Tuesday, March 29, 2011

Ask CRRC | Sampling Weights II

Now, let’s move from this simple example to the Caucasus Barometer. When selecting respondents for the CB, CRRC first divides the country into nine geographic divisions: the capital, urban-northeast, urban-northwest, urban-southeast, urban-southwest, rural-northeast, rural-northwest, rural-southeast and rural-southwest. Within each of these nine groups, nine separate samples of voting precincts are randomly selected. Households are randomly selected within each selected voting precinct. Then, a single adult respondent is randomly selected within each selected household.


How do we calculate the sampling weight of a CB respondent? First we calculate the probability that the respondent was in our sample. There are three steps to this process since there are three stages of random selection. Suppose that a respondent is a woman who lives with her husband, her husband’s mother and father, and her two young children in an apartment in voting precinct #2 of district #3 in Saburtalo. Voting precinct #2 of district #3 has 624 households and 18 interviews will be completed there. Tbilisi has 712 voting precincts in total, of which 50 are selected for sampling.

The first step is to calculate the chance that the woman’s voting precinct was selected for sampling, which is 50 in 712. The second step is to calculate the chance that her household was selected once her voting precinct had already been selected, which is 18 in 624. The third step is to calculate the chance that the woman herself was selected once her voting precinct and household had been selected, which is one in four—the four being the four adult members of her family. We can put those three selection probabilities together by multiplying them. This gives us the chance of this woman being interviewed:


The number of adult Georgians that she represents can be calculated as


Individuals living in different regions of the country, in different voting precincts, and in different size families have different probabilities of being selected for the sample. Thus, they have different sampling weights. Therefore, it is important to use an appropriate data analysis program and to use the sampling weights when making estimates about the greater Georgian population from the CB sample.

No comments: