Data Analysis Assignment 3

Data Assignment 3 is due ???

Overview

For this assignment, we will be using Rounds 5, 6, and 7 of the Afrobarometer Surveys. The Afrobarometer data contains information on political attitudes, values, and beliefs from representative population samples in 36 African countries. We will combine the Afrobarometer data on information about female heads-of-state in Africa to estimate the impact of female leaders on gender attitudes in a two-way fixed effects framework.

Getting Your Data

You can download the Afrobarometer from here. There is a single merged file for each round (containing data from all the countries included in that round). You need to download the data from Rounds 5, 6, and 7. The data is in SPSS format, but you can convert it to Stata format using the import spss command. It makes sense to save a separate Stata data set for each round because the same variable name is sometimes used for a different variable in different rounds.

Data on female heads-of-state in Africa is available here. The data is in Excel format.

Cleaning Your Data

Start by creating a cleaning do file. Record each step involved in cleaning your data in the do file, so that you can replicate them easily.

The first step is to keep only the variables you need from each round of the Afrobarometer data. The following variables are common to all the Afrobarometer data sets:

• COUNTRY is a numeric country code
• DATEINTR is the date of the Afrobarometer survey interview
• URBRUR is an indicator for urban vs. rural locations
• Q1 is the respondent’s age
• Q97 is the respondent’s level of education
• Q101 is the respondent’s gender

You also need to keep data on our key outcomes of interest. In each round, respondents were asked whether they agreed more with the statement

Men make better political leaders than women, and should be elected rather than women.

or the statement

Women should have the same chance of being elected to political office as men.

Responses to this question appear as variable Q22 in Round 5, as Q18 in Round 6, and as Q16 in Round 7.

To build your combined data set, keep only the variables you need from each round and then use Stata’s append command to put them together. Then familiarize yourself with the variables in your combined data set. Generate new, clean variables for use in your analysis - setting any variables to missing as needed (for example, if someone answers “Don’t Know”).

The variable DATEINTR is in Stata’s date format, which can be a challenge (though there is plenty of online documentation). To create numeric variables for the month and the year in which a survey interview took place, use the code

gen year = year(dofc(DATE))
gen month = month(dofc(DATE))

To generate a dummy variable for interviews that occured when a female head-of-state was in power, you might want to merge in the data from the spreadsheet on female leaders. Or, you might just want to code this variable by hand, since the number of female leaders is small. To do this, you will need to use Stata’s date format (again). Here is how I would code up a dummy variable for observations in Malawi during the period when Joyce Banda was the president:

gen joycebanda = 0
label var joycebanda "Joyce Banda was president"
replace joycebanda = 1 if (country=="Malawi" & dofc(DATE)>td(7apr2012) & dofc(DATE)<td(31may2014))

You can extend this code to instead indicate whether any female leader was in power.

Analysis

Your analysis should include the following:

1. A summary statistics table
2. A regression table
3. A falsification test using your main specification, but with different outcome variables
4. A falsification test using an alternative definition of treatment

Your summary statistics table should include data on all the variables used in your analysis. It should be nicely formatted (that goes without saying at this point).

Your main regression table should report estimates of the causal impact of having a female head-of-state on beliefs about women politicians. In addition to the independent variable of interest, you should include country and year fixed effects and cluster your standard errors at the country level. You should also control for age and educational attainment. Column 1 should look at the entire sample, Columns 2 and 3 should show separate results for men and women, and Columns 4 and 5 should show separate results for urban and rural areas.

Your third table should include a falsification test: you should look at the “impact” of having a female head-of-state on respondent gender, age, educational attainment, and rural or urban residence in a two-way fixed effects framework (controlling for year and country fixed effects).

Your fourth table will look at a different falsification test: does having a female head-of-state one year in the future impact gender attitudes? To explore this hypothesis, you need to create a dummy for having a female head-of-state next year (i.e. in the year after the survey). Regress gender attitudes on this variable, controlling for country and year fixed effects.

Presentation

You should present your results in a 3 to five page note (i.e. a mini-paper), as you have for your other assignments.

Your note should begin with an Introduction. Motivate your analysis in reference to the research question. Do not begin with a narrative opening sentence like “For this assignment, I decided to look at…” In your first sentence, explain why the analysis you are about to present is of interest to an economist studying gender. Then use the remainder of your introductory paragraph to explain what you will do, and why you think your results are important.

A Data section should come next. Describe the data sets used in your analysis, and present your summary statistics. How do country-years with a female head-of-state differ from the rest of the sample?

The rest of your note should describe your Analysis. Present your results. Devote at least a paragraph to each table of results. Explain the regression specifications that you use. Discuss your coefficient estimates and their level of statistical significance, and summarize the main conclusions from each table.