Using the regression models in part (1c), test whether the election outcome at the discontinuity has statistically significantly different effects for Black and White individuals in these states.

Instructions:

Upload Files:
For the assignment, submit a pdf document that combines the following:
A PDF containing the answers to your questions, with the adequate explanations
and interpretations;
The code (do–file) that you used for the analysis, with comments for steps on each
section;
The log file automatically generated by your code (text file).
You will have to upload these documents before the deadline on Quercus.

[Note: hand in your own solutions]

1. Political Parties and Labor Market Outcomes: Evidence from US States [15 pts]
In an interesting paper, Beland (AEJ: Applied Economics, 2015), estimates causal
impacts of the party allegiance (Republican or Democratic) of US governors on labor–
market outcomes of different groups in society; specifically, Black and White adults in
each state.

He matches gubernatorial elections information in each of the 50 states with March
Current Population Survey (CPS) data with labor market information (e.g., labor force
participation, employment, hours worked, earnings) of a representative sample of Black and White individuals for the years 1977 to 2008.

The goal of this exercise is to implement the RDD in Beland (2015) to estimate these
causal effects for states in which a Democrat gubernatorial candidate barely won an
election relative to those in which a Republican candidate barely won.

Dataset
You should use the Stata dataset posted on the course website to do this, following the
suggestions below. The dataset available for the assignment is a version of the one used by Beland (2015) where we have taken the mean of each variable per state and year, for each group of white and black individuals.1

Below is a description of each variable in the dataset before collapsing to the state–year–ethnic group cell means.

storage display value
variable name type format label variable label

year2 float %9.0g Year
state2 float %9.0g State ID
black2 byte %9.0g Black individual (1/0)
employed float %9.0g Employment status (1/0)
age float %8.0g Individual’s age (years)
age2 float %9.0g (mean) age2
age3 float %9.0g (mean) age3
age4 float %9.0g (mean) age4
female float %9.0g Gender F=1/M=0
separated float %9.0g Separated (1/0)
divorced float %9.0g Divorced (1/0)
widowed float %9.0g Widowed (1/0)
nevermaried float %9.0g Never married (1/0)
elementary float %9.0g Separated (1/0)
somecol float %9.0g Some college (1/0)
coll float %9.0g College graduate (1/0)
moreba float %9.0g More than college degree (1/0)
demwon float %9.0g Democrat candidate won (1/0)
blackdemwon float %9.0g Black ind x Democrat won (1/0)
marginvvv float %9.0g Margin of victory (votes)
marginvictory2 float %9.0g (Margin of victory)^2
marginvictory3 float %9.0g (Margin of victory)^3
dMVv float %9.0g demwon x Margin of victory
dMV2v float %9.0g demwon x (Margin of victory)^2
dMV3v float %9.0g demwon x (Margin of victory)^3
blackMVv float %9.0g Black ind. x Margin of victory
blackMV2v float %9.0g Black ind. x (Margin of victory)^2
blackMV3v float %9.0g Black ind. x (Margin of victory)^3
blackdMVv float %9.0g Black ind. x demwon x Margin of victory
blackdMV2v float %9.0g Black ind. x demwon x (Margin of victory)^2
blackdMV3v float %9.0g Black ind. x demwon x (Margin of victory)^3
marginggg float %9.0g Margin of victory grouping cells

Sorted by: state2 year2

Questions

[Hints: use collapse command, with option “by(marginggg black2) cw”. After
running regressions on collapsed data and using the predict command to generate
fits of the regression models, use the following graph command to generate each
figure: twoway (scatter y x if z==1, xline(c_o)) (line y_hat x if z==1)]

(c) Using the larger dataset made available for the assignment, generate estimates of
the discontinuity in employment rates of Black individuals using linear, quadratic,
and cubic polynomial models on the running variable (margin of victory, or
marginvvv), including state and year fixed effects.

Assess the robustness of the estimates with and without predetermined individual controls (e.g., gender, age, education, marital status). Replicate the empirical exercise for White individuals. Interpret the results. [5 pts]

[Hint: Estimate OLS regression models with i.state2 and i.year2 fixed effects,
cluster standard errors at the state2 level]

(d) Generate analogous RD estimates using the local polynomial RD estimation using
the Calonico, Cattaneo and Titiunik (2016) optimal bandwidth and robust (bias–
corrected) confidence intervals procedure, for Black and White individuals
separately. Interpret the results. [2 pts]

[Hint: use rdrobust command using vce(cluster state2) option]

(e) Estimate the balance of predetermined covariates around the discontinuity as tests
of the RDD continuity assumption using the available data, for Black and White
individuals separately.

For simplicity, you can use the Calonico, Cattaneo, and
Titiunik (2016) rdrobust procedure. Interpret the results. What other tests could
be conducted if you had access to the raw individual–level data? [3 pts]

(f) Bonus question: Using the regression models in part (1c), test whether the election
outcome at the discontinuity has statistically significantly different effects for
Black and White individuals in these states. [No hints] [Bonus: 3 pts]

2. Vulnerability in the Brazil Semi–Arid and Political Clientelism
A long–standing literature seeks to understand whether and how poverty and vulnerability affect political clientelism—the exchange of contingent benefits for political support.

Yet it is challenging to establish a causal relationship, in part due to the difficulty of
disentangling the role of poverty and vulnerability from those of various unobserved
determinants of the phenomenon, such as voters’ beliefs, attitudes and preferences.

In a recent working paper, Bobonis et al. (2021) study this question in the context of
Northeast Brazil, a drought–prone region where vulnerability is prevalent. They employ a large–scale randomized control trial designed in collaboration with a Brazilian NGO to construct private rainfed water cisterns for individual households.

Each cistern captures up to 16,000 liters of water from rainfall or water truck deliveries. The cisterns construction was assigned randomly to neighborhood clusters among a predetermined set of 40 municipalities. They measure individuals’ requests for private help from local political representatives during the electoral campaign in the 2012 municipal elections.

The goal of this exercise is to implement the reduced form (RF) and IV strategy from this project to estimate the causal effects of getting a cistern on these measures of clientelist requests.

Dataset
You should use the Stata dataset posted on the course website to do this, following the
suggestions below. The dataset available for the assignment is a version of the one used by Bobonis et al. (2021).

Below is a description of each variable in the dataset.
obs: 2,809 Post Electoral Data organized by individual vars: 44 2 Nov 2021 22:48
(_dta has notes)

storage display value
variable name type format label variable label

water_tank byte %8.0g LABE Owns cistern (1/0) clusters float %9.0g Neighborhood Cluster of HH ask_any_priv_~p byte %8.0g Ask for Any Private Help
treatment byte %8.0g Treatment (1/0)
ImunID1 byte %8.0g mun ID 1
ImunID2 byte %8.0g mun ID 2
ImunID3 byte %8.0g mun ID 3
ImunID4 byte %8.0g mun ID 4
ImunID5 byte %8.0g mun ID 5
ImunID6 byte %8.0g mun ID 6
ImunID7 byte %8.0g mun ID 7
ImunID8 byte %8.0g mun ID 8
ImunID9 byte %8.0g mun ID 9
ImunID10 byte %8.0g mun ID 10
ImunID11 byte %8.0g mun ID 11
ImunID12 byte %8.0g mun ID 12
ImunID13 byte %8.0g mun ID 13
ImunID14 byte %8.0g mun ID 14
ImunID15 byte %8.0g mun ID 15
ImunID16 byte %8.0g mun ID 16
ImunID17 byte %8.0g mun ID 17
ImunID18 byte %8.0g mun ID 18
ImunID19 byte %8.0g mun ID 19