## 5. ESTIMATING CAUSAL EFFECTS WITH OBSERVATIONAL DATA

## 5.3 THE EFFECT OF RUSSIAN TV ON UKRAINIANS' VOTING BEHAVIOR

## Set the working directory
setwd("~/Desktop/DSS") # example of setwd() for Mac 
setwd("C:/user/Desktop/DSS") # example for Windows

## Load the dataset
uas <- read.csv("UA_survey.csv") # reads and stores data

## Understand the data
## (Read about description of variables and unit of observation)
head(uas) # shows first observations

## Identify the types of variables included
## (character vs. numeric; binary vs. non-binary)

## Identify the number of observations
dim(uas) # provides dimensions of dataframe: rows, columns

### 5.3.1 USING THE SIMPLE LINEAR MODEL TO COMPUTE 
### THE DIFFERENCE-IN-MEANS ESTIMATOR

## Option A: Compute the difference-in-means estimator directly
mean(uas$pro_russian_vote[uas$russian_tv==1])-
  mean(uas$pro_russian_vote[uas$russian_tv==0])

## Option B: Fit linear model
lm(pro_russian_vote ~ russian_tv, data=uas) 

### 5.3.2 CONTROLLING FOR CONFOUNDERS 
### USING A MULTIPLE LINEAR REGRESSION MODEL

## Compute correlation
cor(uas$within_25km, uas$russian_tv)

## Create two-way table of frequencies
table(uas$within_25km, uas$russian_tv)

## Fit linear model
lm(pro_russian_vote ~ russian_tv + within_25km, data=uas) 

## 5.4 THE EFFECT OF RUSSIAN TV ON UKRAINIAN ELECTORAL OUTCOMES


## Load the dataset
uap <- read.csv("UA_precincts.csv") # reads and stores data

## Understand the data
## (Read about description of variables and unit of observation)
head(uap) # shows first observations

## Identify the types of variables included
## (character vs. numeric; binary vs. non-binary)

## Identify the number of observations
dim(uap) # provides dimensions of dataframe: rows, columns

### 5.4.1 USING THE SIMPLE LINEAR MODEL TO COMPUTE 
### THE DIFFERENCE-IN-MEANS ESTIMATOR

## Create pro-Russian change variable
uap$pro_russian_change <- uap$pro_russian - uap$prior_pro_russian

## Create histogram
hist(uap$pro_russian_change)
mean(uap$pro_russian_change)
median(uap$pro_russian_change)

## Fit linear model
lm(pro_russian_change ~ russian_tv, data=uap)
mod_ukr0 <- lm(pro_russian_change ~ russian_tv, data=uap)

### 5.4.2 CONTROLLING FOR CONFOUNDERS 
### USING A MULTIPLE LINEAR REGRESSION MODEL

## Compute correlation
cor(uap$within_25km, uap$russian_tv)

## Fit linear model
lm(pro_russian_change ~ russian_tv + within_25km, data=uap) 

mod_ukr <- lm(pro_russian_change ~ russian_tv + within_25km, data=uap) 

summary(mod_ukr)

library(modelsummary)
modelsummary(mod_ukr, stars=T)
modelplot(mod_ukr)

mods = list("base"=mod_ukr0, 
            "control"=mod_ukr)

modelsummary(mods, stars=T)
modelplot(mods)

library(marginaleffects)
plot_predictions(mod_ukr0, by="russian_tv")
plot_predictions(mod_ukr, by=c("russian_tv","within_25km"))
plot_predictions(mod_ukr, condition="russian_tv")


library(performance)
check_model(mod_ukr)
check_outliers(mod_ukr)
check_heteroscedasticity(mod_ukr)

compare_performance(mod_ukr0, mod_ukr)

library(report)
report(mod_ukr)