Sampling
Study Population
The data is only meant to represent the 308 localities surveyed. The results cannot be generalized to a larger population. The objective was not to produce estimates of national means and totals, but to estimate the parameters of an analytical model of program impact.
Sampling Procedure
In the present application, the approach that is being used, in lieu of randomization, to select a control sample is statistical matching. A matched-pairs design was used, matching 174 (154 plus 20 replacements) treatment localities to 174 control localities using nearest-neighbor matching. Sampling was restricted, as mentioned earlier, to localities having population 1,000 or more (according to the 2000 Census) and to the 20 largest localities in each district.
The treatment population included all localities within 120 minutes estimated travel time of the nearest MiDA program road, and the control population included all localities located more than 120 minutes estimated travel time from the nearest MiDA program road. (The estimated travel times were calculated using a GIS model of the Ghana road network (documented separately).) This resulted in resulted in population sizes of 675 treatment units and 848 control units. Sampling was restricted to all of the country except Western Region.
Matching was based on a number of variables, including population, travel time to Accra, travel time to the nearest MiDA feeder road, and physiographic data.
The sample localities occur at all distances from the program roads, since it was desired to have substantial variation in the travel time to the program roads
Because of the sample design process, the sample has reasonable spread, balance and orthogonality for a large number of design variables. Also, the sample includes a control sample for which the units are individually matched to units in the treatment sample. The sample will be a very good one for use in estimating an analytical model showing the relationship of program impact (price changes) to the Ghana MiDA feeder-road improvements, and for estimating a double-difference estimate of program impact.
Deviations from Sample Design
Of the 308 sampled localities only one locality was removed from the sample because we were unable to locate it. This locality, Choo #0155, was not located and was removed along with its matching pair, Sabiye #0159. These localities were replaced with Suame #0812 and Ogbodzo #1264. All other localities were located and surveyed.
Weighting
No weighting is used in the dataset.
Documentation in PDF