代写THdata留学生、代做c++、代写Java/Python编程、代写datafile

日期：2019-06-26 10:45

Problem I (11 Points)

For this problem, joint work with up to 3 other students who are currently enrolled in this course

is possible. Please state the student matriculation numbers of your co-authors in your solution.

Load the datafile THdata new.csv. It contains a cross-section of 1040 working individuals with the

following variables:

Table 1: Description of Variables

Variable Label Description

’wage’ Yi real log hourly wage in Euro

’high’ Di = 1 if unit completed a higher education track (Bachelor/equivalent or higher)

= 0 else

’iq’ Xi1 = score on cognitive ability test at age 16

’motedu’ Xi2 = years of education of unit’s mother

’exp’ Xi3 = potential years of working experience

’exp2’ X2

i3 = potential years of working experience squared

’localINC’ Zi1 = real average labor market income of unit’s district at age 16 to 19

(corrected for long run income level trends)

’localUE’ Zi2 = average unemployment rate of unit’s district at age 16 to 19

(corrected for long run unemployment rate trends)

’tuition’ Zi3 = real tuition fee paid during higher education

’college’ Zi4 = 1 if unit lived nearby a college at age 16 to 19

= 0 else

You would like to investigate to causal effect of higher education on labor market earnings of

individuals.

1. Assume that university degrees are randomly allocated over the sample. Calculate the sample

average treatment effect and evaluate its economic and statistical significance.

2. Compare the measures for cognitive ability and mother’s education between treatment and

control group using means and the normalized difference. What do you conclude?

3. For the remainder of this questions, assume that university degrees are independent of potential

outcomes given ’iq’ and ’motedu’. Estimate the propensity score using a probit model

and briefly interpret the results.

4. Estimate the propensity score density for both treated and control units. What can you say

with respect to overlap and common support?

5. Estimate the average treatment effect using:

Stratification with 10 propensity score strata of equal length

Nonparametric imputation using the propensity score (“Kernel Matching”)

Inverse probability weighting

2

and compare them to a i) the estimate in question 1) under independence and ii) the naive

regression estimate assuming homogeneous treatment effects. What do you conclude?

6. Estimate the treatment effect on the treated using inverse probability weighting. Do higher

education graduates benefit more from higher education compared to non-graduates?

Problem II (4 Points)

You are not permitted to collaborate on this problem. Hand in your own, unique solution. Min./max.

number of words: 600/1200. You may use a limited amount of equations and formula.

Discuss the following statement:

Evaluating a treatment effect always requires overlap and conditional independence. Conditional

independence is a strong requirement on the information content of the observable covariates. Thus,

the more confounding variables we can control for, the better we can rule out selection bias.

3