Experimental Design
Treatment will be assigned at the level of a water tank; that is, the government will assign water tanks to be repaired or restored in a certain phase or not until a later phase. In the first, non-experimental component of the study, the sample is composed of tanks that have already been taken up for repair or restoration. We will identify a sample of 300 comparable treatment and control tanks from phases 1 and 2, stratified by Assembly Constituency (the lowest unit where decisions are made with regard to which tanks are to be selected under specific phase), using propensity score matching method. Phase 1 tanks, most of which have been restored as of summer 2016, will serve as treatment tanks, and phase 2 tanks, most of which have yet to be restored, will serve as control tanks. The observable variables on which the tanks will be matched on include: a) observable tank characteristics such as the command area, catchment area, tank storage capacity, bund length, number of feeder and irrigation channels, b) estimated/contracted cost of rehabilitation and contractor characteristics where available, and c) village (corresponding to the tank location) level characteristics such as population distribution along ethnic lines, political affiliation of village leader (gram panchayat members and President), village level voting share to the party of the local Member of Legislative Assembly (MLA), village level infrastructure, and market connectivity.
In the second, experimental component of the study, the researchers will use a random encouragement design ensuring at least 150 of 300 sample tanks are rehabilitated in phase 3 in 2017. For this second component, we will work with the Government of Telangana to identify a sample of 400-500 tanks such that the tank villages have no other tanks that have been rehabilitated before. This strategy is required to minimize potential spillovers from earlier phases through hydrological and local economy channels. From the sample, we will randomly assign 200 tanks to be rehabilitated in phase 3 and then encourage the government’s tank selection team to implement restoration of most of the assigned tanks. This is because there are engineering reasons that could imply rehabilitation of certain tanks to not be economically feasible. The rehabilitated tanks from our selected list under phase 3 will form our treatment group, which we will compare with the remaining tanks in our sample that serve as control and therefore will not restored during 2017.
A sample of 5 farmers per tank will be randomly selected among all farmers with plots in the tank’s command area (a term for the area served by a tank used by hydrologists and engineers). These will constitute our subject population. Subjects in the tank command area for the 600 tanks in our study (300 in the non-experimental component, 300 in the experimental component) will be informed whether the water tank that services their farms has been taken up for repairs. In the experimental component, we will explain that the order of repairs was randomly assigned, whereas in the non-experimental component, we will explain that the order was based on a set of government guidelines. In both components, the subjects will be asked to answer survey questions on agriculture production on their plots in the tank command area as well as their plots elsewhere. Additionally, a village level survey will be administered to village leaders/village officials to obtain village economy level data.