INTRODUCTION

Different types of evaluation designs exist to study a cause-and-effect relationship. There are two requirements to create a design. One is to have comparison groups. The basic approach is to form two groups, one with and another without a treatment whose effect is being measured. These two treatment groups are denoted as treatment (T) for the one that receives it and the other as control (C) that remains without it. The basic idea is that units in the two groups are set up in such a manner so that their outputs can be compared for estimation of the causal effect of the treatment.

The formation of a design is complete when the requirement of creating a design, that is, assigning units to the two treatment arms, has been completed. This latter aspect provides the sanctity to the comparison to reveal the causal effect. The issue of assignment of units will be taken up in the next chapter. The present chapter elaborates the three basic comparisons that are widely used in the development of evaluation designs.

To choose a particular design in a given situation, one needs to consider the total error comprising both bias and error that is expected in estimating the effect under it, along with overall cost of the study. Apart from a description and estimation of the impact, in a design, the chapter includes discussion on the biasing effects that can arise and ways and means to handle them. It also specifies the computation of standard error of an estimate. In addition to a discussion on the comparison groups, the chapter provides elaboration on

1. • measuring main and interaction effects,

2. • bias and error in measurement of a treatment effect,

3. • sources of bias,

4. • internal and external validity and

5. • output and its timing.

MEASURING MAIN AND INTERACTION EFFECTS

Measurement of effect of a treatment needs at least two observations (rather two sets of observations) on a relevant output. It requires estimates of the output for T and C so that a comparison between the two estimates would provide the effect of treatment. The inherent assumption, in this comparison, is that the two estimates would be same if the treatment has no effect on the outcome of interest.