Finance

Finance

Development economists are often interested in the determinants of production in impoverished parts of the world. An area of particular focus in recent years has been agricultural output.  When farming is successful, there are large positive spill-over effects for local economies. Conversely when crops fail or livestock die, famine and associated problems such as the spread of disease often follow. Consequently, by identifying policies that lead to increased agricultural yields, economic research can contribute to substantial welfare gains.
Your task is to assess the effectiveness of an agricultural assistance program implemented in Tanzania. The assistance program involved visits to small family farms where farmers were provided with some training and were given access to fertilizer. Data were taken on farms that did and did not participate in the program, and information was collected twice – shortly before the program was implemented, and one year afterwards.
The data are available in the workfile Agriculture.wf1. Dummy variables program and after identify farms that participated, and the time period after the program was implemented. Agricultural output is measured in US dollars per person-hectare, and is stored in the variable yield. You also have information on the age of the farmer (age), the size of the farm (size), an index summarizing labour input (labour), rainfall in the last 12 months (rainfall), and an indicator of whether or not the farmer has completed seven years of schooling (school).

1.    To assess the effectiveness of the program a difference-in-differences estimator can be used. Briefly explain the concept of a “parallel trend” for this estimator and what this assumption implies for the agricultural assistance program.

2.     Why can’t we estimate the effectiveness of the program by simply averaging the output rates of farms that did and did not participate in the program? Briefly explain.

3.    Calculate the difference-in-difference estimate of the effectiveness of the program (in terms of output) based only upon average rates before and after implementation.

4.    Give the regression equation that would be used to calculate the full difference-in-difference estimate employing the full set of control variables. Also estimate this equation and interpret the results. Why is this estimate preferable to the estimate given in part 3?

5.    Briefly perform some diagnostics on your model.  Do the coefficients make sense? Are there any signs of misspecification? Are the residuals in line with your assumptions?