Econ 275 (Baseler): Problem Set 3

** Due Thursday April 3 at the beginning of class**

** Please use a pen and not a pencil if you hand-write**

1 What is the impact of microfinance?

What impact do micro-loans have on the welfare of borrowers? This problem illustrates different methods of evaluating impact.

1.1 Newspaper article: Huge gains for microloan clients

“Statistics released today reveal that prominent South African bank, SAB, and its new micro-lending program to working class clients in Johannesburg and Cape Town has led to a 78 percent increase in client household incomes. When clients approached SAB, their typical monthly income was 3070R; when households were interviewed 12 months after taking their loans, their incomes had jumped to 5454R. This provides only further evidence to what has already been recognized – largely via the 2006 Nobel Peace Prize to Muhammad Yunus and his revolutionary Grameen Bank – as one of the powerhouse new methods to fight poverty and raise incomes worldwide.”

• What do you think of this article’s claim?

1.2 Letter to the Editor

I have been a follower of the microfinance revolution since its beginnings in the 1970s. And while I’m happy to see South Africa join the rest of the world in exploring this exciting new space, I have some qualms with your newspaper’s article from last week. Indeed I’ve done my own calculations – and I found a healthy impact of 51 percent, not 78 percent. Indeed, while SBA borrowers had an income of 5,454R after 12 months, those who did not take loans had an income of 3,616R at that time. So the correct comparison is 5,454 to 3,616 and that is a 51% difference. Still something to be proud of, but almost 30 percent less than what you found!

• What do you think of this reader’s claim?

2 Kenya household dataset

This problem uses household data collected by KNBS (Kenya National Bureau of Statistics) be-tween 2015 and 2016. On Blackboard, under the problem set 3 folder, download the datasets called HH_Information.dta and Consumption_aggregate.dta. Also download the do-file called PS3_dofile to get yourself started. The survey questionnaire (“Questionnaire_Q1B.pdf”) is also available and may be helpful for finding the right variables. The section “STATA TIPS” at the end of this problem set contains some notes on the dataset and how to handle some of the variables.

* Please attach your completed do-file to the end of the problem set. Please do not hand in a log file. *

1. A number of variables are expressed in Kenyan shillings (KSh or KES). What was the exchange rate between the USD and KSh around the time of the survey? What does 100 KSh at the time correspond to in USD at the same time?

2. Researchers often do not use purely random sampling. For example, we might over-sample poor households if we want to study poverty - that means that the probability of appearing in the dataset is higher for poor houseohlds than rich households. The reason we do this is because we want to have enough poor households in our sample to make precise statements about them. In this case, we have to re-weight our data to reflect the fact that the sample is non-random; otherwise, the mean income in the sample will be much lower than the popu-lation mean. The data you are working with for this problem set used a two-stage sampling design: that is, they first selected regions within Kenya, and then randomly selected an equal number of households per region. They reason they did this is that they wanted to have enough households even in places with low populations. The do-file posted on Blackboard gives you the commands to re-weight the data to adjust for this sampling design. Without the weights, the sample would over-represent households in sparsely populated areas. Cal-culate the weight-adjusted average household size using “svy: mean” and compare it to the unweighted average. Which one is higher? Why might that be? Hint: Remember that the design over-sampled households in sparsely-populated areas.

3. The do-file includes a command to merge the household information data with data on con-sumption aggregates (i.e., consumption totals across household members). How many house-holds are there in the HH_Information dataset? For how many of those households do we have information from the Consumption_aggregate dataset? Hint: Look at the _merge_consumption variable created within the merge command.

4. During the industrial revolution, urban mortality was higher than rural mortality across a broad set of contexts, in part because sanitation in cities was much worse. Test whether access to improved sanitation technology is significantly different in urban vs. rural areas for the following outcomes:

(a) Presence of an improved toilet (flush toilet or Ventilated Improved Pit (VIP) latrine)

(b) Presence of a hand-washing station near the toilet

(c) Access to improved waste disposal (any method other than dumping or burning in the open)

5. The Consumption_aggregate dataset includes information at the household level on total consumption expenditure and total food expenditure per adult equivalent (consumption per adult equivalent is similar to consumption per capita, but adjusts for the fact that children do not need to consume as much as adults). Generate a new variable (call it food_share) that measures the share of total consumption that the household spends on food.

(a) Plot the mean of food_share for urban, rural, and peri-urban locations (use the variable “eatype”). I suggest you use the “graph bar” command (type “help graph bar” to get instructions) along with the “over” option. There’s no “graph” command in the svy prefix that we’ve been using, so you’ll have to enter the survey weights directly into the command. So, for example, “graph bar var1 [pweight=weight], over(var2).” Save this graph and include it with your submission.

(b) In which areas (rural, urban, or peri-urban) do people spend the highest share of their budget on food? The lowest? Why do you think this might be?

(c) What is the definition of peri-urban? Based on this measure, do peri-urban places look more like rural areas or urban areas?

(d) Regress food_share on the logarithm of per adult equivalent total consumption (the denominator in food_share). Don’t forget the “svy:” prefix! Is your estimated coefficient positive or negative? Does that mean that food is a necessary or a luxury good? Hint: Look up the definitions of necessity and luxury goods if you aren’t familiar with them. To create a log transform, use “gen newvar = log(oldvar).”

STATA TIPS

• Remember that often variables have “value labels.” This means that when you look at tabula-tion tables of numeric variables, Stata shows you labels like “Yes” and “No” even though the underlying data is a number. To see the values associated with each label, use the “codebook” command. For a variable called var1, I would type “codebook var1.”

• Be careful not to run a regression on a Yes-No variable with code 1 for Yes and 2 for No. In that case a positive coefficient means the “No” group has a higher mean! You’ll want to generate a new variable equal to 1 when the answer is Yes, and 0 when the answer is No. You can also recode the original variable, but this can be dangerous when you forget which variables you’ve recoded.

• You may find the command “lookfor” useful for finding the right variables. “lookfor” searches for instances of whatever string you specify in variable names and labels. E.g., “lookfor consumption.”

• To collapse a variable that takes on several values to one that takes on only two (ex: for questions 4a and 4c), use the symbol | which means “or.” Example: to generate a variable named var2 which is equal to 1 when var1 is 1, 2, or 4, and 0 otherwise, write “gen var2 = (var1==1 | var1==2 | var1==4).”