*** Course: FLS6183
*** Class: Causal Inference
*** 02/2018
*** Natalia Moreira

** In this lab, we will use the example developed in chapter 2 of the book "Quantitative Social 
** Science: An Introduction" - K. Imai.

** The first part of this lab follows the step-by-step of chapter 2 in order to
** calculate the difference-in-difference approach.

** The second part of this lab compares the results using the DiD approach and 
** the bivariate regression approach.  

** Unlike the chapter, we will stress how hypothesis tests are undertaken in both approaches. 


*** Part I. Difference-in-Difference

clear

* Define working directory
cd ""

* Import .csv data
import delimited "minwage.csv"

* Save .dta data
save "minwage.dta", replace

* Create dummy variables for the two states 
gen minwageNJ = 0
replace minwageNJ = 1 if location != "PA"

gen minwagePA = 0
replace minwagePA = 1 if location == "PA"

gen NJ = 0
replace NJ = 1 if location != "PA"

gen PA = 0
replace PA = 1 if location == "PA"

* Compare the proportion of restaurants whose wage is less than $ 5.05
tab wagebefore if minwageNJ==1
tab wageafter if minwageNJ==1
tab wagebefore if minwagePA==1
tab wageafter if minwagePA==1

* Compare the proportion of full-time employees in NJ and PA after increase the MW
gen fullpropafterNJ = fullafter / (fullafter+partafter) if minwageNJ == 1
gen fullpropafterPA = fullafter / (fullafter+partafter) if minwagePA == 1
sum fullpropafterNJ fullpropafterPA

* Compute the difference-in-means
gen fullpropafter = fullafter / (fullafter+partafter)

sort minwageNJ
by minwageNJ: sum fullpropafter
ttest fullpropafter, by(minwageNJ) 

* Full-time employment proportion in the previous period for NJ
gen fullpropbefore = fullbefore / (fullbefore+partbefore)

* Naive Test// Compute the difference-in-means NJ
ttest fullpropbefore=fullpropafter if minwageNJ==1

return list // Returning results
				
gen NJbeforemean = r(mu_1)
gen NJaftermean = r(mu_2)

gen NJdiff = NJbeforemean - NJaftermean
tab NJdiff

* Naive Test//  Compute the difference-in-means PA
ttest fullpropbefore=fullpropafter if minwagePA==1

return list
				
gen PAbeforemean = r(mu_1)
gen PAaftermean = r(mu_2)

gen PAdiff = PAbeforemean - PAaftermean
tab PAdiff

* Testing mean difference
ttest NJdiff == PAdiff

***************************************************************************

*** Part II. Regression Model 

* We need to reshape and reload the data to use a regression to analylze the same question.
use "minwage-reshaped.dta", clear

* Naive regression

reg fullprop NJ time

* Interaction

gen did = 0
replace did = time*NJ 

reg fullprop NJ time did

* Let´s check if the results are the same calculating diff-in-diff 

sscinstall diff
diff fullprop, t(NJ) p(time)