Causal_Inference_HW3: Difference in Difference
Paulina Cano
April 11, 2019
PART 2 OF ASSIGNMENT
LOADING DATA
Adding state control variables. I extracted some data from University of Kentucky Center for Poverty Research national Welfare Database (UKCPR_National_Data_Final_Update). From the dataset, please first identify three important socioeconomic and/or demographic factors at the state level, and then merge these variables to the main eitcR.dta file.
library(dplyr)## Warning: package 'dplyr' was built under R version 3.5.3##
## Attaching package: 'dplyr'## The following objects are masked from 'package:stats':
##
## filter, lag## The following objects are masked from 'package:base':
##
## intersect, setdiff, setequal, unionlibrary(haven)
library(stargazer)##
## Please cite as:## Hlavac, Marek (2018). stargazer: Well-Formatted Regression and Summary Statistics Tables.## R package version 5.2.2. https://CRAN.R-project.org/package=stargazerdata<-read_dta('C:/Users/PCMcC/Documents/Causal Inference/Homeworks/Assignment 3/eitcR.dta')
data2<-read.csv("C:/Users/PCMcC/Documents/Causal Inference/Homeworks/Assignment 3/UKCPR_National_Welfare_Data_Final_Update_20180116_0.csv")VARIABLE SELECTION FROM UKCPR-NATIONAL WELFARE DATA SET
The following variables are selected from “data2” for our analysis: unemployment rate, minimum wage and povert rate
attach(data2)
vars<-cbind(state, year, state_name, unemployment_rate, poverty_rate, state_min_wage)
data3<-data.frame(vars)MERGE DATASETS AND PRINT OF COLUMN NAMES TO CONFIRM ALL VARIABLES MERGE
I include the statement all.x = TRUE in order to avoid duplicates and maintain my original sample size of N=1887.
merge_data<-merge(data,data3, by = c("state", "year"),all.x = TRUE)
#rm(data2, data3)EVALUAITON OF DATA AND CONFIRM INTEGRITY OF MERGE
Evaluation of the merge shows that the number of observations is consistent for all columns. I filter complete cases in order to eliminate states in which I was unable to get information leaving me with a sample of 5,809 individuals.
library(stargazer)
df <- data.frame(merge_data)
df<-df %>% filter(complete.cases(.))
cols <- c('children', 'nonwhite', "earn", "age", "ed", "work", "unearn", "unemployment_rate", "poverty_rate", "state_min_wage")
stargazer(df[, cols], type = "text", summary.stat = c("N","min", "p25", "median", "p75", "max", "mean", "sd")
)##
## =============================================================================================
## Statistic N Min Pctl(25) Median Pctl(75) Max Mean St. Dev.
## ---------------------------------------------------------------------------------------------
## children 5,809 0 0 1 2 9 1.189 1.404
## nonwhite 5,809 0 0 1 1 1 0.542 0.498
## earn 5,809 0.000 0.000 1,624.494 14,439.450 366,095.500 10,280.960 17,760.500
## age 5,809 20 27 35 44 54 35.514 10.118
## ed 5,809 0 7 10 11 11 9.019 2.462
## work 5,809 0 0 0 1 1 0.469 0.499
## unearn 5,809 0 0.1 4.3 7.4 134 5.472 7.307
## unemployment_rate 5,809 2.900 5.000 6.100 6.900 9.200 6.100 1.356
## poverty_rate 5,809 7.500 10.100 11.800 14.400 25.500 12.405 3.018
## state_min_wage 5,809 1.600 4.250 4.250 4.250 5.250 4.162 0.552
## ---------------------------------------------------------------------------------------------RECODING OF VARIABLES BY NUMBER OF CHILDREN AND ADD THEM TO THE WORKING DATAFRAME
library(dplyr)
library(car)## Loading required package: carData##
## Attaching package: 'car'## The following object is masked from 'package:dplyr':
##
## recodenames(df)## [1] "state" "year" "children"
## [4] "nonwhite" "finc" "earn"
## [7] "age" "ed" "work"
## [10] "unearn" "state_name" "unemployment_rate"
## [13] "poverty_rate" "state_min_wage"#Number of children Variables
df$withchildren<-ifelse(df$children>=1,1,0)
df$onechild<-ifelse(df$children==1,1,0)
df$twochildrenormore<-ifelse(df$children>=2,1,0)
df$nochild<-ifelse(df$children==0,1,0)
df$childrenx <- Recode(df$children, recodes = "1:9='withchildren'; 0='nochilren'; else=NA")
df$childrenx<-as.factor(df$childrenx)
#Employment
df$rec_work<-ifelse(df$work==1 & df$earn >=1, 1,0)
df$rec_nowork<-ifelse(df$work==0 & df$earn<=0,1,0)
#Pre-Treatment and Expansion Variables
df$expansion<-ifelse(df$year>=1993,1,0)
df$pretreat<-ifelse(df$year<=1992,1,0)CALCULATION OF SAMPLES MEANS BY NUMBER OF CHILDREN AS CONDITION INCLUDING ADDITIONAL SUMMARY STATISTICS
ch0 <- subset(df, children==0)
cols <- c('nonwhite', "earn", "age", "ed", "work", "unearn", "unemployment_rate", "poverty_rate", "state_min_wage")
stargazer(ch0[, cols], type = "text", summary.stat = c("N","min", "p25", "median", "p75", "max", "mean", "sd"))##
## =====================================================================================
## Statistic N Min Pctl(25) Median Pctl(75) Max Mean St. Dev.
## -------------------------------------------------------------------------------------
## nonwhite 2,564 0 0 0 1 1 0.445 0.497
## earn 2,564 0 0 7,067.5 20,003.1 176,050 13,925.040 19,645.530
## age 2,564 20 29 41 49 54 39.068 10.937
## ed 2,564 0 7 10 11 11 8.670 2.789
## work 2,564 0 0 1 1 1 0.549 0.498
## unearn 2,564 0.000 0.000 2.222 6.874 134.058 5.131 8.629
## unemployment_rate 2,564 2.900 5.000 6.100 7.000 9.200 6.098 1.395
## poverty_rate 2,564 7.500 10.200 11.800 14.500 25.500 12.536 3.071
## state_min_wage 2,564 1.600 4.250 4.250 4.250 5.250 4.135 0.567
## -------------------------------------------------------------------------------------ch1<- subset(df, children==1)
cols <- c('nonwhite', "earn", "age", "ed", "work", "unearn", "unemployment_rate", "poverty_rate", "state_min_wage")
stargazer(ch1[, cols], type = "text", summary.stat = c("N","min", "p25", "median", "p75", "max", "mean", "sd"))##
## =====================================================================================
## Statistic N Min Pctl(25) Median Pctl(75) Max Mean St. Dev.
## -------------------------------------------------------------------------------------
## nonwhite 1,249 0 0 1 1 1 0.525 0.500
## earn 1,249 0 0 2,159.0 14,176.4 366,095 10,300.080 19,575.000
## age 1,249 20 24 34 42 54 33.862 9.862
## ed 1,249 0 7 10 11 11 9.219 2.176
## work 1,249 0 0 0 1 1 0.486 0.500
## unearn 1,249 0.000 0.295 3.684 6.383 92.926 4.812 6.006
## unemployment_rate 1,249 2.900 5.000 6.100 6.900 9.200 6.100 1.332
## poverty_rate 1,249 7.500 10.100 11.800 14.400 25.500 12.429 3.048
## state_min_wage 1,249 1.600 4.250 4.250 4.250 5.250 4.131 0.593
## -------------------------------------------------------------------------------------ch2<- subset(df, children>=2)
cols <- c('nonwhite', "earn", "age", "ed", "work", "unearn", "unemployment_rate", "poverty_rate", "state_min_wage")
stargazer(ch2[, cols], type = "text", summary.stat = c("N","min", "p25", "median", "p75", "max", "mean", "sd"))##
## ========================================================================================
## Statistic N Min Pctl(25) Median Pctl(75) Max Mean St. Dev.
## ----------------------------------------------------------------------------------------
## nonwhite 1,996 0 0 1 1 1 0.675 0.468
## earn 1,996 0.000 0.000 0.000 6,572.419 118,847.800 5,587.935 11,986.770
## age 1,996 20 26 32 37 54 31.983 7.288
## ed 1,996 0 9 10 11 11 9.342 2.100
## work 1,996 0 0 0 1 1 0.357 0.479
## unearn 1,996 0 2.7 5.6 8.2 81 6.324 6.006
## unemployment_rate 1,996 2.900 5.000 6.100 6.900 9.200 6.103 1.320
## poverty_rate 1,996 7.500 10.100 11.200 14.400 25.500 12.223 2.923
## state_min_wage 1,996 1.600 4.250 4.250 4.250 5.250 4.216 0.499
## ----------------------------------------------------------------------------------------CALCULATION OF SAMPLES MEANS BY EMPLOYMENT STATUS AS CONDITION INCLUDING ADDITIONAL SUMMARY STATISTICS
#Women employed
em1 <- subset(df, work==1 & earn>=1)
cols <- c('nonwhite', "earn", "age", "ed", "work", "unearn", "unemployment_rate", "poverty_rate", "state_min_wage")
stargazer(em1[, cols], type = "text", summary.stat = c("N","min", "p25", "median", "p75", "max", "mean", "sd"))##
## ===============================================================================================
## Statistic N Min Pctl(25) Median Pctl(75) Max Mean St. Dev.
## -----------------------------------------------------------------------------------------------
## nonwhite 2,726 0 0 0 1 1 0.451 0.498
## earn 2,726 1.023 5,719.886 12,995.950 23,325.260 176,050.200 17,946.030 18,564.300
## age 2,726 20 27 36 45 54 36.050 10.203
## ed 2,726 0 7.5 10 11 11 9.241 2.282
## work 2,726 1 1 1 1 1 1.000 0.000
## unearn 2,726 0 0 0.6 4.9 108 3.469 6.371
## unemployment_rate 2,726 2.900 5.000 6.100 6.900 9.200 6.034 1.409
## poverty_rate 2,726 7.500 10.200 11.800 14.500 25.500 12.537 3.141
## state_min_wage 2,726 1.600 4.250 4.250 4.250 5.250 4.144 0.586
## -----------------------------------------------------------------------------------------------#women Unemployed
em0<- subset(df, work==0 & earn==0)
cols <- c('nonwhite', "earn", "age", "ed", "work", "unearn", "unemployment_rate", "poverty_rate", "state_min_wage")
stargazer(em0[, cols], type = "text", summary.stat = c("N","min", "p25", "median", "p75", "max", "mean", "sd"))##
## ==============================================================================
## Statistic N Min Pctl(25) Median Pctl(75) Max Mean St. Dev.
## ------------------------------------------------------------------------------
## nonwhite 2,547 0 0 1 1 1 0.631 0.483
## earn 2,547 0 0 0 0 0 0.000 0.000
## age 2,547 20 26 33 43 54 34.672 9.849
## ed 2,547 0 7 10 11 11 8.902 2.521
## work 2,547 0 0 0 0 0 0.000 0.000
## unearn 2,547 0.000 3.672 5.967 8.485 134.058 6.914 6.756
## unemployment_rate 2,547 2.900 5.000 6.100 6.900 9.200 6.154 1.301
## poverty_rate 2,547 7.500 10.100 11.800 14.200 25.500 12.298 2.858
## state_min_wage 2,547 1.600 4.250 4.250 4.250 5.250 4.172 0.519
## ------------------------------------------------------------------------------Annual employment rates by year (1991-1996) for single women with children (treatment) and without children (control).
plot<-aggregate(df$work, by=list(df$year, df$childrenx), FUN=sum)
#install.packages("ggplot2")
library(ggplot2)## Warning: package 'ggplot2' was built under R version 3.5.3ggplot(plot, aes(fill=Group.2, y=x/5809*100 , x=Group.1)) +
geom_bar(position="dodge", stat="identity") + scale_fill_manual(values=c( "#E69F00", "#56B4E9"))+ theme_minimal() + scale_y_continuous(limits=c(0,5)) + scale_x_continuous(breaks=seq(1991, 1996, 1)) +
ggtitle("Annual Employment Rates by Year (1991 to 1996) for Single Women by Treatment") +
theme(plot.title = element_text(face="bold", size=10, hjust=0)) + theme(legend.title=element_blank()) + ylab("Percentage Rate")+xlab("Year") + geom_text(aes(label=round(x/5809*100,2) ),position=position_dodge(0.9))
## Unconditional difference-in-difference estimates
of the effect of the 1993 EITC expansion on employment of single women. Your table should be parallel to the format found in a typical differences-in-differences paper. Calculate estimates with all women with children as the treatment (single women with no children as the control), women with children as the control. Discuss these results.
CREATE AN INTERACTION AND ESTIMATE THE DID ESTIMATOR
#Create the interaction
did<-df$withchildren * df$expansion
#DID Estimator
didreg = lm(work ~ nochild + expansion + did, data = df)
summary(didreg)##
## Call:
## lm(formula = work ~ nochild + expansion + did, data = df)
##
## Residuals:
## Min 1Q Median 3Q Max
## -0.5672 -0.4117 -0.4032 0.4624 0.5968
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.41167 0.01397 29.469 < 2e-16 ***
## nochild 0.15548 0.02115 7.351 2.24e-13 ***
## expansion -0.02955 0.02013 -1.468 0.142
## did 0.02109 0.02688 0.785 0.433
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.4941 on 5805 degrees of freedom
## Multiple R-squared: 0.02044, Adjusted R-squared: 0.01994
## F-statistic: 40.38 on 3 and 5805 DF, p-value: < 2.2e-16Regression with conditional difference-in-difference estimates including individual-level controls.
reg1 <- lm(work ~ withchildren + did + scale(age) + nonwhite + ed, data = df)
summary(reg1)##
## Call:
## lm(formula = work ~ withchildren + did + scale(age) + nonwhite +
## ed, data = df)
##
## Residuals:
## Min 1Q Median 3Q Max
## -0.6872 -0.4519 -0.2966 0.4858 0.8614
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.429474 0.026100 16.455 < 2e-16 ***
## withchildren -0.120437 0.017453 -6.901 5.73e-12 ***
## did 0.001549 0.017551 0.088 0.9297
## scale(age) 0.016070 0.006803 2.362 0.0182 *
## nonwhite -0.148992 0.013040 -11.426 < 2e-16 ***
## ed 0.020761 0.002643 7.854 4.76e-15 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.486 on 5803 degrees of freedom
## Multiple R-squared: 0.05259, Adjusted R-squared: 0.05178
## F-statistic: 64.43 on 5 and 5803 DF, p-value: < 2.2e-16Regression with conxtetual factors (includes interaction term “withchildren”)
reg2 <- lm(work ~ (did + scale(age) + nonwhite + ed +unemployment_rate + poverty_rate + state_min_wage)*withchildren, data = df)
summary(reg2)##
## Call:
## lm(formula = work ~ (did + scale(age) + nonwhite + ed + unemployment_rate +
## poverty_rate + state_min_wage) * withchildren, data = df)
##
## Residuals:
## Min 1Q Median 3Q Max
## -0.7146 -0.4393 -0.2748 0.4771 0.8404
##
## Coefficients: (1 not defined because of singularities)
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.4859358 0.1021325 4.758 2.00e-06 ***
## did -0.0403472 0.0201922 -1.998 0.045746 *
## scale(age) -0.0059237 0.0089029 -0.665 0.505840
## nonwhite -0.1377841 0.0195321 -7.054 1.94e-12 ***
## ed 0.0261223 0.0034443 7.584 3.88e-14 ***
## unemployment_rate -0.0050752 0.0071542 -0.709 0.478100
## poverty_rate 0.0007088 0.0033598 0.211 0.832923
## state_min_wage -0.0189002 0.0174667 -1.082 0.279265
## withchildren 0.0260217 0.1428319 0.182 0.855445
## did:withchildren NA NA NA NA
## scale(age):withchildren 0.0516190 0.0138346 3.731 0.000192 ***
## nonwhite:withchildren -0.0112088 0.0265281 -0.423 0.672658
## ed:withchildren -0.0106774 0.0053962 -1.979 0.047897 *
## unemployment_rate:withchildren -0.0314584 0.0103581 -3.037 0.002400 **
## poverty_rate:withchildren 0.0064888 0.0045983 1.411 0.158259
## state_min_wage:withchildren 0.0224460 0.0243120 0.923 0.355917
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.4844 on 5794 degrees of freedom
## Multiple R-squared: 0.0603, Adjusted R-squared: 0.05803
## F-statistic: 26.56 on 14 and 5794 DF, p-value: < 2.2e-16PLACEBO TREATMENT
#DID PLACEBO Estimator
#Create the interaction
did2<-df$withchildren * df$pretreat
#DID Estimator
didreg2 = lm(work ~ withchildren + pretreat + did2, data = df)
summary(didreg2)##
## Call:
## lm(formula = work ~ withchildren + pretreat + did2, data = df)
##
## Residuals:
## Min 1Q Median 3Q Max
## -0.5672 -0.4117 -0.4032 0.4624 0.5968
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.53759 0.01237 43.467 < 2e-16 ***
## withchildren -0.13438 0.01660 -8.098 6.76e-16 ***
## pretreat 0.02955 0.02013 1.468 0.142
## did2 -0.02109 0.02688 -0.785 0.433
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.4941 on 5805 degrees of freedom
## Multiple R-squared: 0.02044, Adjusted R-squared: 0.01994
## F-statistic: 40.38 on 3 and 5805 DF, p-value: < 2.2e-16reg3 <- lm(work ~ did2 +scale (age) + nonwhite + ed, data = df)
summary(reg3)##
## Call:
## lm(formula = work ~ did2 + scale(age) + nonwhite + ed, data = df)
##
## Residuals:
## Min 1Q Median 3Q Max
## -0.6668 -0.4394 -0.3063 0.4858 0.8706
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.399261 0.025935 15.395 < 2e-16 ***
## did2 -0.061853 0.015796 -3.916 9.12e-05 ***
## scale(age) 0.030115 0.006589 4.571 4.96e-06 ***
## nonwhite -0.167803 0.012875 -13.033 < 2e-16 ***
## ed 0.019316 0.002650 7.289 3.53e-13 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.4885 on 5804 degrees of freedom
## Multiple R-squared: 0.04286, Adjusted R-squared: 0.0422
## F-statistic: 64.98 on 4 and 5804 DF, p-value: < 2.2e-16reg4 <- lm(work ~ (did2 + scale(age) + nonwhite + ed +unemployment_rate + poverty_rate + state_min_wage)*withchildren, data = df)
summary(reg4)##
## Call:
## lm(formula = work ~ (did2 + scale(age) + nonwhite + ed + unemployment_rate +
## poverty_rate + state_min_wage) * withchildren, data = df)
##
## Residuals:
## Min 1Q Median 3Q Max
## -0.7146 -0.4393 -0.2748 0.4771 0.8404
##
## Coefficients: (1 not defined because of singularities)
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.4859358 0.1021325 4.758 2.00e-06 ***
## did2 0.0403472 0.0201922 1.998 0.045746 *
## scale(age) -0.0059237 0.0089029 -0.665 0.505840
## nonwhite -0.1377841 0.0195321 -7.054 1.94e-12 ***
## ed 0.0261223 0.0034443 7.584 3.88e-14 ***
## unemployment_rate -0.0050752 0.0071542 -0.709 0.478100
## poverty_rate 0.0007088 0.0033598 0.211 0.832923
## state_min_wage -0.0189002 0.0174667 -1.082 0.279265
## withchildren -0.0143255 0.1425014 -0.101 0.919928
## did2:withchildren NA NA NA NA
## scale(age):withchildren 0.0516190 0.0138346 3.731 0.000192 ***
## nonwhite:withchildren -0.0112088 0.0265281 -0.423 0.672658
## ed:withchildren -0.0106774 0.0053962 -1.979 0.047897 *
## unemployment_rate:withchildren -0.0314584 0.0103581 -3.037 0.002400 **
## poverty_rate:withchildren 0.0064888 0.0045983 1.411 0.158259
## state_min_wage:withchildren 0.0224460 0.0243120 0.923 0.355917
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.4844 on 5794 degrees of freedom
## Multiple R-squared: 0.0603, Adjusted R-squared: 0.05803
## F-statistic: 26.56 on 14 and 5794 DF, p-value: < 2.2e-16