Data
====

1. nesYYYY_vote.data.R, where YYYY = 1952, 1956, ..., 1996, 2000
Data from the National Election Study
  - N     : number of observations
  - vote  : 1: George Bush, 0: Bill Clinton
  - income: income percentile
             1: 0-16th
             2: 17-33rd
             3: 34-67th
             4: 68-95th
             5: 96-100th

2. wells.data.R
Decisions of households in Bangladesh about wether to change their source of
drinking water.
  - N        : number of observations
  - arsenic  : the arsenic level of respondent's well in units of hundreds of
               micrograms per liter
  - assoc    : 1 if any members of the household are active in community
               organizations
  - dist     : the distance in meters to the closest known safe well
  - educ     : years of education of the head of household
  - switched : 1 if household i switched to a new well

3. separation.data.R
An example of data for which a logistic regression model is nonidentifiable
  - N : number of observations (60)
  - x ~ N(mean = 1, sd = 2)
  - y : 0 if x < 2, else 1

Models
======

Logistic regressions
1. One predictor
  nes_logit.stan    : glm(vote ~ income, family=binomial(link="logit"))
  separation.stan   : glm(y ~ x, family=binomial(link="logit"))
  wells_dist.stan   : glm(switched ~ dist, family=binomial(link="logit"))
  wells_dist100.stan: glm(switched ~ dist/100, family=binomial(link="logit"))

2. Multiple predictors with no interaction
  wells_d100ars.stan: glm(switched ~ dist/100 + arsenic, family=binomial(link="logit"))
  wells_dae.stan    : glm(switched ~ dist/100 + arsenic + educ/4, 
			  family=binomial(link="logit"))

3. Multiple predictors with interction
  wells_daae_c.stan       : glm(switched ~ c_dist100 + c_arsenic + c_dist100:c_arsenic
                                           + assoc + educ4, 
			        family=binomial(link="logit"))
   where:
   educ4 <- educ / 4
  wells_dae_c.stan        : glm(switched ~ c_dist100 + c_arsenic + c_dist100:c_arsenic 
                                           + educ4, 
			        family=binomial(link="logit"))
  wells_dae_inter.stan    : glm(switched ~ dist/100 + arsenic + educ/4 
                                           + dist/100:arsenic, 
			        family=binomial(link="logit"))
  wells_dae_inter_c.stan  : glm(switched ~ c_dist100 + c_arsenic + c_educ4 
                                           + c_dist100:c_arsenic + c_dist100:c_educ4
                                           + c_arsenic:c_educ4, 
			        family=binomial(link="logit"))
  wells_interaction.stan  : glm(switched ~ dist/100 + arsenic + dist/100:arsenic, 
			        family=binomial(link="logit"))
  wells_interaction_c.stan: glm(switched ~ c_dist100 + c_arsenic 
                                           + c_dist100:c_arsenic, 
				family=binomial(link="logit"))
   where:
   c_dist100 <- (dist - mean(dist)) / 100
   c_arsenic <- arsenic - mean(arsenic)
  wells_predicted.stan    : glm(switched ~ c_dist100 + c_arsenic + c_educ4 
                                           + c_dist100:c_arsenic + c_dist100:c_educ4 
                                           + c_arsenic:c_educ4, 
				family=binomial(link="logit"))
  wells_predicted_log.stan: glm(switched ~ c_dist100 + c_log_arsenic + c_educ4 
                                           + c_dist100:c_log_arsenic 
                                           + c_dist100:c_educ4 
                                           + c_log_arsenic:c_educ4, 
				family=binomial(link="logit"))
    where:
    c_log_arsenic <- log(arsenic) - mean(log(arsenic))