One question that remains to be answered is: How do we estimate or quantify the impact of a randomized experiment?

We have given details about the conditions the experiment must meet in terms of the selection of the treatment and control groups. We also know that the assignment between groups must be random in order to have "identical" groups before the intervention.

But all this ... how do we use it?

We want the assess a program by analyzing its impact on an outcome indicator. If we go back to our previous example, the outcome indicator would be height, and the program would be the school meals program.

We then estimate the impact by taking the average outcome of the group that received the treatment, minus the average outcome of the group that did not receive the treatment.

The latter group tells us how much a child would have grown without the treatment and additionally takes into account special conditions that occur at the time of the intervention (all other factors that we must isolate), which is why we remove this growth in order to find the impact attributable exclusively to the program (or treatment).

If we use graph A as a given result, we find that both children in the treatment group and the control group grew between 2016 and 2018.

As mentioned previously, the beneficiary children grew 10 cm on average. This is because they started with an average height of 120 cm in 2016, and after two years, in 2018, they reached 130 cm. On the other hand, the children in the control group grew 8 centimeters on average, because they started with an average height of 120 cm (equal to that of the children in the treatment group), and two years later their average height was 128 cm.

Therefore, we may conclude that in the absence of the program the growth would be 8 cm and with the school meals program the growth would be 10 cm, which means that the program has a positive impact of 2 cm on average.