This question is about sample size.

We are doing a SMART nut survey in a commune in Conakry, Guinea. So we will be putting in all the numbers in ENA in terms of cluster effect, precision, est. population, approx. prevalence to get our sample size.

Now suppose we wanted to do a survey of all Conakry (5 communes) in order to be able to find significant differences in terms of malnutrition rates between the communes. So, this is not the same thing as doing a city-wide survey to get an average malnutrition rate for the whole city.

How do we calculate the sample size for each commune to make sure we will have enough statistical power? Does it in fact work out to be the same number of samples as if we were doing 5 separate surveys? I hope not…

If you want to do the survey in all Conakry, it would depend on the variability that you would expect to have between different communes in terms of malnutrition rate (in the case of nutrition survey.) If there is a big variability between the communes, then your sample will be very large because you'll have to choose a high design effect. Otherwise, you can do it in all Conakry by entering all the villages and cities in the same ENA file and do it as one survey, but then you won't be able to have the rates by communes because it was not weighted for the size of every commune.

If you want to have the malnutrition rate for evey commune, you'll have to do 5 seperate surveys in 5 seperate ENA files and then you transfer all the data into excel and merge all those files and then calculate the weight of every commune with EPI-Info or SPSS to find the overall malnutrition rate. This method will allow you to have the GAM for every commune and the overall GAM of all Conakry.

It will be easier once SMART version 2 is made available on next spring, and it will allow for stratification and weighting.