I noticed today a huge disparity in the results for the so-called generic congressional ballot polls that illustrates the problems that afflict many of the highly publicized polls. Reuters-Ipsos asked about 900 registered voters if they prefer the Democrats or Republcans for Congress and Democrats won by 48% to 44%. Rasmussen asked 3000 likely voters the same question and the result favored the Republicans by 44% to 38%. In other words, there was a full ten percentage points difference in the results of the two polls. It is true that Reuters questioned registered voters and Rasmussen spoke to likely voters who normally are slightly more Republican than registered voters as a whole, but this is not enough to explain the difference. The answer came, however, when I looked at the composition of those who were polled. Rasmussen adjusts the numbers of Democrats and Republicans in the sample so that they follow the percentages who so identified during the previous month. In other words, Rasmussen uses the identification given by about 50,000 respondents the previous month to weight the results of its poll. Reuters, however, just picked people at random and took their answers without adjusting for party preference. As a result of this different methodology, Reuters included more than 10% more Democrats in the sample than did Rasmussen. Since Democrats tended to heavily favor Democrats for Congress, this different methodology totally explains the differences in the results.
The use of skewed sample groups is a common event in many polls. Next time you see a poll with an unbelievable result, keep this in mind. Most likely, the result was manufactured by using a skewed sample.
No comments:
Post a Comment