Causal Inference and Data Analytics

1 Form of government, corruption and freedom of speech

We are interested in the relationship between democracy and corruption as well as freedom of speech. To study this relationship, we go back to the Quality of Government Basic Dataset (see the qogProjectUpdated/data folder). We will use the ht_regtype1 variable (see the QoG codebook) to classify the countries into democracies and non-democracies (autocracies, etc).

a. Define a new binary variable democracy that is 1 in case the country is a democracy (according to ht_regtype1) and 0 otherwise.

Answer the following questions, both graphically (i.e. by showing suitable plots) and by employing hypothesis tests. For hypothesis testing, carefully follow the steps we outlined during the lecture, and clearly formulate the (null and alternative) hypotheses. Hint: Consult the “Hypothesis Testing Cheat Sheet” in the Canvas Module 2 folder.

b. How does corruption (as measured by the Bayesian Corruption Index, BCI) differ between democracies and other forms of government? Compare the distributions of the two classes of government graphically, and find appropriate tests by consulting cheat sheet. Specifically, (i) test whether the mean BCI is different between democracies and non-democracies. (ii) Are the two distributions different? Interpret the results!

c. How does freedom of speech differ between democracies and non-democracies? The variable ciri_speech categorizes countries into three different groups regarding freedom of speech. Produce a 2 (democracy, non-democracy) × 3 (freedom of speech) contingency table and find a suitable way to present the differences graphically. Consult the cheat sheet to find a test that can answer the question whether there is an association between the form of government and freedom of speech. Administer the test and interpret the results.

1

2 Does Islamic political control affect women’s empowerment?

In this part, we have a look at the causal research question studied by Meyersson (2014) “Islamic Rule and the Empowerment of the Poor and Pious”, Econometrica, 82(1), pp.229-269. More specifically, we study the question whether Islamic political control (as measured by whether the Islamic party won local elections) affects whether women complete the secondary school or not. We are not digging very deep into this research question, but apply a simple (parametric) regression discontinuity method to the data collected by Myersson

(2014). A preprocessed and reduced data set is enclosed to the original announcement of the problem set (see Mysersson2014.csv). The original data can be found on the Econometrica website.

The dataset contains observations on municipalities in Turkey (each row refers to a municipality). It contains two variables:

• schoolFemales: the secondary school completion rate for 15 to 20 years old women

• margin: The margin of the Islamic party’s win or loss in the 1994 election. The variable is coded similar to the Democrats’ margin in the “Causal Inference in Practice” document: Values larger than 0 mean that the Islamic party won, and values smaller than 0 mean that the Islamic party lost.

a. Generate a (binary) treatment variable islamicWin indicating whether or not the Islamic party won the 1994 election. Summarize the variable.

b. We start with a naive approach and compute secondary school completion rates (see the variable schoolFemales) conditional on regions where the Islamic party won and lost in 1994 (hint: use group_by()). Is the difference in women’s secondary school completion rates between regions where the Islamic party won and regions where it did not win a plausible estimate of a causal effect? Where do you see problems with this approach? Argue!

c. Apply the regression discontinuity method. We first need to define which observations to the left and right of our cutoff (the margin of 0) we take into account. We set the bandwidth to [−0.2,+0.2]. Estimate a linear regression model with schoolFemales as a dependent variable. Report and interpret the results (Remark: Focus on the coefficients, their sign, magnitude and significance. You can ignore other parts of the summary() output).

d. Present the results graphically (Hint: You may follow the “Causal Inferen