There is a significant mathematical methodology error in Pollster.com's simplistic approach to "averaging" recent polls. It is so bad and unprofessional, it cannot go without comment since the lesson we might learn may serve us well in the coming days and the next election.

I won't go into the details here, but suffiice it to say their simplistic approach to averaging is essentially misleading. Under certain mathematical conditions, averaging can be mathematically sound way of coming up with more accurate estimates.

One condition under which averaging accomplishes this goal is when the samples essentially are of the same type, and in particular drawn from the same population. This is not a necessary condition, but the problem gets much harder when it does not apply. On the other hand, using samples from different populations is one of those "red flag" situations to those skilled in the mathematics of the problem in which one must exercise some care in doing.

Another consideration is sample size, and those skilled in the art always factor in sample sizes when combining multiple estimates to attempt to produce a more accurate estimate. The pollster.com folks demonstrate no such awareness or care of either issues in their presentation of the data.

If one looks carefully at the list of pollster.com results, they include without an indication of differentiation in their averaging process, polls which are samples of likely voters ("LV") and registered voters ("RV"). These are very different populations and there is little that is mathematically sound in simply averaging them and arguing this gives a more accurate estimate.

Nonetheless, they blithely include the latest Mason-Dixon results which are both relatively small samples and of random voters, with more numerous results of bigger samples of likely voters. They then give uncritical play to the stories of the last few days of a press that sensationalizes the horse race aspect of an election for their own purposes. Aside from the fact this approach by the press exerts it's own influence on public opinion, the uncritical presentation by pollster.com of the data and this last raft of "the numbers are tightening" news stories is at best unprofessional and misleading.

What we need to get an accurate picture of trends and an estimate of the vote count, but won't get because pollsters quit doing such samples on the weekend before an election, is a poll of a large sample of likely voters one day before the election. We also need to know how many of those voters have already cast their ballots using some type of early voting mechanism and when they actually voted so we can correct for the date. We should see credible political scientists and pollsters arguing right now about why we don't get quality polling before the election, and how we can use exit polling to complete the polling picture, so we have a credible public audit on the results. But we don't because they have their own agendas, as does the news media that makes ad revenues by hyping the horse race.

On this blog, I have from the beginning of this election season remained skeptical we will take either the Senate or the House. A big reason is that we have a lot of people in the political science/pollster community and the news media who, for their own careerist reasons, have been faithless with the electorate in terms of communicating accurate and meaningful information about the candidates and the actual sentiments of the electorate, as well as about the numerous flaws with electronic voting and tabulating systems (at least until this last week anyway) and VBM.

Those are the main reasons I have insisted on the mathematical merits from the get-go that well run exit polls whose raw data is available to all are essential audit checks on this election. It's more than a little significant that the media has chosen to embargo exit poll results until 5:00PM EST so they won't be caught off-guard by possibly meaningful disparities between reported vote counts and the exit polling, and professional pollsters who demonstrate a certain lack of mathematical sophistication like the pollster.com guys and "paul" have just been vehement in attacking exit polling without providing any substantive technical arguments, or even particularly skilled logical arguments, against it.

pollster.com is a great site for raw data. One can play around with their unsophisticated and misleading approach of averages if one likes, but one should not attempt to draw any conclusions, particularly about trends, because the averaging results can so easily be misleading.

<hr/>