Sean_Posey_HW4

Readings

Krueger 1999

1.) What is the causal link the paper is trying to reveal

1. Effect of class size from kindergarten to third grade on students’ performance on standardized tests and does adding an aide to large classrooms reduce the negative impacts of large classrooms.

2.) What would the ideal experiment to test this causal link

1. The ideal experiment is to have the same individual attend different size classes and obtain their performance. This is impossible because 1 individual cannot be both treated and untreated therefore, we do not get perfect counterfactuals. Realistically the ideal experiment would be a randomized control trial with a blocked randomization design where students are first grouped by a variable that affects outcome such as past performance, IQ, etc. and then individuals in each group are randomly split between treatment and control.

3.) What is the identification strategy

1. The authors conducted a randomized control trial where individuals in select schools in Tennessee were randomly assigned to classes with different size classes, small (13-17 students), regular (22-25 students), and regular with aide (22-25 students). The authors did not have baseline data on student’s performance so were unable to balance the sample on the outcome variable.

4.) What are the assumption/threats to this identification strategy

1. Differential attrition, students enter in and out of treatment. Students who may enter into a small class may have previously been in a large classroom at a different school and vice versa, this creates a bias in an unknown direction.

Ashenfelter and Krueger 1994

1.) What is the causal link the paper is trying to reveal

1. Educations impact on wages

2.) What would the ideal experiment to test this causal link

1. An ideal experiment is to have a single individual both go to school and not go to school for varying years so that innate ability and other non-observable characteristics such as social skills do not impact wage.

3.) What is the identification strategy

1. For this paper they used twins to control for the non-observables since twins should be a strong counterfactual.

4.) What are the assumption/threats to this identification strategy

1. Measurement error. They claim that they find no evidence that unobserved ability is positively related to the schooling level completed, however this seems unlikely.

2. Another potential issue is sampling. In the sample, the authors stated, more women and whites than in CPS sample. This may result in a non-representative sampling. Furthermore, identical twins are not a perfect counterfactual leading to unknown biases.

Replication

twins<-read.csv('./AshenfelterKrueger1994_twins.csv')
#twins<-as.data.table(twins)
colnames(twins)<-c('famid','age', 'educ.T1', 'educ.T2', 'lwage.T1','lwage.T2', 'male.T1','male.T2','white.T1','white.T2')
long<-reshape(twins, direction='long',
              varying=c('educ.T1', 'lwage.T1', 'male.T1','white.T1', 'educ.T2','lwage.T2','male.T2','white.T2'),
              timevar='twin',
              times=c('T1','T2'),
              v.names=c('educ','lwage','male','white'),
              idvar=c('famid','age'))
long<-long[order(long$famid),]

twins$D.Ed=twins$educ.T1-twins$educ.T2
twins$D.lw=twins$lwage.T1-twins$lwage.T2
FD2<-lm(D.lw~D.Ed,data=twins)
stargazer(FD2,type='html')

	Dependent variable:
	D.lw
D.Ed	0.092***
	(0.024)
Constant	-0.079*
	(0.045)
Observations	149
R2	0.092
Adjusted R2	0.086
Residual Std. Error	0.554 (df = 147)
F Statistic	14.914*** (df = 1; 147)
Note:	p<0.1; p<0.05; p<0.01

The coefficient on eduction, B=0.092, should be interpreted as 1 more year of education leads to an 9.2% increase in wages holding all else constant.

long$age2<-long$age^2/100
Table3.1<-lm(lwage~educ+age+age2+male+white, data=long)
stargazer(Table3.1,type='html')

	Dependent variable:
	lwage
educ	0.084***
	(0.014)
age	0.088***
	(0.019)
age2	-0.087***
	(0.023)
male	0.204***
	(0.063)
white	-0.410***
	(0.127)
Constant	-0.471
	(0.426)
Observations	298
R2	0.272
Adjusted R2	0.260
Residual Std. Error	0.532 (df = 292)
F Statistic	21.860*** (df = 5; 292)
Note:	p<0.1; p<0.05; p<0.01

The coefficient on the education variable should be interpreted that for a 1 additional year of education, your wage should increase by 8.4% holding all else constant. The coefficients on controls shouldn’t be interpreted closely as they are controlling for other factors correlated with wage but not with education such as innate ability, social skills, social norms, etc.