Research strategy

Before embarking on an impact assessment study, as discussed in the Study Design guide, it is important to consider beforehand

• how the data might be analysed
• how observational units are to be defined for the purpose of analysis
• what approach should be used for defining appropriate baseline values against which the results of the intervention can be compared.

One method for obtaining baseline data is to collect data from a parallel set of subjects that do not receive the intervention. One herd of cattle used for tsetse control purposes at Ghibe (not one of those in the present case study) was raised on a plateau in an area above a valley where the remaining monitoring herds grazed. Tsetse control applied to the plateau area had no effect in the valley where tsetse control was not being applied at the time. Thus, it was decided to use data collected from herds in the valley as controls. Mean values from these 'neighbouring' herds were used as a covariate in an analysis of covariance to adjust for random year-to-year variations in each of the response variables.

But the values in the two areas were poorly correlated presumably reflecting differences in environment and in management practices. Likewise the terrain and vegetation grazed by the herds also could not be considered identical in every respect. Thus, this approach was of no value.

Another problem that can arise when control and intervention farmers are studied in parallel is that that the intervention starts to be adopted by the control farmers when they see the benefits of the intervention (see Case Study 6).