Tuesday, June 15, 2004


At least two of them are, anyway. David Winston of Roll Call--hyped by Matt Drudge--proves he does not understand basic concepts of statistics, namely methods of statistical sampling. Just because a sample population does not reflect demographic trends in the population as a whole does not mean that a sample is biased. (In this instance, Winston gets his panties in a ruffle because the sample of a recent LA Times poll included 38% Democrats and 25% Republicans).

Here is how statistics work:

Some small portion of a population is chosen as a sample. Statistics are applied to the sample and the results are used to state something about the original population. Choosing a sample so as to replicate exactly the demographic attributes of the population in which one is interested is what statisticians call biased sampling because hand picking a sample to reflect the underlying population biases the results. When each member of a population has an equal chance of being selected for a sample, statisticians call it unbiased sampling even if it is not a perfect representation of the underlying population.

I do not have enough information on the sampling methodology of the poll in question, though it does appear to be unbiased. The composition of the sample, however, certainly does raise questions as to the statistical significance of the results.

While not biased, the results of the poll probably mean, in the words of Kurt Vonnegut, doodley-squat. This would be a fair way to criticize of the poll. Crying partisanship is not.

I'm still putting my money on the Iowa Electronic Market as the predictor of choice in who is leading the race for the White House. Markets do tend to aggregate information.


Post a Comment

Links to this post:

Create a Link

<< Home