[Home Page] [Previous Page]
[Old Roundtable Archives] [New Roundtable]
Posted by Mike Curtis on February 09, 2000 at 07:45:18:
In Reply to: Re: Now, you really didn't mean this, did you? posted by anonymous on February 06, 2000 at 02:05:05:
Recognizing that I may be misunderstanding your statement here, I don't think I can agree with it. No amount of weighting can compensate for an inadequate sampling frame. To say that it can implies that... 1) All members of our population are represented in the Internet sample, though perhaps not in correct proportions, and weighting can be used to correct these proportions. 2) We know what the correct proportions are with respect to the variables of interest in our study. If we knew that #1 was true, and we knew the answer to #2, we wouldn't need to do the research in the first place. No amount of weighting to demographic, psychographic or lifestyle characteristics will give your results external validity in any statistically measurable way. Taking a simplistic example, let's say I want to know about flourescent light usage in households. I conduct an Internet survey, then compare my demographic data back to my population and see that my sample is better educated, better off financially, younger and more likely to be male than my population as a whole. I weight the data such that my sample now looks like the population on these characteristics. I still do not have any idea whether or not my sample is biased compared to the total population with respect to flourescent light usage. That is, I cannot say that the expected value of usage in my sample is the same as actual usage in the population, with the only difference being uncertainty due to sampling error. Does it mean we shouldn't use the Internet for research? No, but it means there continues to be serious research limitations associated with it that must be traded off against the limitations of other modes of data collection because we can't apply a simple statistical fix to the problems.
Subject: