Sign up to save your podcasts
Or
Share Introduction to Impact Evaluation - M2_L3
Share to email
Share to Facebook
Share to X
Share to email
Share to Facebook
Share to X
Or
Introduction to Impact Evaluation - M2_L3
Ways to randomize
Flipping a coin, as we have already seen, can randomly assign individuals to two groups.
A lottery where all the names of potential participants (those individuals who will be part of the treatment group and the control group) are put into a bag. The names of the first number of people to be part of the treatment are drawn. If, for example, there are 200 potential participants, the first 100 names are assigned to the treatment group and all those remaining in the bag are assigned to the control group.
Another way, if we want to select half of the participants for the treatment group and the other half for the control group, is to determine that people with IDs ending in an even number will participate in the treatment and odd numbers will not participate in the treatment. This selection mechanism is completely verifiable and is a condition independent of the individuals’ decision. In other words, there is no way to modify it in order to be included in the treatment or the control group.
Produce a random number (it can be done in Excel or in a statistical package), sort it and take one yes, one no. Or take the first half for one group and the second half for another group.
The positive aspect of these randomization methods is that they are completely transparent and verifiable for any individual who is present during randomization or who wants to verify it in the case of identification documents or the random number produced by a computer.
Being able to verify randomization is extremely important, as it is proof that the evaluator or policy implementer does not want to or has no interest in favoring or disfavoring any individual by giving him/her the program or treatment.
Additionally, any of these randomization methods has the benefit that the selection is not related to the outcome variable. For example, the first letter of your name does not make you grow faster or slower, nor does the last number of your ID, and so on.
There may be cases where we want to randomize into more than two groups. In this scenario we must find a way to randomize correctly. A coin would not be useful in this situation. Let's think, for example, that we want to randomize into six groups, in which case a dice would be ideal to perform a random assignment.
View all episodes
By DDICWays to randomize
Flipping a coin, as we have already seen, can randomly assign individuals to two groups.
A lottery where all the names of potential participants (those individuals who will be part of the treatment group and the control group) are put into a bag. The names of the first number of people to be part of the treatment are drawn. If, for example, there are 200 potential participants, the first 100 names are assigned to the treatment group and all those remaining in the bag are assigned to the control group.
Another way, if we want to select half of the participants for the treatment group and the other half for the control group, is to determine that people with IDs ending in an even number will participate in the treatment and odd numbers will not participate in the treatment. This selection mechanism is completely verifiable and is a condition independent of the individuals’ decision. In other words, there is no way to modify it in order to be included in the treatment or the control group.
Produce a random number (it can be done in Excel or in a statistical package), sort it and take one yes, one no. Or take the first half for one group and the second half for another group.
The positive aspect of these randomization methods is that they are completely transparent and verifiable for any individual who is present during randomization or who wants to verify it in the case of identification documents or the random number produced by a computer.
Being able to verify randomization is extremely important, as it is proof that the evaluator or policy implementer does not want to or has no interest in favoring or disfavoring any individual by giving him/her the program or treatment.
Additionally, any of these randomization methods has the benefit that the selection is not related to the outcome variable. For example, the first letter of your name does not make you grow faster or slower, nor does the last number of your ID, and so on.
There may be cases where we want to randomize into more than two groups. In this scenario we must find a way to randomize correctly. A coin would not be useful in this situation. Let's think, for example, that we want to randomize into six groups, in which case a dice would be ideal to perform a random assignment.