Comments Posted By DrSteve
Displaying 11 To 11 Of 11 Comments
I'm an econometrician. I do sampling design, among other things, for a living. The biggest potential issues I see here are (1) whether the clusters were actually chosen randomly (there's no real way to document this); (2) whether there's overcounting of shared members from families spilling across households (selection of adjacent structures beginning with a randomly selected house); (3) whether the protocol for reselecting clusters deemed unsuitable post-selection has any impact on results.
Here's my reaction to my own objections: (1) there's no way to tell; (2) a search for duplicate death-certificate records would clear this up quickly if the researchers retained this documentation; (3) this might be a problem though some (three) of the clusters picked in lieu of the originals seem to have been dropped for "misattribution." I'm not sure it's documented how many (in total) of the original 50 clusters were subject to the protocol that allowed researchers to choose the next nearest population area instead.
I don't like cluster sampling, as it results in less precision per fixed amount of sample and is highly sensitive to the representativeness of the units sampled, but I generally also have the luxury of recommending something else to my clients. What ultimately needs to be done is a study with a much larger sample and a more robust sampling methodology.
Comment Posted By DrSteve On 11.10.2006 @ 14:59
