I have been examining several population surveys conducted in Australia dealing with health, disability and the like. Ideally, in Australia the sample population should have a profile reflected in each jurisdiction. What is evident is that the sampling used in one survey had no relationship to the population distribution. I have a particular interest in the interface between ethnicity, disability and homelessness.
It is not just in Australia that one would like the sample profile to match the population profile. The classic way to try and correct for visible bias is to weight the sample so that what I am calling “visible bias” are aligned. By visible bias I mean variables such as gender and age for which there is probably good data from the population.
Of course, the trap in weighting is the invisible biases that get swept along behind the scenes when the sample is weighted.
One of the seminal books on the topic was W. Edward Deming’s “Statistical Adjustment of Data” in 1943. He wrote this when he was, I seem to recall, at the US Bureau of Census.
These days, I suspect that the majority of surveys get some kind of statistical adjustment, often using what is called “rim weighting” where the sample is weighted to match a set of marginal values of known variables…. my “visible biases”.
Does this help or have I totally lost the point of your query?
Viewing 4 posts - 1 through 4 (of 4 total)
The forum ‘Default Forum’ is closed to new topics and replies.