PyFixest 0.10.7: Basic Did2s support by s3alfisc · Pull Request #194 · py-econometrics/pyfixest

s3alfisc · 2023-10-22T09:35:01Z

Adds support for experimental event_study() function, which supports Difference-in-Differences estimation via two-way fixed effects and Gardner's two-stage estimator (2021).
To support the implementation of Gardner's estimator, now allows to estimate 'shallow' models as feols("Y ~ 0 | f1).

To Do's:

allow for covariates
allow for custom clustering variables
implement proper event study with leads and lags. Pyfixest's i() operator currently only supports one reference level, do I have to adjust it to add a second level? Check.
Write unit tests.

Example:

%load_ext autoreload
%autoreload 2
from pyfixest.experimental.did import event_study
from pyfixest.summarize import summary
import pandas as pd

df_het = pd.read_csv("pyfixest/data/df_het.csv")

fit_twfe = event_study(
    data = df_het,
    yname = "dep_var",
    idname= "state",
    tname = "year",
    gname = "g",
    estimator = "twfe"
)

fit_did2s = event_study(
    data = df_het,
    yname = "dep_var",
    idname= "state",
    tname = "year",
    gname = "g",
    estimator = "did2s"
)

summary([fit_twfe, fit_did2s])

# ###
#
# Estimation:  TWFE
# Dep. var.: dep_var, Fixed effects: state+year
# Inference:  CRV1
# Observations:  46500
#
# | Coefficient   |   Estimate |   Std. Error |   t value |   Pr(>|t|) |   2.5 % |   97.5 % |
# |:--------------|-----------:|-------------:|----------:|-----------:|--------:|---------:|
# | zz00_treat    |      2.135 |        0.044 |    48.803 |      0.000 |   2.049 |    2.220 |
# ---
# ###
#
#   Estimation:  DID2S
# Dep. var.: dep_var
# Inference:  CRV1 (GMM)
# Observations:  46500
#
# | Coefficient   |   Estimate |   Std. Error |   t value |   Pr(>|t|) |   2.5 % |   97.5 % |
# |:--------------|-----------:|-------------:|----------:|-----------:|--------:|---------:|
# | zz00_treat    |      2.152 |        0.048 |    45.208 |      0.000 |   2.059 |    2.246 |
# ---

s3alfisc · 2023-10-22T09:40:40Z

@kylebutts on one hand, mostly fyi =) But also, some of the code is more or less directly translated from your R package, which I mention in the codebase. I think it would be best practice to actually add your name to the list of package authors (this is what I do with in my R packages if I more or less copy code from another package). What do you think?

kylebutts · 2023-10-23T17:14:14Z

I'm not sure how multiple inheritance works in python, but is there a possibility to pass first_stage and second_stage directly? For example, I'm not sure how I could specify subgroup-specific time-trends in that notation. The nice thing about did2s (IMO) is the flexibility in modeling $Y_{it}(0)$.

Though, I do see how it's nice to have a single event_study option. Could you have a did2s command that event_study calls with the standard TWFE formula? Happy to try and help, but my python skills are not very strong.

s3alfisc · 2023-10-23T21:27:04Z

Hi Kyle, Thanks for offering your help! I might have some questions on reference levels for the event study with leads and lags down the line =)

I thought about having a custom did2s function yesterday and basically came to the same conclusion as you: I should definitely support it! It indeed offers a loot of versatility that event_study() cannot provide (as you mention, subgroup specific time trends, "pooling" of time intervals in event studies, triple DIDs and probably quite a few other things I am not aware of). It should be fairly straightforward to set up - basically I need to copy the event_study() interface and call it did2s and add formula arguments to it + DID2S.estimate(). And that's it I think =)

s3alfisc · 2023-10-23T21:33:56Z

Ah, maybe not as easy as I thought: mapping from an event study interface to calling did2s is fairly straightforward, but not the other way around. But should still be manageable =)

kylebutts · 2023-10-24T03:07:57Z

What do you think is hard about making a did2s command? Your code looks pretty generic. From what I can tell, the main change you would need to make is in the vcov part:
https://github.com/s3alfisc/pyfixest/blob/f03deb3b8015efbdf7cfac9cda14f5c10db37830/pyfixest/experimental/did.py#L287-L291

s3alfisc · 2023-10-24T07:19:11Z

Yes, you're right, it's just that the closer time gets to midnight, the faster I type and the worse I think 😅

basically, the steps are the following:

create a did2s function
Initiate the DID2S object
Overwrite the formula attributes with the user selected formulas
update estimate() and vcov()

Really not too hard :)

codecov · 2023-10-30T19:49:15Z

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (0c3b162) 0.00% compared to head (7398f5e) 0.00%.

❗ Current head 7398f5e differs from pull request most recent head 04e4140. Consider uploading reports for the commit 04e4140 to get more accurate results

Additional details and impacted files

@@      Coverage Diff      @@
##   master   #194   +/-   ##
=============================
=============================

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

kylebutts · 2023-10-30T20:48:15Z

lol I'm getting push notifications of each attempt @s3alfisc 😅 godspeed on this

s3alfisc · 2023-10-30T20:57:17Z

Oh no 🙈 I had my fingers crossed GitHub would stop notifying you after the 10th attempt or so 🙃 sorry about that! I'm definitely done for today 😅

* PyFixest 0.10.7: Basic Did2s support (#194) * initial commit * reproduce did2s.r results * add twfe estimator * create treatment var internally * some defensive input checks * acknowledge kyle as copyright holder * create experimental folder * update docs * update version * add did2s function * get event_study() did2s to work again * delete did2s function * add basic tests * patch version * delete pyaches * update news * update starting message * update readme * update pyproject.toml * ci tests on mac as fixest already compiled * test on windows * back to testing on linux * first install python, then R deps * fix yaml syntax * yaml is too hard * fix typo * PyFixest 0.10.8: More flexible DID2s support (#206) * initial commit * reproduce did2s.r results * add twfe estimator * create treatment var internally * some defensive input checks * acknowledge kyle as copyright holder * create experimental folder * update docs * update version * add did2s function * get event_study() did2s to work again * delete did2s function * add basic tests * patch version * delete pyaches * update news * update starting message * update readme * update pyproject.toml * ci tests on mac as fixest already compiled * test on windows * back to testing on linux * first install python, then R deps * fix yaml syntax * yaml is too hard * some work towards a more flexible API * get did2s function prototype to run * event_study() and did2s() both run through * add examples to did2s() and event_study() * more tests for did2s * pass tests * two reference levels * fix small bug re _drop_ref * get did2s() to run * fix did2s vcov bug * add tests * update did example * further updates * update i-tests * bump version and update readme * readme tweaks * update news * update readme * update readme * dont gitignore figures folder * PyFixest 0.8.10.1: add black workflow (#207) * add black action * run black * revert notebook * PyFixest 0.10.8.2: close #208 (#209) * check for did2s if treatment is boolean * update docs * ci only when push to master * bump version * small fixes * formatting * PyFixest 0.10.9: Improve etable() (#210) * first step to prettier etable * add tabulate dependency, fix etable fixef bug * fix bug with no fixed effects * fix no model has fixef exception * bump poetry version * fix hline placement bug * formatting * make plot nicer * add vcov info * reformat * reformat * update news * PyFixest 0.10.10 (#212) * attempt to fix #211 * cleanup * clarification * fix i() bug for i(var), not for i(var1, var2) * add drop_intercept argument to feols, fepois * fix error with did2s inference * add error when 0 or -1 in first stage * fix tests, output warning for i(var1, var2) syntax * format code * update figure + fix small bug * no 0 in second stage did test * bump version * deprecate i(var1, var2) syntax, update news * format * bring back i(var1, var2) * format * fix i(var1, var2) interaction - unit tests pass * formatting * bump version * Fix poisson bug (#215) * update docs * add _depvar attribute to Feols object * use _depvar attribute * move separation checks out of Fepois to fix #138 * revert changes to performance method * temporarily do not report r squared metrics * fix separation bug #138 * formatting * update docs * njit and parallelize crv1 inference #204 * formatting * bump version

* initial commit * reproduce did2s.r results * add twfe estimator * create treatment var internally * some defensive input checks * acknowledge kyle as copyright holder * create experimental folder * update docs * update version * add did2s function * get event_study() did2s to work again * delete did2s function * add basic tests * patch version * delete pyaches * update news * update starting message * update readme * update pyproject.toml * ci tests on mac as fixest already compiled * test on windows * back to testing on linux * first install python, then R deps * fix yaml syntax * yaml is too hard

s3alfisc added 9 commits October 21, 2023 18:50

initial commit

abd31d1

reproduce did2s.r results

fa7dfc3

add twfe estimator

99ad650

create treatment var internally

acf761a

some defensive input checks

258b381

acknowledge kyle as copyright holder

1018f78

create experimental folder

13c49d9

update docs

9429cf3

update version

f03deb3

s3alfisc mentioned this pull request Oct 22, 2023

Difference-in-Differences Support #78

Closed

5 tasks

s3alfisc linked an issue Oct 22, 2023 that may be closed by this pull request

Did2s (Gardner) #195

Closed

s3alfisc added 5 commits October 28, 2023 16:01

add did2s function

250710c

get event_study() did2s to work again

30e2f90

delete did2s function

65f3e23

add basic tests

7e83905

patch version

1a999e7

s3alfisc changed the title ~~PyFixest 0.10.5~~ PyFixest 0.10.7: Basic Did2s support Oct 29, 2023

s3alfisc added 7 commits October 29, 2023 20:42

solve merge conflicts

77fd0be

delete pyaches

74e2162

update news

b3a93cc

update starting message

09da8e1

update readme

7031577

update pyproject.toml

7398f5e

ci tests on mac as fixest already compiled

3e5d134

s3alfisc added 5 commits October 30, 2023 21:04

test on windows

edd9640

back to testing on linux

79ad27c

first install python, then R deps

1ec2f7d

fix yaml syntax

284dcaa

yaml is too hard

04e4140

s3alfisc merged commit dccce10 into master Nov 1, 2023

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

PyFixest 0.10.7: Basic Did2s support#194

PyFixest 0.10.7: Basic Did2s support#194
s3alfisc merged 26 commits intomasterfrom
did

s3alfisc commented Oct 22, 2023 •

edited

Loading

Uh oh!

s3alfisc commented Oct 22, 2023

Uh oh!

kylebutts commented Oct 23, 2023

Uh oh!

s3alfisc commented Oct 23, 2023

Uh oh!

s3alfisc commented Oct 23, 2023

Uh oh!

kylebutts commented Oct 24, 2023

Uh oh!

s3alfisc commented Oct 24, 2023

Uh oh!

codecov bot commented Oct 30, 2023 •

edited

Loading

Uh oh!

kylebutts commented Oct 30, 2023

Uh oh!

s3alfisc commented Oct 30, 2023

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

s3alfisc commented Oct 22, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

s3alfisc commented Oct 22, 2023

Uh oh!

kylebutts commented Oct 23, 2023

Uh oh!

s3alfisc commented Oct 23, 2023

Uh oh!

s3alfisc commented Oct 23, 2023

Uh oh!

kylebutts commented Oct 24, 2023

Uh oh!

s3alfisc commented Oct 24, 2023

Uh oh!

codecov bot commented Oct 30, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

kylebutts commented Oct 30, 2023

Uh oh!

s3alfisc commented Oct 30, 2023

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

s3alfisc commented Oct 22, 2023 •

edited

Loading

codecov bot commented Oct 30, 2023 •

edited

Loading