* USP Prof. Lorena G Barberia
* 2019 
* Interaction Lab - Answer Key

version 15
clear 
set more off

use "C:\Users\Lorena\Dropbox\2019 USP Methods II\Labs\Lab Class 7 - Interactions\Hillary example\nes1996_modif.dta", clear

* Part 1: X and Z are dummy variables; 	


* As we continously emphasize in this class, you must always know your data.  

*Women's Movement
sum wmnmvt, detail
kdensity wmnmvt
tab wmnmvt

*Income
sum faminc, detail
kdensity faminc
tab faminc

* In the first exercise, it is required to create dummy variable from the varibles already available in the data set.
* Hint: To create variables you can use de "generate" command.  "Replace" will help you while creating a variable from one already 
* present in the data set.

* In this data set, there are many missing values for variables. 
* We need to make sure to keep missing values as missing in creating dummy variables.

*B - Estimate the additive model
* Run the additive model using the dummy variables created above.

*C- Estimate the interative model
* There is more than one way to estimate interactive models. Here, we will describe some of them:

* Creating an interactive variable manually

gen wmn_hiincome_dummy =  wmnmvtdummy*hiincomedummy

* After creating the interactive variable, you may run your regression.

* Using "margins" command
findit margins
* Look for stata documentation of "margins" command before using it.

regress hillary i.wmnmvtdummy##i.incomedummy

* E -  Calculating Predicted Values
* In this exercise, there are a few command that you may use to calculate the predicted values.
* Using Stata's margins command

*Always before runnig margins command, you must run your regression
regress hillary i.wmnmvtdummy##i.incomedummy

* Non-Feminist and Low Income
margins, at( wmnmvtdummy = (0) incomedummy = (0) )

* Feminist and Low Income
margins, at( wmnmvtdummy = (1) incomedummy = (0) )

* Non-Feminist and High Income
margins, at( wmnmvtdummy = (0) incomedummy = (1) )

* Feminist and High Income
margins, at( wmnmvtdummy = (1) incomedummy = (1) )

* A more succint version of the command

margins wmnmvtdummy incomedummy wmnmvtdummy#incomedummy, atmeans

* It is also possible to use "lincom" command

* Non-Feminist and Low Income
lincom 1*_cons

* Feminist and Low income
lincom 1*_cons + 1*wmnmvtdummy

* Non-Feminist and High Income 
lincom 1*_cons + 1* hiincomedummy

* Feminist and High Income 
lincom 1*_cons + 1* wmnmvtdummy +1*hiincomedummy + 1*wmn_hiincome_dummy 

********************************************************************************************
* Part 2: X is a dummy and Z is a categorical variable; 

* A

* With Stata commands treating income as categorical
gen wmnmvtdummy_faminc=wmnmvtdummy*faminc
regress hillary wmnmvtdummy faminc wmnmvtdummy_faminc
regress hillary i.wmnmvtdummy##i.faminc

* B
* Margins plot command allows you to draw a graph to analyze the average predicted effect of income influences 
* how feminist and non-feminists rate Hillary

regress hillary i.wmnmvtdummy##c.faminc
margins, at(faminc=(1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24) wmnmvtdummy=(0 1))
marginsplot, xdimension(faminc)

* Or, equivalently
margins, at(faminc=(1(1)24) wmnmvtdummy= (0 1)) 
marginsplot

* C 

* using command for adjacent values to assess if differences are statistically different from 0. 
regress hillary i.wmnmvtdummy##i.faminc		
margins ar.faminc, dydx(wmnmvtdummy) contrast(nowald)
marginsplot, yline(0)

* With interflex 
ssc install interflex, replace all

interflex hillary wmnmvtdummy faminc, vce(r) ylab(hillary) dlab(wmnmvtdummy) xlab(faminc)


********************************************************************************************
* Part 3: X is a continuous variable and Z is a categorical variable

* A:

* Creating interactive term manually:
gen wmnmvt_faminc=wmnmvt *faminc 
regress hillary wmnmvt faminc wmnmvt_faminc
 
* Using margins command
regress hillary c.wmnmvt##i.faminc
 
 * B
 
 * In Stata margins code 
regress hillary c.wmnmvt##i.faminc
margins, dydx(wmnmvt) at(faminc=(1(7)24) wmnmvt=(0(20)100))  
 
* Simulating a one unit change in wmnmvt 
regress hillary c.wmnmvt##i.faminc
quietly predict double wmnmvt0
quietly replace wmnmvt=wmnmvt+1
quietly predict double wmnmvt1
gen double diff = wmnmvt1-wmnmvt0
sum diff

********************************************************************************************
* Part 4:  X and Z are continuous variables 

* A

* Additive Model 
regress hillary wmnmvt faminc 

* Interaction Model
regress hillary c.wmnmvt##c.faminc 

*B -  Calculating Interaction Effects using Margins

regress hillary c.wmnmvt##c.faminc  
margins, dydx(wmnmvt) at(faminc=(1(1)24)) // dydx is the partial derivative
regress hillary c.wmnmvt##c.faminc  
margins, dydx(faminc) at(wmnmvt=(0(10)100)) // dydx is the partial derivative

// Using margins and marginsplot

regress hillary c.faminc##c.wmnmvt

* (Case 1)
margins, dydx( faminc ) continuous at( wmnmvt = (0(10)100) )

marginsplot, xdimension(at(wmnmvt)) recast(line) recastci(rline) ///
ciopts(lpattern(dash)) ytitle(Marginal Effect of Income) ///
xtitle(Women’s Movement) ///
title(Marginal Effect of Income on Hillary) ///
subtitle((Given variation in Women’s Movement Score))


* Using interflex 

interflex hillary wmnmvt faminc , type(linear) xlab(faminc) ylabel(me of wmnmvt on hillary)

interflex hillary faminc wmnmvt, type(linear) xlab(wmnmvt) ylabel(me of income on hillary)