PPGA503 HW5 prompt 2022 Due Nov 28 This assignment will build on HW4, specifically by utilizing more advances techniques to increase robustness and generate more nuanced insights. Please use dataset hw5. Please add the new commands for hw5 onto the .do file you wrote for hw4 (meaning you are extending your existing .do file). You will turn in both a write-up of your analysis and the complete .do file you used for hw4 and hw5. As a reminder, the dataset is a modified version of the World Bank’s Indonesia Database for Policy and
Economic Research (INDO-DAPOER). You can access the background material in the following places:
here and here. The dataset and a list of variables are available on Canvas (under week 11 module). The dataset includes a significant number of variables covering health, education, governance, economic, development, and natural resource attributes at the district level in Indonesia. Note that Indonesia, aside from having the fourth largest population in the world (spread across thousands of islands stretching approximately 5k kilometers from west to east), is also among the most decentralized countries in the world. This creates immense variation in both governance and developmental outcomes across the districts.1 There are slightly over 500 districts in one of two types: kota (city) are more urbanized, while
kabupaten (regency) are typically more rural. Provinces, of which there are just under 40, form the meso (middle) tier of government. As this is an extension of hw4, we continue to focus on variation in health levels, meaning Morbidity rate (in %) remains the DV. 1. We noted that Indonesia is a highly decentralized country. Districts, which provide a significant portion of healthcare provision, rely on several sources of funding, most notably (1) Local funding for health (2) a Specific Allocation Grant for Health (Basic Services), which is a federal-level grant that the Indonesian government allocates to districts’ budgets for healthcare spending. Please include these two variables into model 4 (from hw4). This will become model 1 in hw5. You may want to (hint, hint!) rescale one or both variables, but ensure that you properly name and label the variables accordingly if you do. What effect do these grants have on healthcare outcomes? Compare the local and national funding impact on morbidity rates — which rupiah is more effective? Do they impact your overall conclusions in a meaningful way? 2. Next, we want to examine how Household Access to Safe Sanitation (%) may affect Morbidity rates. In this case, it is reasonable to theorize that access to safe sanitation and its impact on morbidity rates depends on urbanization — see graph below for an initial exploration of this relationship. So, we decide to use the Kota dummy (which takes a value of 1 for kota and 0 for kabupaten) to create an interaction term Kota * Household Access to Sanitation to add to our new model 1. This is our model 2. After you run this model, use the margins command (margins VAR, dydx(VAR)) to estimate two slopes.
1 If you are really intrigued by the case and interested in further reading, you could give this a try: “Indonesia’s Decentralization Experiment: Motivations, Successes, and Unintended Consequences” by Ostwald, Tajima, and Samphantharak (2016).


3. Finally, I’d like you to critically think about our model. Given your (whether intuitive or expert) understanding of variation in health outcomes, is there anything important missing from the model? Go back to the variable list to see whether it can be refined further. You may add up to three additional variables, as you see fit. Estimate the new model and include the findings as model 3. Interpret the results in 200 words or less. Be sure to address what has changed, what hasn’t, etc. 4. As you’ll recall, we’ve noted that OLS can be highly subject to bias from outliers that exert a disproportionate influence on outcomes. Please assess for this (use the lvr2plot) to help you. If you are concerned about any observations, take a look at what they have in common. Can you control for what makes them unusual? Please take action if necessary. In your writeup discuss: (1) what you found and how, and (2) what you did in response, in less than 200 words. If this involves changes to your model, please make those changes, but don’t yet include them into your regression table. 5. Now let’s look for violations of the GM assumptions that may be biasing our findings. Assess for heteroskedasticity and multicollinearity. As before, describe what you found (and how), then what you did in response, in less than 200 words. If you made changes to your model from question #4 and/or #5, please show the updated findings in your regression table as model 4. 6. Let’s think like policy makers. Let’s assume for a moment that a given amount of funding could achieve one of three things: a. 10% increase in puskesmas density b. 5% increase in physician density

c. 2% increase in national grant for healthcare (basic services) Across all of Indonesia (meaning model 4, but without the controls for the different islands), which of these interventions would you predict might have the greatest effect on decreasing morbidity? (Hint: use the post-estimation margins command to specify values. Remember that if you don’t specify a value for a given variable, Stata will use the mean value). Now, refine your recommendations by assessing whether they hold (or change) depending on whether we’re looking at rural areas (population density == 100) or urban areas (population density == 5,000). Keep your response to 200 words or less. Good luck and hope you find some of this enjoyable (or at least interesting!) !