Assignment 2

The Impact of Community Driven Development after Conflict

Overview

Community Driven Development (CDD) programs have become increasingly popular in international development over the past several decades. CDD programs directly involve impacted communities in the selection, design, and implementation of development projects. Usually, this also involves building upon or creating community level organizations (or committees) that are involved with the projects, with the goal of creating democratic and inclusive decision-making processes. See this 3ie working paper for discussion of some common features of CDD.

Given these design features, proponents of CDD argue that these programs should (a) improve the quality and impact of development projects (e.g., public goods) AND (b) improve social cohesion, inclusion, and local governance. For this reason, CDD has been implemented in a number of post-conflict settings, where the hope is that CDD can both improve development outcomes and contribute to rebuilding social cohesion and governance (which are often harmed by conflict).

Does CDD improve development outcomes? Does it improve social cohesion after conflict? Your goal in this assignment is to answer these questions, using data from an evaluation of a CDD program in post-conflict Sierra Leone.

The relevant files for this assignment are available in the classwork folder in the homework-2 file.

The GoBifo Program in Sierra Leone

Sierra Leone was devastated by a civil war lasting from 1991 to 2002. In the post-conflict context, the Ministry of Local Government and Community Development implemented the GoBifo (“forge ahead”) CDD project in Sierra Leone from 2005-2009 (with support from the World Bank and other international donors). See this report for a detailed summary of the program.

The program had two key components:

  • Block grants of about 4,667 dollars (about 100 per household in each community) that communities could allocate to local development projects, skills trainings, and small-business investment.

  • “Technical assistance that promoted democratic decision-making, the participation of socially marginalized women and youth in local politics, and transparent budgeting practices” (Casey, Glennerster, and Miguel 2012)

The GiBifo program also included an impact evaluation, conducted by economists Katherine Casey, Rachel Glennerster, and Edward Miguel. The academic paper generated from the project is in the References section (Casey, Glennerster, and Miguel 2012).

Wait until after you complete your assignment to read the full paper.

Importantly, the impact evaluation included randomization: communities in the study were randomly assigned to receive the GoBifo program, or not. In this sense, the impact evaluation was similar to a clinical drug trial, where some subjects are randomly assigned to receive the medicine while others do not receive it. This aspect of the design makes it easier to assess the causal impact of this project: we can compare outcomes in the treatment (GoBifo program) and control groups, and test for whether any differences we observe are likely to be due to chance.

Load Data and Packages

You can use the code below to load the packages and data. Make sure the .csv file is saved in your RProj folder for this project.

library(tidyverse)
library(tidymodels)
myData <- read_csv("gobifo_data.csv")

Data

“gobifo_data.csv” provides a condensed version of the data collected by Casey, Glennerster, and Miguel (2012) for their evaluation of the project. The study team gathered literally hundreds of outcome variables, and the paper includes an interesting discussion of how they analyze the data and interpret the results given the large number of outcomes they measured.

The unit of analysis (or row in the dataset) is the village.

For our purposes, we will focus on a smaller set of variables, summarized below:

  • Treatment variable (named t in the data): this equals treatment if the community was randomly assigned to be in the program, and control if not. As a character variable, tidymodels will automatically recognize it as a factor. But be sure to use quotation marks when working with the two categories, e.g. filter(t == "treatment").

  • Community Decision Making Infrastructure: Is there a village development committee in the community? (vdc)

  • Public Goods: Is there a functioning primary school in the community? (f_psch)

  • Inclusion, Gender equality in participatory decision making (role_wmn).

    • “Enumerator account of how actively women participated in the deliberation compared to men, ranging from 5 = no difference between women and men to 1 = women not active at all compared to men”

  • Social Capital: Trust in other in the community (trust_own)

  • Conflict: Percentage of respondents in the village that reported they had NO conflicts/disputes with others that required outside intervention (no_conflict)

Part 1: Setup and Load Data (10 points)

  • Create a folder and an RProj file for this assignment. Save “gobifo_data.csv” in that folder. Create a new Quarto document where you will complete the work for this assignment. Make sure to include a title, your name, and the date. (1 point)
  • Create a code chunk that loads the packages you will need: tidyverse and tidymodels (1 point)
  • Use read_csv() to read the “gobifo_data.csv” data in to RStudio. Save the dataset as an object called “myData” (3 points)
  • Examine the data. How many observations (villages) are there in the study? You can use glimpse() to determine the number of rows. How many were in the treatment (t) group (in the program) and how many were in control? You can use summarize() for this calculation. (5 points)

Part 2: Generate Estimates and 95 Percent Confidence Intervals (30 points)

There are five response variables in the data. Each of these corresponds to a different outcome of interest: village decision making infrastructure (vdc), public goods (f_psch), inclusion/gender equality (role_wmn), social capital and trust (trust_own), and social conflict (no_conflict).

  • For each of the outcomes: use group_by() and summarize() to calculate the mean of each response variable in both treatment and control villages. (5 points)
  • Select TWO outcome variables of interest to you. For each, use tidymodels to generate 95 percent confidence intervals around these estimates. You should do this for villages in treatment and control separately. (10 points)
  • Store your estimates and the upper and lower bounds of your confidence intervals for future use. Label this object estimates. (5 points)
  • Write a short paragraph that correctly interprets the confidence intervals in treatment and control groups for both outcomes you selected. Based on your confidence intervals, what is your initial conclusion about whether the program had an impact on the outcome variables you selected. (10 points)

Part 3: Graph your estimates and confidence intervals (20 points)

  • For both outcomes you chose, graph your estimate and 95 percent confidence interval in treatment and control groups. You can produce a bar plot or use geom_point, whichever you think best displays the information. This should generate 2 separate graphs, one for each outcome. For full credit, produce graphs that you could use professionally: that is, these graphs should use appropriate labels, colors, and a nice theme (15 points)
    • Bonus 2 points if you can figure out how to include all of this information on one graph!
  • Write a short paragraph interpreting the graphs: what do they teach us about the impact of the program on the two outcomes in treatment vs control. (5 points)

Part 4: Conduct Hypothesis Tests (30 points)

In this section, you wil conduct hypothesis tests that assess whether the GoBifo program had an impact.

  • For the two outcomes you selected: Clearly state the null hypothesis and the alternative hypothesis for each test. Write these out in a list. (5 points)
  • For each outcome, calculate the treatment effect of the program: this is the mean in treatment minus the mean in control. What are the treatment effects? (5 points)
  • For each outcome, use tidymodels to conduct a hypothesis test to assess whether the treatment effects you estimate are likely to be due to chance. For each, produce the p-value. (10 points)
  • Select one outcome: make a histogram of the null distribution from your hypothesis test. Adjust the bar width as needed. Include a dotted vertical line with your estimate on the plot. Generally, what does this graph tell you about how likely your treatment effect is to be due to chance alone? (5 points)
  • Interpret both p-values in language that people who know nothing about statistics might understand. For each outcome (response) variable, do you reject the null hypothesis? (5 points)

Part 5: Interpretation and Conclusion (10 points)

Write 2-3 paragraphs that summarize the results of your analysis and accurately interpret them. What is the big picture takeaway of your analysis? What does your analysis teach us about the impact of this CDD program? Are there policy or program implications of the findings for CDD in post-conflict settings?

Extra Credit

  • There are three outcome variables you did not select: create visualizions that display the means and 95% confidence intervals in treatment and control groups for these three outcomes. (2 points)
  • Conduct hypothesis tests for each the three additional outcomes. What do you conclude from each test? (2 points)
  • Write a paragraph updating your conclusion about the impact of the program and your interpretation of all of the evidence (2 points)
  • Create a single visualization that could “tell the whole story” of the results from this study. (3 points)
Submission Instructions

Head over to Blackboard and go to the Homework 2 assignment. There you will find a link that you can use to upload your project folder.

If you worked with a partner, only one of you should submit the project folder but both of you should still make a submission to Blackboard so that we have a submission to grade. If you have worked with someoone but still have a substantially different project, you can submit separately.

The person submitting the project folder should take a screen shot of the confirmation message (“Finished Uploading”) and share it with the other group members. This screen shot will be what you submit in Blackboard.

Next, click “Create Submission” and write a statement stating whether you worked on the project independently or whether you are submitting jointly with someone else. Also note whether you used a large language model (LLM) like ChatGPT to help you with the work and which parts of the assignment you used the LLM for. It is fine to use an LLM to help with your code but you must acknowledge the use of it in your submission.

Then, upload the screenshot you saved in the first step. Now submit the assignment.

Please remember to upload your entire project folder and not just the Quarto document.

References

Casey, Katherine, Rachel Glennerster, and Edward Miguel. 2012. “Reshaping Institutions: Evidence on Aid Impacts Using a Preanalysis Plan.” The Quarterly Journal of Economics 127 (4): 17551812.