Americans want the broom, prefer Democratic Congress

Comments

3 comments posted
AP-Ipsos

Amazingly enough, that 49% to 36% poll rating matches the makeup of the sample. In short, when 49% of the people polled identify themselves as Democrats & 37% of the people polled identify themselves as Republicans, I'd expect that poll to turn out that way.

Considering that the official statistics on party registration on Election Day, 2004 was an even 37%-37% split, this AP-Ipsos poll is meaningless.

The truth is that I've been a poll studier for quite awhile now & AP=Ipsos is the least reliable of all polls because (a) they vastly oversample Democrats each time out and (b) the questions they ask are designed to elicit the desired response.

If you're going to go by polls, trust Rasmussen & Gallup's own poll. Most media polls are worthless, too.

Gary Gross's picture
Posted by Gary Gross (not verified) on 9 January 2006 - 1:21am
Poll validity

Rassmussen in particular had excellent predictive ability in the last election. Virtually all of the actual vote percentages were within the margin of error of their predictions. They follow good statistical methodology and are very transparent in its application.

One of the reasons so many polls are inaccurate is they are based on formulas derived in the 60's. All polls use a form of stratified random sampling instead of a pure random sample. However, for this to work you must accurately identify the strata. Democrats have tended to be oversampled for four reasons.

First, Democrats used to be a much larger portion of the population in comparison to Republicans. The ratios have since evened out. Second, even when an area had the same percentage of Democrats and Republicans, it was valid to oversample the Democrats because they were able to get out a larger percentage of their base to vote on election day. That is no longer true. Data from 2004 suggests the parties are either at parity or the Republicans have a slight advantage here. This particular methodological error caused a lot of the election day voting polls to yield biased results in the 4-8% range. Third, polls are conducted by telephone, but only on land lines, not cell phones. This is causing all kinds of errors in the sampling, but the bias is harder to predict. Socioeconomic, regional and time of year factors all heavily influence the results. Fourth, depending on the contact method, it has been found that Democrats and Republicans actually bother to answer the poller's questions at a different rate. This was especially true of election day voting polls. Democrats were much more likely to actually stop and talk to the poller than Republicans were. This throws a monkey wrench into the entire stochastic response assumption and requires more training of polling personnel. Something most polling agencies, especially media, do not adjust for.

The net result of these changes is that accurate polling requires a good mathematical understanding of the underlying voter strata. Many of the polling organizations just plain don't do all their homework and allow bias and heteroscedasticity errors to unduly influence their results.

This is not to suggest that polling now is less accurate than it was in the past. It wasn't that accurate in the past. Witness Dewey-Truman. The difference now is we pay more attention to it and lots more groups conduct polls. However, many of those groups do a poor job of it.

Sorry post was so long, but it is hard to condense statistics terms into English without confusing the hell out of everyone.

Southern Male's picture
Posted by Southern Male on 9 January 2006 - 2:24am
Regarding polls
The truth is that I've been a poll studier for quite awhile now & AP=Ipsos is the least reliable of all polls because (a) they vastly oversample Democrats each time out and (b) the questions they ask are designed to elicit the desired response.

First of all, rejecting a poll because you don't agree with the results is rather silly. A lot has changed since election day, or, um, don't you look at the polls?

Second, if you think only some polls pay attention to how they pose the questions and how that may elecit certain results, then I'd like to know where you got your rose-colored glasses.

I don't trust any polls beyond a certain point, really. But they are news, and they do seem to affect the political climate.

media girl's picture
Posted by media girl on 9 January 2006 - 2:49am