Many microfinance programs are introduced in a staggered manner. That is, one group of locations get the program today, another gets the program in 6 months, and so on. Suppose you are asked by your country’s government to evaluate the impact on poverty of a microfinance scheme that was provided to region A in January 2000, region B in July 2000, region C in January 2001, and region D in June 2001. Suppose that (i) you have data on poverty rates in all four regions in January of each year 1999-2002, and (ii) if microfinance has any impact on poverty, it will occur within 6 months.

(a) Describe a difference-in-difference procedure to estimate the impact of microfinance on poverty rates. What will you measure? When? What will the treatment group(s) and control group(s) be?

(b) Suppose that in January 2000, households in regions B, C and D were notified that they would not receive the program until the later scheduled date. As a result, these households began to work even harder and earn more income than before they learned about the forthcoming program. This is an example of what type of bias/effect? Will this effect lead you to over- or under-estimate the true impact of microfinance access on poverty? What would be one solution for addressing this bias?

(c) Suppose that region A is poorer than region B, which is poorer than region C, which is poorer than region D. This suggests that policymakers chose to place the program in poorer regions first. This is an example of endogenous program placement, which can bias our estimates of program impact. Suppose that you can go back in time and design a randomized control trial (RCT) to study the impact of microfinance. How would the design compare to the approach in part (a)?

(d) Suppose that region B is adjacent to region A and some households in region B gain access to the program as early as January 2000. This is an example of what type of problem in program evaluation? How will this affect the estimated impact of the program if we just focused on the difference-in-difference estimate for regions A and B? [Hint: how will YB1- YB0 look in this case compared to a case where region B is very far from region A and no households in B receive the program until July 2000?]

(e) Suppose that it is now the year 2005, the program exists in every region, and you are interested in studying the impact of microfinance on business creation. Someone notices that the microfinance institutions in region B only provide credit offers to households with landholdings less than 0.5 hectares. How would you use this rule to estimate the program impact? Describe the empirical strategy: define the treatment and control groups, state the relevant assumptions, etc.