Question

The Millenium Villages Project (MVP) and the controversy around the measurement of its impacts. This question will explore that more thoroughly as a way of getting you to think more deeply about the causal inference methods.

(a.) The MVP aims to tackle all the root causes of poverty at once by implementing multi- faceted assistance: medical centers, education initiatives, support for improved agricul- tural productivity, etc. Describe one advantage and one disadvantage of an evaluation of the package as a whole compared with an evaluation of the individual component interventions.

(.b) Putting that critique aside, suppose that we are interested in measuring the effect of being a Millenium Development Village (the “treatment” variable, D) on the incidence of mortality for children under 5 years (child mortality). Using both notation and a short description, describe the

i. true causal effect for village i
ii. average counterfactual outcome for villages that never get MVP support

iii. average causal effect for villages with MVP support

(c.) Suppose that a researcher decides to do a “before-after” comparison of the mortality rate in 2009 in MVP villages with the mortality rate in 2006 in the same villages. Write down the assumption needed for the before-after comparison to represent the true causal effect of MVP support. Is this assumption likely to be satisfied?

Answer 1

a.) The Millennium Villages Project (MVP) is aimed at improving poverty situation in African countries covering the area of primary education, primary health, clean and safe drinking water, sanitation, infrastructure development, business development, Agricultural support such as the provision of fertilizers, insecticides, etc. The critics around this project are that it may be expensive in comparison to taking up individual programs aimed at one of the selected areas above. For example, excessive use of fertilizers to increase agricultural productivity may have negative externalities. The positive side of the program is that it ensured sanitation, reduction in malaria incidents by 50% as measured by Overseas Development Institute. The increased agricultural production catered to the needs of school meals programs etc.

b.) The program effectiveness can be measured through a regression equation where response variable will be mortality rates while dependent or predictor variables will be sanitation cost incurred, number of children undergone vaccination, etc. Mathematically it can be represented as equation 1

b II) average counterfactual outcome for villages that never get MVP support will be alpha

biii) Yhat is the average causal effect for villages with MVP support. This is computed given equation 1. One can add as many independent variables as possible as long as they are theoretically believed to have an effect on mortality.

where beta1 is the coefficient associated with sanitation cost and beta2 is a number of children vaccinated for Malaria. These coefficients indicate the impact on response variable, i.e., the mortality rate for a unit change in predictor variables. Alpha represents intercept, i.e, mortality rates when no sanitation and vaccination is done. Different equations may be constructed for different villages or villages can go as one of the predictor variables. i.e., x3.

The equation looks like as follows, in case of the image not properly visible.

Y(Mortality)= alpha + Beta0*X1 + Beta1*X2+Error

c.) There are two ways of comparing before and after MVP implementation on mortality rates.

First method: collect 3 of monthly mortality figures leading up to 2006 and again from 2007 to 2009 as the second group. This will constitute 36 months of before data points and 36 data points after. Check for normality and conduct paired "t" test to confirm the effect of MVP. If p-value is less than 0.05(assuming 5% significance level) then MVP is effective. if group1 and/or group 2 is non-normal, use Moods Median test which is a non-parametric test.

Second method: Conduct a proportion test. First proportion being mortality up to 2006 and second being between 2006 & 2009. If the test is significant at 5% level (i.e., p<0.05) then MVP is effective.