State, and War:

Applied to international politics this becomes, in words previously used to summarize Rousseau, the proposition that wars occur because there is nothing prevent them. Rousseau’s analysis explains the recurrence of war without explaining any given war. He tells us that war may at any moment occur, and he tells us why this is so.

That something could happen is not an argument why it does or does not happen.² I would say this is an exceptionally lazy argument, except I’d have to qualify it as an argument.³ A la Wagner (2007) and my training in the Correlates of War tradition, just know that we conceptualize war as “severe” subsets of violent confrontations (i.e. not “cola wars”). We operationalize them by fatalities. You can learn more about this here with respect to inter-state conflicts. War may make the state and the state may make war, but “war” has never required sovereign states in order to happen.⁴ Anarchy, a constant feature of the international system as we may describe it, is clearly not a sufficient condition for war. It’s not necessary for war either. Saying this, as he did, with such conviction belies the nonsense that it is.

Also, take this from Theory of International Politics as the seminal statement of the balance of power thesis (p. 119):

Balance-of-power theory claims to explain a result (the recurrent formation of balances of power), which may or may not accord with the intentions of any of the units whose actions combine to produce that result. To contrive and maintain a balance may be the aim of one or more states, but then again it may not be. According to the theory, balances of power tend to form whether some or all states consciously aim to establish and maintain a balance, or whether some or all states aim for universal domination.

Ignore, for the moment, the omission of agency for states or heads of state in a tradition that goes out of its way to treat humans like they were objects in motion. That’s a disease that is endemic to this tradition, which it states up front as a disability to absolve itself from thinking about anything internal to the state in any detail. But what even is this? What does it mean that something “tends to form”, “which may or may not accord with the intentions of any of the units?” Feel free to read the few pages before it where he assumes a system of independent states and the mechanisms at their disposal to achieve these ends “in more or less sensible ways” (whatever that means). You’re still left with the question of what on earth Waltz is trying to say. What is he trying to say? I’m putting a few words into the mouth of the late Bear Braumoeller (2013, p. 14), but it’s supremely rich that Waltz spends the bulk of his first chapter (second section, especially) complaining that competing theories of international politics to his are so imprecise that hypotheses cannot be stated in ways that would make falsification possible. Waltz is just as imprecise as what he’s criticizing.⁵

I never met Waltz. He retired before I entered the profession and passed away in 2013. I’ve not encountered many people in my field rolling their eyes at him as if he were a troll like Stephen A. Smith. However, accusing others of making imprecise arguments by making an even louder imprecise argument is something I’d imagine Stephen A. Smith doing if he were on an “Author Meets Critics” panel at ISA. I just can’t be as deferential in the absence of information to John Mearsheimer, though. The mechanism is the same, but louder and more bombabstic. Like Stephen A. Smith, Mearsheimer has a shtick. He has a few of them, actually, and he’s the primary inspiration for the image.⁶ Let’s start with this opening salvo from “Why the Ukraine Crisis is the West’s Fault” (which, Jesus Christ, that article title).

Putin’s pushback should have come as no surprise. After all, the West had been moving into Russia’s backyard and threatening its core strategic interests, a point Putin made emphatically and repeatedly. Elites in the United States and Europe have been blindsided by events only because they subscribe to a flawed view of international politics. They tend to believe that the logic of realism holds little relevance in the twenty-first century and that Europe can be kept whole and free on the basis of such liberal principles as the rule of law, economic interdependence, and democracy.

But this grand scheme went awry in Ukraine. The crisis there shows that realpolitik remains relevant—and states that ignore it do so at their own peril. U.S. and European leaders blundered in attempting to turn Ukraine into a Western stronghold on Russia’s border. Now that the consequences have been laid bare, it would be an even greater mistake to continue this misbegotten policy.

“Only the U.S. and the West do dipshit things in the name of domestic politics and ideology while poor hapless Russia is simply responding to ironclad laws of nature.” This is not a serious statement. This is a shtick, like Stephen A. Smith in a cowboy hat. It’s a recurring gag. It’s the IR analogue to Andrew Dice Clay’s “nursery rhymes” bit. I would say it’s funnier than Clay’s “Mother Goose” but Clay’s most grotesque and woefully cringe material isn’t a pretense for an ethnic cleansing campaign. If this isn’t supposed to entertain, was it supposed to seriously inform? I would think the answer is an obvious no. It’s not supposed to explain the underlying logic of what the U.S. and Europe have done. It sure as hell doesn’t explain what Russia has done or is doing. It’s designed to get the kind of engagement and recognition like the kind the Russian Ministry of Foreign Affairs would provide. And they have provided it. On more than one occasion.

Btw, this is pages 2-3 of Tragedy of Great Power Politics.⁷

There are no status quo powers in the international system, save for the occasional hegemon that wants to maintain its dominating position over potential rivals. Great powers are rarely content with the current distribution of power; on the contrary, they face a constant incentive to change it in their favor. They almost always have revisionist intentions, and they will use force to alter the balance of power if they think it can be done at a reasonable price. At times, the costs and risks of trying to shift the balance of power are too great, forcing great powers to wait for more favorable circumstances. But the desire for more power does not go away, unless a state achieves the ultimate goal of hegemony. Since no state is likely to achieve global hegemony, however, the world is condemned to perpetual great-power competition.

This unrelenting pursuit of power means that great powers are inclined to look for opportunities to alter the distribution of world power in their favor. They will seize these opportunities if they have the necessary capa bility. Simply put, great powers are primed for offense. But not only does a great power seek to gain power at the expense of other states, it also tries to thwart rivals bent on gaining power at its expense. Thus, a great power will defend the balance of power when looming change favors another state, and it will try to undermine the balance when the direction of change is in its own favor.

Mearsheimer is complaining that the United States did what we would expect them to do in the 1990s, under the circumstances. States maximize relative power (p. 163, same book), except when they shouldn’t, for ~ reasons ~. Tom Pepinsky notes how frustrating it is that Mearsheimer is able to effortlessly weave between arguments of what states do and what states should do, often in the same paragraph, and in ways that contradict each other. Frustrating? Yes. But this is the dude wearing Machiavelli’s robe as a greeting. If we’re asking for an explanation, we’re not getting it. If we’re asking for sincerity, that’s on us. He’s content to talk about it, though for what I’m sure is a hefty speaker’s fee. If you want some insight into Russia’s invasion of Ukraine, I can point you to Olga Chyzh’s commentaries on it. I can point you to the special issue that came out last year in Conflict Management and Peace Science that explored the fundamental bargaining problems that incentivized Russia for war and make an off-ramp difficult to find. Olga’s brilliant and publicly accessible and your university library already paid for access to this special issue (along with the rest of the journal/publisher’s catalog). Even if you find yourself behind The Guardian’s paywall or aren’t at a university with a SAGE or EBSCOhost subscription, you will learn more about the situation at a fraction of the cost. Please contact Raleigh Addington at Chartwell Speakers if you’d like to pay more to learn less, though.

Again, from Tragedy of Great Power Politics (pp. 30-32), where Mearsheimer says he’s proposing an explanation of the behavior in the international system.

The first assumption is that the international system is anarchic, which does not mean that it is chaotic or riven by disorder. It is easy to draw that conclusion, since realism depicts a world characterized by security competition and war… The second assumption is that great powers inherently possess some offensive military capability, which gives them the wherewithal to hurt and possibly destroy each other… The third assumption is that states can never be certain about other states’ intentions. The fourth assumption is that survival is the primary goal of great powers… The fifth assumption is that great powers are rational actors. They are aware of their external environment and they think strategically about how to survive in it.

As emphasized, none of these assumptions alone dictates that great powers as a general rule should behave aggressively toward each other. There is surely the possibility that some state might have hostile intentions, but the only assumption dealing with a specific motive that is common to all states says that their principal objective is to survive, which by itself is a rather harmless goal. Nevertheless, when the five assumptions are married together, they create powerful incentives for great powers to think and act offensively with regard to each other. In particular, three general patterns of behavior result: fear, self-help, and power maximization.

This is a description of the international system masquerading as an explanation of the international system.⁸ It doesn’t explain the behavior of states in it, but talks about essential assumptions we make about the absence of a global sovereign in a system where the primary units have instruments for harm. Like Waltz’s (1959) appeal to Rousseau, Mearsheimer writes this to imply he’s said something more than he has. It’s a clever writing device to say something with a demonstrative authority to belie the nonsense it is. Talking heads on cable television—like Stephen A. Smith—do this all the time with their delivery of the spoken (if not written) word. The international system has always been anarchic. All states—let alone great powers—possess the capacity for harm toward others (whether it’s horses or nuclear weapons). We can qualify, perhaps, how much weight we want to put on the certainty assumption, but sure, probability is never 0 or 1 in a real world governed by at least a little noise/chance.⁹ States exist to not go the way of the Polish-Lithuanian Commonwealth, which was conspired out of existence in 1795. Got it. Trivial, but got it. States are rational/strategic actors. Sure, Mearsheimer never belabors what he means by that, and likely did not consult Tsebelis (1989) for caveats, but we’ll worry about those details later. We’ve identified five constants that (near direct quote) “when married together, create powerful incentives for great powers to think and act offensively with regard to each other.” I’ve no idea what that is supposed to mean (i.e. this is more of a word salad than you appreciate after a first reading), but we’ve identified behavior that varies across time and space as functions of five things we believe do not vary. To borrow lingo popular in the drag community, Mearsheimer thinks he ate with that passage and left no crumbs. I’d say there were crumbs but the damn thing fell on the ground and we’re almost 25 years past the five-second rule. The flies already got to it.

“Such a distinguished professor at a prestigious university can’t simply be a troll” asks the curious reader who did not enter graduate school in the afterglow of the so-called Perestroika movement in the discipline. The discipline leveraging greater clarity and technological sophistication became quite the existential threat to the engagement troll that suddenly finds itself with a diminished platform. No, the engagement trolls still feature prominently in the outlets more accessible to the general public (whether periodicals like Foreign Affairs or cable television networks like CNN). They’re still getting their speaker fees and are getting enough of them to justify paying for an agent as a business expense. They still found ample time to publish complaints about “rigor” in theoretical argumentation before courting the same community to complain about the quantitative folks. The blue hairs doing this are fine, but their graduate students found it tough-sledding to run back the adviser’s playbook in an academic climate more interested in explanations than conjecture. I’m going to link to Will Moore’s 2015 blog post on this and implore you to read it in its entirety if you haven’t. I came for the stuff I already knew and figured out pretty quickly in graduate school about the Mearsheimer/Walt types. I stayed (as you definitely should!) for the details I did not know about a hiring spat involving UC San Diego from the early 1990s. Get a load of this.

: Totally normal stuff. Not at all ridiculous.

This is the behavior of someone tacitly acknowledging that what he offers was in less demand and does not want to operate on market conditions. It’s the confession of a professor whose PhD products were becoming uncompetitive in a changing job market that became more interested in signals than noise. It’s tough to read those letters from Mearsheimer and Walt and feel anything but an intense level of second-hand embarrassment for these two in sending it. Departments were increasingly swiping left and it bothered these people enough to do this. This was an era before widespread internet use, but this is, spiritually, identical to “whatever you’re ugly anyway.” Rejection fed derision. I don’t know if either sender wore a fedora in front of their typewriter.

It’s not that these people lost a platform. They still had outlets to, again, complain about scientific modeling and their name still carries enough wait to repeat their basic complaints in outlets like European Journal of International Relations or Perspectives on Politics. There are also Foreign Affairs and Foreign Policy too, which are a bit more geared for the general public than academic readers. Even if you humor the notion that they were being “deplatformed” (to use that language), it’s more the case that their energies shifted to arguably bigger (and certainly more lucrative) venues that were more content with content for content’s sake. Cable news or various periodicals/newspapers won’t demand answers the same way International Studies Quarterly or International Organization would. It’s incidentally a bigger stage too. It’s difficult to be sympathetic at all with these complaints under these circumstances. If Stephen A. Smith left ESPN, he’d still have a platform somewhere (Skip Bayless still does!). However, the troll would still complain, much like a troll like Megyn Kelly complains about being “canceled” by NBC on a new platform on SiriusXM. Perhaps the new platform requires even less effort, but the bottom line isn’t adversely affected. God forbid we be “rigorous” though in saying what we mean, meaning what we say, and bringing evidence to support the claims we want to make.

Yes, that was a clumsy pivot to another instance that is not as widely known, but it’s troll-like behavior emerged in a symposium that was required reading or me in graduate school. This was, for lack of a better term, the “rigor” debate in International Security 1999. It was introduced by Stephen Walt and the subsequent symposium later that year is a sight to behold if you know how to read between the lines. To summarize, Walt was complaining that the discipline’s move toward theoretical clarity (i.e. actually being clear in your assumptions and what you’re arguing) was cramping his style. Here is the official complaint, on p. 8:

My argument is straightforward. The central aim of social science is to develop knowledge that is relevant to understanding important social problems. Among other things, this task requires theories that are precise, logically consistent, original, and empirically valid. Formal techniques facilitate the construction of precise and deductively sound arguments, but recent efforts in security studies have generated comparatively few new hypotheses and have for the most part not been tested in a careful and systematic way. The growing technical complexity of recent formal work has not been matched by a corresponding increase in insight, and as a result, recent formal work has relatively little to say about contemporary security issues.

It’s tough for me to take the “allies balance against threat” guy seriously when he follows that with this tasty lick on p. 13.¹⁰ Emphasis added.

These criteria provide a set of hurdles that any social science approach must try to overcome. Although all three are important, the latter two criteria—originality and empirical validity—are especially prized. A consistent, precise yet trivial argument is of less value than a bold new conjecture that helps us understand some important real-world problem, even if certain ambiguities and tensions remain. Similarly, a logically consistent but empirically false theory is of little value, whereas a roughly accurate but somewhat imprecise theory may be extremely useful even though it is still subject to further refinement.

Translation, a la Stephen A. Smith: “I’d rather just talk about how the Cowboys suck than to explain why they do.”

There’s also this one on the folk theorem (pp. 18-19).

Another reason why logical consistency is not enough is the well-known problem of ”multiple equilibria.” Over the past three decades, game theorists have devised ways to build more realistic models by relaxing certain key assumptions (such as the belief that the players have full information). Unfortunately, these more complicated games often contain several equilibrium solutions (i.e., solutions a rational actor would not depart from unilaterally), which means that logical deduction alone cannot tell you which outcome is going to occur. This problem is compounded by the so-called folk theorem, which says that in repeated games with incomplete information and an appropriate discount for the future payoffs, there are always multiple Nash equilibria. Although it is sometimes possible to identify which equilibria will be preferred—Schelling’s famous discussion of ”focal points” was an important effort in this area—”formal mathematical game theory has said little or nothing about where these expectations come from.”

The lay reader may not have raised the kind of quizzical eyebrow I raised reading this, but here’s a way of contextualizing this from my perspective. I tell students “simple is good; simplistic is bad.” All modeling is an exercise in simplification, taking a complex social “object” (e.g. “war”, “state”, “democracy”) and reducing it to essential components that get at the core of the thing in question. It’s why I make allusions to maps or model airplanes as metaphors for what we do and what we want students to think about doing. We can make things more complex and potentially risk generalizability, but I assure you that placing heretofore unmodeled multiple equilibria selection at the core of a process is not that complex. What Walt wants instead is not simplification, but gross simplification. I like the “luddite” metaphor that Niou and Ordeshook (1999) use in response to this. Walt yearns for simpler days of yesteryear where we could be comfortable not thinking about hard stuff or details. It’s just as incurious as an appeal to Rousseau.

I read the full symposium in response to this when I was in graduate school and would encourage you to do as well if you have not already. Bueno de Mesquita and Morrow (1999) largely take Walt (1999) to task for misreading what he cites and for being alarmingly uninterested in 1) the implications of these contributions and 2) making consistent arguments. Martin (1999) repeats much of the same, adds the observation that people like Walt (1999) aren’t being “canceled” (in modern lingo), while also digging into how eager Walt (1999) seems to say “we already knew that” in response to efforts to clarify or move beyond what counted as classic security studies 40 years before the article. Zagare (1999) notes the whole of Walt (1999) is an unstructured mess where nothing follows whatever point Walt wants to communicate.¹¹ Powell (1999), who always did great/challenging work for the -ism field, clarifies how valuable the formal modeling stuff is and offers the pithy complaint that Walt got 44 pages for his screed while every other author in the symposium got ten.

But I want to conclude with Niou and Ordeshook (1999), as I think their “luddite” metaphor is more a four-letter word than seven-letter word when you spell it out. Consider how they close their retort in this (truncated) passage spanning pages 94-96.

The study of politics is, as we argue elsewhere, a field more akin to engineering than to science. Of necessity, our discipline must deal with phenomena that are both too complex for simple, closed-form analysis and too complex for the imprecision of other approaches. This, perhaps, is the attractiveness of imprecision and journalistic discourse-it gives the impression of understanding without revealing the inherent inadequacies of our ideas.

[…]

Finally, we are puzzled most of all by Walt’s assertion that “formal rational choice theorists have been largely absent from the major international security debates of the past decade (such as the nature of the post-Cold War world; the character, causes, and strength of the democratic peace; the potential contribution of security institutions; the causes of ethnic conflict; the future role of nuclear weapons; or the impact of ideas and culture on strategy and conflict)” (p. 46). Even if we were to agree with this statement, we would add that the contributions of Walt’s “other approaches” to this list of security issues escape us as well. But the list is revealing, for it is the product of someone concerned not with science and empirical regularity as those terms need to be understood for the development of cumulative knowledge, but instead with the commentary and informal discussion we find in newspapers and popular journals that has too long appeared under the label “political science.” Such discussion and commentary may be entertaining and even sometimes enlightening, but it remains mere journalism until it can be given the solid scientific grounding that formal theorists pursue.

The references to “newspapers” and “journalism” is thinly veiled if you know how to read between the lines. You know that Stephen A. Smith would still call himself a “journalist” too. There might be a bit of a hubris in the political scientist that dismisses the endeavors of journalism as 100% description and 0% explanation. That may not be 100% warranted, but it feels like it’s at least 99% warranted in the American context when even “description” is a charitable euphemism for what transpires. The Trump-covering ~~stenographer~~“journalist”, the flamboyant ESPN personality, and Stephen Walt aren’t explaining anything. Not a damned thing. It’s even more frustrating when we care deeply about the thing in question. But they are certainly talking about it. Any yahoo with a keyboard or some other type of creative outlet can do that, though.

I know I’m writing this mostly for me, but it’s a point I want to impress upon students (even if my illustration is very niche). The questions we’re asking about the world around us are far more important than the Dallas Cowboys, which makes the demand for explanation infinitely greater and more important than the demand for conjecture. I want explanations. Be wary of those who only offer the ability to talk about it. I’m not saying explanations are easy, but I am saying they are at a greater premium than whatever a booking agent would quote you for a speaker fee. Try to answer the question; don’t beg it.

Notice I said nothing about the merits of Mike McCarthy as play-caller. Cowboy fans seemed to love complaining about Mike McCarthy as a play-caller and how uninspiring his route tree was for his wide receiver corp. In the case above, my eyes typically gravitate toward the caliber of the offensive (and defensive) line. It’s not lost on me that the Cowboys were at their most recent best when their offensive line consisted of all-pros like Travis Frederick, Zack Martin, and Tyron Smith. I’m deliberately downplaying “play-calling” (which is often a generic complaint when specific things aren’t working) and perhaps the caliber of the wide receivers because my perspective places considerable emphasis on play in the trenches. I’ll also confess that the finer points of route trees and coverage shells are beyond my comprehension. Urban Meyer once said “If you want to have a bad team, have a bad defensive line.” I love the sentiment from an offense-oriented coach, but I politely disagree. I think it’s the offensive line. That’s at least my perspective, but I’ll defer to Meyer because he knows more than me. Notice, though, I’m being upfront about my perspective and am acknowledging its limitations. That translates to international relations scholarship. See also the fourth bullet point here. ↩
There is very real Stephen A. Smith energy in that passage. In the second clip in the video above, Smith summarizes some plight of the Dallas Cowboys by saying “What can go wrong, will go wrong. It’s just who the Cowboys are.” Again, maybe it’s entertaining, but it sure as shit isn’t informative and doesn’t explain anything. ↩
For those of us who have worked on territorial conflict, the realist dismissal of knowing any particular war and why it happened (i.e. “without explaining any given war”) is particularly infuriating. Every war might as well be World War 1. “Geopolitics” (a term I’ve always found weird) kind of nudged us to think about this stuff and it predates Kenneth Waltz getting his dissertation at Columbia. Lewis Fry Richardson had already been working on this topic as we might know it now. I think Statistics of Deadly Quarrels was published posthumously, but he had stuff in print precededing Waltz’s statement. The Correlates of War project would be about 15 years away from producing the data and more general statements that might provide better context to what Richardson was doing. Perhaps finding it intellectually incurious is strong language on my part, but it’s hard not to feel that way as a graduate student after reading Diehl (1992). Gate-keeper effects are strong in this profession and are often unstated. Keep this footnote in mind for when I reference the 2015 Will Moore post. ↩
However, it is still imperative to understand that what we know about a war will importantly hinge on whether all, some, or none of the participants are states. Don’t pool types of war together. Please don’t. ↩
This passage is inspired/derived from an email to a student on this exact topic. It’s interesting that Waltz criticizes the ambiguity of “polarity” and its multiple definitions in this book when he’s comparably evasive about what he means by it. This is a long-running complaint of mine that we don’t appreciate Wagner (1993) enough. We have no firm definition of what was “bipolarity” and why it should have mattered at all for the behavior of states living in it. We variously define it as a system of just two states (which is wrong and has never existed), a system of two hostile blocs (which begs the question about the behavior we want to explain), and a system where two states basically guaranteed their own survival from other states (which does not imply a hypothesis of the rivalry that they had and why they cared about each other at all). This will matter a great deal for hypotheses about what “tends to form” irregarding what “may or may not accord with the intentions” of states (again, whatever that means). If we don’t know what it was, why should we believe any behavior followed from it? I don’t think Bruce Bueno de Mesquita published this 1999 ISA conference paper in anything other than his textbooks, but notice that bipolarity (whatever that was) led to peace through certainty. So the story goes. That works if and only if power were equal in a system of two units. Not “balanced”. Equal. The moment we relax that and note it was balanced “enough” is the moment we betray a hypothesis about what “tends to form.” Suddenly, imbalances “tend to form” and now it may have led to the “long peace” through uncertainty. I can’t falsify this fucking hypothesis. Isn’t that a problem, according to Waltz (1979, chp. 1)? It would be if you wanted to explain something. It wouldn’t be if you were content to just talk about it. ↩
If my academic website had a 1990s splash page followed by a Photoshop job of me cosplaying Niccolo Machiavelli, you would not take me seriously. You would deride me for this obviously clownish behavior. Why is he any different? Why does he get a pass for this conceit? Lots of people say they’ve read The Prince, but has anyone who has said that asked themselves why Machiavelli wrote it and what he might have been trying to accomplish? Are you asking me to believe in the sincerity of the messenger under these circumstances? Do you? Come on, guys… ↩
I will never pass up an opportunity to note that Tragedy of Great Power Politics was given a front-cover endorsement by Samuel Huntington, who at the time was in between making an evidently ridiculous argument with a dog whistle and an even more ridiculous argument with a foghorn. “Game recognize game” is a roundabout way of putting it… ↩
I attribute this sentence to an undergraduate class I took with Alexander Wendt at The Ohio State University, which I think was the first undergraduate class he taught after moving there. What follows in this paragraph is mostly inspired by Wagner (2007). You should all read Wagner (2007). ↩
You can conjure the hypothetical MAGA red hat on the way to the voting booth in November being visited by Christmas ghosts, or blinded by the Almighty, to change their intended vote. The probability of such a late reckoning emerging as an accumulation of the inherent contradictions that underpin this devotion the leader is asymptote to 0. But it’s not 0. ↩
Walt (1985) is 40 pages of a closed circle. “Threat” = “perceived intentions.” I perceived an “intention”, and now it’s a threat, and it’s totally a system-level variable and not a domestic factor. Got it. “Power -> threat”, but don’t call it “balance of power”. Call it “balance of threat”. Also got it. I will leave aside the important point that alliances are more than just something like the “all the homies hate Germany” Franco-Russian alliance. Alliances are vehicles to coordinate mutual foreign policy. Russia, for example, routinely uses them as a form of bilateral relationship management and not threat-balancing, per se. Alliances might also be a means to settle pressing foreign policy questions that allies themselves have. These can be the familiar entente (like the one the Brits and French signed after the Fashoda crisis). They could also be the document to outright transfer territory or settle a dispute over it. Alliances are a classically “realist” topic, but the realist treatment of alliances is quite simplistic. ↩
A realist offering an argument with a conclusion that does not logically follow the premises. Hey, that only happens all the time… ↩