How to Randomize – Part 1
I must confess part of the process I go through every week to write these blog posts – I re-read whatever it was that I posted the week before. I have to remind myself what had been most on my mind a week prior.
This week something unfortunate happened — and I must confess as well that it’s happened before – the stuff I wrote last week ended up being pretty similar to what I was inspired to write about this week!
If you take a look at last week’s post, maybe you’ll understand and forgive me for my lack of creativity. It’s just that we’re pretty busy right now with one very major thing – getting ready for the MPAT. There’s a lot of work to do related to it, and it does take most of my time, so I suppose I am challenged to focus in on one thing in particular to write about. I guess I feel like if I am going to write about the MPAT, I must write about everything.
BUT, because I did that last week, it is not an option this week. So, I decided I should tell you a bit about one of our challenges – good old randomization. Everyone who conducts any sort of study with using some sample of some whole faces this challenge, and I think we are going to be fine in the end. However, we have a couple of perhaps slightly unique issues that do affect the process we’re engaged in now – planning, of course.
Rogonga, Jamie, Alasdair, and I have done some good collaborative work in creating a sampling frame for our two sub-locations, as I have mentioned here.
The grand-total number of households across both areas is 1,694. To start off with and, we acknowledge, be a little aggressive with our sample size, we decided to go with a confidence level of 99% and a confidence interval of 5. That means our total sample size is going to be 478 of those households.
The households are spread across 24 villages. According to the process outlined in the MPAT User’s Guide, we essentially re-named the villages “clusters”, and used a good randomization process (outlined in the User’s Guide) to pick 15 of the 24 clusters within-which to conduct the MPAT. At the moment, we are finalizing the number of households per cluster that we will survey based on the sizes of the clusters. For instance, if a cluster contains less than 40 households, we will likely survey a sample of 15 households in that cluster, and if it contains more than 80 households, we will likely survey 45 households in that cluster.
So, are you with me? I’ve attempted to outline the process of getting from our big area of two sub-locations to the villages within those sub-locations that we will survey, and now we’re at the point where we’re deciding on the number of households to survey in each village.
The final final step is deciding on which households to survey. The actual doors or doorways upon which our enumerators will knock. Believe it or not, this might be a little more complicated than it seems for a couple of reasons: (1) we do not have good maps of our two sub-locations and (2) the village boundaries are not well-defined agreed-upon lines of distinction like the Ole Ohio River that is flowing ever nearer by my desk as I write this and the rains continue where I am….
To be Continued Next Week….