代写Econ 275 (Baseler): Problem Set 2代写Web开发

2025.04.22 - 首页 >> Python编程

Econ 275 (Baseler): Problem Set 2

Total Points: 100

1 Information frictions in the migration decision (40 points)

Although incomes are increasing rapidly in many developing countries, this growth has typically been much faster in urban regions. Today, rural workers in Kenya earn about half what their urban counterparts do, even after controlling for differences in education. Why don’t more people move to cities?

In January 2017, I ran an experiment to test whether bad information about urban wages restricts migration out of rural villages in Kenya. I identified 497 households in rural Western Kenya who had at least one member of “migration age", i.e., between 18 and 35 years of age. I randomly gave out information about typical earnings in 3 big cities: Nairobi (the capital), Kisumu, and Eldoret. I also told households about employment levels for migrants, common jobs in each city, and food price differences.

The dataset migration_survey.dta includes the following data on these 497 households:

Variable name	Definition	Source
b__c1	Number of children under 5 in household	Baseline Survey
b__c2	Number of children between 5 and 11 in household	Baseline Survey
b__c3	Spending on vegetables last week (KSH)	Baseline Survey
b__c8	=1 if household saved any money in past month	Baseline Survey
b__c9	If b__c8=1, amount saved; o.w., missing	Baseline Survey
b__c12	Total spending on school fees in past year	Baseline Survey
b__hh2	Respondent age	Baseline Survey
b__hh3	Respondent gender	Baseline Survey
b__num__rightage	Number of people aged 18-35 in household	Baseline Survey
treatment	=1 if got labor market info	Tracking data
migrated__n__num	Number of migrants sent to Nairobi since 2018	1-Year Follow-up

The reason why I assigned the information treatment randomly is that I wanted to create two groups of households comparable along all dimensions except for whether they knew the information about urban earnings. This random variation in access to information enables me to test whether access to information is a binding constraint in households’ migration decision.

1. Verify that randomization indeed created two comparable groups. For this, you want to check whether the treatment status is correlated with baseline characteristics—check each of the characteristics listed in the table above. Present the results in a table like Table A1 in this paper: https://assets.aeaweb.org/asset-server/files/19428.pdf. You can make the table in Excel if you want. (6 points)

2. How many migrants, on average, did each household in the control group send to Nairobi in the year following the experiment? How many migrants did the treatment group households send on average? (6 points)

3. What is the impact of the information experiment on migration to Nairobi? Is the effect statistically significant at the 10% level? Is the effect size big or small? (6 points) (hint: you should run regressions to answer this question and the following ones.

4. Define a “rich” household as one whose savings in the past month is higher than the sample median, and a “poor” household one whose savings was below median. Is the treatment effect on migration different for poor vs. rich households? Is this difference statistically significant

at the 10% level? (10 points) hint: To see whether the treatment effects are different, you could run the regression of migration on treatment separately for rich and poor households. However, this won’t tell you whether any difference you see is statistically significant. To accomplish this, you should use an interaction term regression to compute the difference in differences between treatment and control households in the rich group RELATIVE TO the poor group.

5. Now let’s use a different proxy for household income: spending on vegetables in the past week.

Is this proxy of household income correlated with migration to Nairobi? Is the correlation significant at the 10% level? Offer one possible explanation for the sign of the coefficient. (6 points)

6. Re-run the regression testing the impact of the information on migration to Nairobi with vegetable spending as a control. What happens to the estimated effect of the information treatment? Are you surprised by this result or not, and why? (6 points)

2 Deworming (40 points)

This exercise is based on Edward Miguel and Michael Kremer, “Worms: Identifying Impacts on Education and Health in the Presence of Treatment Externalities,” Econometrica 72(1): 159-217, 2004. The goal is to think about how attrition and non-compliance affect the way to estimate the impact of a program or intervention.

Background Worm infections account for over 40 percent of the global tropical disease burden. Infections are common in areas with poor sanitation. More than 2 billion people are affected. Children, still learning good sanitary habits, are particularly vulnerable: 400 million school-age children are chronically infected with intestinal worms.

Worms affect more than the health of children. Symptoms include listlessness, diarrhea, ab- dominal pain, and anemia. Beyond their effects on health and nutrition, heavy worm infections can impair children’s physical and mental development and reduce their attendance and performance in school.

Deworming treatment, in the form of oral medication, kills worms in the body (but it does not prevent re-infection). The WHO recommends presumptive school-based mass deworming in areas with high prevalence.

Measuring the impact of deworming children You are looking at the health effects of a school-based deworming program. Your outcome of interest is the worm load (severity of worm infection). You’ve measured it at both baseline and endline. You did not visit students’ home to test them but instead went to the school and measured the worm load for all children found there.

Worm loads are scaled as follows: Heavy worm infections = score of 3; Medium worm infections = score of 2; Light infections = score of 1.

There are 60,000 children: 30,000 in treatment schools and 30,000 in comparison schools. The treatment was randomized across schools.

1. Suppose protocol compliance is 100 percent: all children who are in the treatment get treated and none of the children in the comparison are treated. Children that were dewormed at the beginning of the school year (that is, children in the treatment group) end up with a worm load of 1 at the end of the year because of re-infection. Children who have a worm load of 3 drop out of school if they are not treated. The number of children found in each worm-load category is shown for both the baseline and endline measurements in this table:

Baseline Endline

Worm load	Treatment	Control	Treatment	Control
3	10,000	10,000	0	0
2	10,000	10,000	0	10,000
1	10,000	10,000	30,000	10,000
Total children measured	30,000	30,000	30,000	20,000

(a) What is the difference in average observed worm load between the two groups at endline? (4 points)

(b) Is this outcome difference an accurate estimate of the impact of the program? Why or why not? (4 points)

(d) How could we get a better estimate of the program’s impact? (3 points)

2. Besides worm load, you also measured test scores.

(a) Would differential attrition (i.e. differences in drop-outs between treatment and compar-ison groups) bias your estimate of the program’s effect on test scores? Why? (5 points)

(b)

3. Let’s now relax the assumption made earlier that compliance was 100%. Suppose none of the children from the poorest families have shoes and so they have worm loads of 3. Though their parents had not paid the school fees, the children were allowed to stay in school during the year. Parental consent was required for treatment, and to give consent, the parents had to come to the school and sign a consent form in the headmaster’s office. However, because they had not paid school fees, the poorest parents were reluctant to come to the school. Con- sequently, none of the children with worm loads of 3 were actually dewormed. Their worm load scores remained 3 at the end of the year. Compliance in the control group was perfect though: No child was treated in any control schools. Finally, attrition is not an issue this time: all the children in the sample at the beginning of the year were followed up, if not at school then at home. The distributions of children by worm load at baseline and endline now look like this:

Baseline Endline

Worm load	Treatment	Control	Treatment	Control
3	10,000	10,000	10,000	10,000
2	10,000	10,000	0	10,000
1	10,000	10,000	20,000	10,000
Total children measured	30,000	30,000	30,000	30,000

(a) Calculate the impact estimate based on the original group assignments. This is an unbi- ased measure of the effect of the program, and it’s called the “Intention to Treat (ITT) estimate” of the program effect. In what ways is it useful and in what ways is it not as useful? (6 points)

(b) You are now interested in learning the effect of treatment on those actually treated (called “treatment on the treated” (TOT) estimate).

i. Five of your colleagues are passing by your desk; they all agree that you should

calculate the effect of the treatment ignoring the 10,000 children in the treatment group who were untreated. Is this advice sound? Why or why not? (3 points)

ii. Another colleague says that it’s not a good idea to drop the untreated entirely; you should use them but consider them as part of the comparison. Is this advice sound? Why or why not? (3 points)

iii. Another colleague suggests that you use the compliance rates, the proportion of peo- ple in each group that did or did not comply with their treatment assignment. You should divide the ITT estimate obtained in (3a) by the difference in the treatment rate (the probability of being dewormed) between the two groups. Is this advice sound? Why or why not? (3 points)

3 Perceived returns and demand for education (20 points)

Read the following paper: Jensen, Robert (2012). “Do Labor Market Opportunities Affect Young Women’s Work and Family Decisions? Experimental Evidence from India,” Quarterly Journal of Economics, 127(2), p. 753-792, available here or on Blackboard (file Jensen-qje_qjs002.pdf under PS2).

Please be as concise as possible in your answers to the questions below.

1. What change in labor market opportunities does the paper exploit to answer the question in its title? (4 points)

2. What was the “treatment” the researcher set-up in the randomized experiment? (4 points)

3. What feature of the study context (rural India) made it such that the author expected the “treatment” to affect family decisions (age at marriage and age at first child)? (4 points)

4. How did the “treatment” affect expectations among women? (4 points)

5. What was the effect of the “treatment” on completed fertility? (4 points)