Are the initial polls 100 correct?

Is there a "red shift" in the US elections - a significant shift in Republican votes between elections and voting? Does this mean fraud?

There is something called a "redshift," but it is caused by unreliable, blue-leaning baseline surveys, not fraud that causes actual results to red-shift.

If you had asked if Surveys in general have a democratic bias compared to the results, you would get a very different answer. However, since you specifically asked for quit polls, this is all about them.

This article, published just before the 2008 election, explains 10 reasons why exit polls are not reliable. You like be informative , but you cannot draw statistical conclusions from it, even for something as basic as who won. I will quote sections that deal specifically with the subject of "redshift".

1. Exit surveys have a much greater intrinsic error rate than regular surveys.

This is due to so-called cluster sampling techniques. Exit surveys are not carried out in all districts, only a fraction of them. Although these districts are randomly selected and intended to reflect their condition as a whole, this provides another possibility for errors to occur (e.g. that a particular district was particularly heavily promoted by one of the campaigns). As a result, the error rates are between 50 and 90% higher than in comparable telephone surveys.

It is a fact of life that almost no media coverage of survey results ever the margins of error discussed. But this is where almost all of the turnover in the surveys comes into play. If one poll predicts that a candidate will get 47% of the vote +/- 1%, and another predicts that the candidate will get 52% of the vote +/- 5%, then when the second poll is published, it will be used throughout Media coverage reports how much the candidate's position has improved, even though the error rates are so high that the candidate (without further information) actually gets closer to 48% of the vote (and therefore likely to lose). Since exit surveys are still higher The rates of error than normal surveys are their numbers all the moreless reliable.

Error bars in statistical results are very important. I can say now, over a year before the next presidential election, that the Republican candidate will get 51% of the vote, +/- 49%. If the media covered my prediction, they would say that I said the unnamed Republican was ahead ... but I really said that it was way too unsafe to say anything at all (other than that Will the Republican) got votes because my floor was 2% overall).

2. Exit polls have repeatedly overestimated the democratic share of the vote.

Many of you will remember that event in 2004 when leaked polls indicated that John Kerry was going to have a much better day than he actually had. However, this phenomenon was hardly unique for 2004. In 2000, for example, surveys on the exit of Al Gore states such as Alabama and Georgia (!) Had won. If you go back and look at The War Room, you'll find that George Stephanopolous and James Carville enjoy polls that show that Bill Clinton won states like Indiana and Texas which, of course, he didn't.

This builds on the first point. Exit polls are unreliable and (for the reasons discussed below) slim democratic. So if the unreliable numbers are skewed one way, they seem actual Results compared to being skewed in the other direction. So what really happens is not that the voices suddenly turn redder than they should, but the initial polls are bluer than they should be .

4. Exit surveys question the definition of a random sample.

Although the baseline polls theoretically established methods of collecting a random sample - essentially where the interviewer approaches every nth person who leaves the polling station - in practice this is difficult to do at a busy polling station, especially when the pollster may be many People has feet away from the polling station itself due to election campaign laws.

5. Democrats may be more likely to participate in polls on exit.

Regarding items 1 and 4 above, Scott Rasmussen noted that Democratic supporters are more likely to agree to take polls on the exit, probably because they are more enthusiastic about the election.

These two mostly stand on their own, but together they somehow explain why exit polls are democratic. In a busy polling station, where there is even an even mix of people, if someone becomes more approachable to the polling station, the more likely they will be interviewed. If more Democrats than Republicans are ready to vote (the link to the evidence of this in the article is dead and the reasons are not discussed) the results will obviously be Democratically skewed.

6. Exit polls may have problems calibrating the results of an early reconciliation.

Contrary to common wisdom, polls try to include people who voted in most (if not all) states prior to election day using a random telephone sample of such voters. However, this requires the polling stations to guess the ratio of early voters to regular voters, and sometimes they don't guess correctly. In Florida in 2000 [absentee votes were essentially Republican], which led to an overestimation of Al Gore's vote and contributed to the state's infamous false call.

7. Even late voters can miss out on exit polls.

By “late” voters, I mean people who come to their polling station in the last hours of the day after the initial elections have not been held. ... this adds another possibility where the sample may not be random, especially in areas with long lines or longer voting hours.

Therefore, both people who voted before election day and people who are about to graduate are not necessarily well represented. I couldn't find any statistics on how these groups overall vote, but they are certainly enough to add even more errors to the baseline survey results (even if they even out from year to year).

In short, any claim based on the difference between the exit poll results and the actual results of an election only shows how unreliable the exit polls are.

The article I quoted ends with a good summary:

10. You will know the actual results soon enough anyway.

Be patient, my friends, and be in luck: in France it is illegal to conduct any poll of any kind within 48 hours of the election. But exit surveys really are more trouble than they're worth, at least as a forecasting tool. An independent panel set up by CNN after the Florida disaster in 2000 recommended that the network completely ignore exit polls when calling certain states. I suggest you do the same.


It would be interesting to know if the same shift is taking place in other countries and in the same "direction" ...


Just as a hint: +/- error means nothing at all, unless you state what error it is. Standard deviation? Standard bug? Offer? Confidence interval? That makes a HUGE difference.


@nico - I've used it in its simplest form (range) as a simple English proxy for the more statistically stable information that actual polls have with their data (which the media then ignore).


@ Bobson sure mine was more of a general comment :)

Christian Sauer

Just as a comment: in Germany almost all polls before the elections show almost no right-wing extremist parties, but they tend to get a lot more votes in the current poll. One explanation might be that "voting rights" are frowned upon and most people don't say they do - but in the dressing room they do. Exit surveys are not legal so I can't comment on them.