How to Randomize – Part 2 (continued from last week’s post)

There is a third reason that I did not list last week that our process for randomization might be a little different from other ones – multiple families living at one homestead. Many of the households that we will be surveying during the MPAT are on homesteads with multiple generations, and every once in a while, multiple wives for one husband. So, when an enumerator walks into a homestead to conduct a survey, in many cases, she will have multiple households to choose from. We must ensure that our process for picking which households to survey remains random and takes this into account.

Here are two potential ways to try to ensure that this happens:

1. Instruct the enumerators extensively on randomization, what it is, why it is important, and how to do it during the four day training session and subsequent pilot that will occur during the first two weeks of the MPAT. Then, have each individual enumerator cover specific geographic areas that do not overlap at all over the course of the conduct of the MPAT. This is a challenge because, as mentioned in my last post, we do not have good detailed maps of our area. Training is only four days long, and the pilot is only two days. I am not sure we would feel comfortable that the enumerators had a very solid understanding of randomization and how to do it with such a short time of training and very little supervision. (There are a whole bunch of other topics that will be covered during training and the pilot.) We have two MPAT supervisors and two permanent Nuru staff members (Jamie and Rogonga), who will be out and about during the MPAT to ensure things are happening as they should. With only four of them and 11 enumerators in a very spread-out geographic area, using this method to ensure randomization would pose too many practical challenges, I believe.
2. Choose the households to survey before conducting the MPAT in a random manner based on our census data in excel (using a random number generator, for instance) and assign specific households to specific enumerators in advance of them conducting the survey. This does not leave the picking of households nor therefore the randomization process up to the enumerators. If we use this method, though, we could have the entire team of enumerators, supervisors, and staffers go to one village on a given day and not have to worry about overlap. The biggest issue with the use of this method alone is that each set of GPS coordinates is only uniquely associated with a homestead, not a household. So when an enumerator shows up at a homestead, even if she has been assigned a specific household within that homestead via a good randomization process, she will not be able to identify that household from amongst the other in a homestead.

Because neither of these two processes I have (somewhat vaguely) outlined will work for us alone, we plan to use a combination of the two and a few other tools. We will assign specific GPS coordinates to enumerators in advance using a random number generator and our census data, we will instruct them during training about randomization, it will be their responsibility to randomly choose households within homesteads when they show up to polygamous ones, and our supervisors and staff will keep a close eye on everything as it is happening.

Thanks for reading!