Why a control group is necessary and what are the conditions it must meet?

We cannot assess the impact of a program by looking only at the outcome of individuals benefited ... we will understand why!

Let’s recall our impact evaluation question: How does the provision of school meals impact the height of beneficiary children?

In the following figure we have the outcome (height) of the children who participated in the school meals program during 2017, the period identified by the green line. The figure shows us that before the program, in 2016, the beneficiary children had an average height of 120 cm. After the program, in 2018, the beneficiary children had an average height of 130 cm.

Given this outcome, could we say that the impact of the school meals program on the height of beneficiary children is 10 centimeters on average per year of intervention?

The answer is NO... because we would expect children to grow up over time anyway, even in the absence of the school meals program.

Let's think: what other factors can modify the normal development of children's height?

Eating habits

Health condition

Physical activity

Between 2016 and 2018 (before and after the intervention), two things happen:

1) School meals program +

2) Other factors that may in turn modify the normal course of the outcome variable (child height).

In the comparison we make in the graph, we do not know whether we are measuring (1) or we are measuring (2), or to what extent both factors are combined.

Therefore, we need to know whether the growth that children experience when they participate in the program is greater than, less than, or equal to what they would experience without the program, the counterfactual. In other words, to know the impact of the program we need to know what would have happened to the height of those children if they had not participated and compare that height with what we can actually observe in 2017.

Since we cannot know what would have happened without the meals program (because the children have already benefited from it) we need a control group (or comparison group). That is, another group of children that has not benefited from the program and that will help us to approximate the counterfactual that we cannot observe.

• What constitutes a good control group?

A group identical to the treatment group composed by non-beneficiaries.

In our example, the best control would be "literally" a group composed by the twin siblings of the children benefited by the school meals program. Since thetwins are not beneficiaries of the school meals program, they will be fed at home.

• The control group must meet the following conditions:

1) It is a group that shows us what would have happened to those treated had they not received the treatment, the counterfactual.

2) The effects of the program should be the same for both groups (treatment or control). This means that if the units in the control group are the ones that receive the treatment (instead of what actually happens: those in the treatment group are the beneficiaries), the impact would be exactly the same as the one we are going to measure.  In other words, the groups are similar to each other, and it is irrelevant which individuals specifically within each group receive the treatment.

3) External factors, which affect all children, should have the same effect on the comparison (control) group and the treatment group.