Polls, Polls, Polls - momentum shifts vs. transitory reversals

[troublemaker]

This is a follow-up on a (widely misunderstood) post from yesterday that asked some questions about where we are. The key point to understand about polls is that they are NOT predictions. They are snapshots of how people would have voted on a certain day. So election results don't show which polls were "right" and "wrong."

It became obvious yesterday that something bad happened last Thursday-Friday and that if the election were held last Thursday we would have been in a difficult environment. National Congressional Generic polls that ended Wednesday were all very good, showing a profound Democratic wave. But all polls that included the end of last week showed a dampening effect so strong that it could have represented a shift in momentum.

History shows that late momentum moves enlarge and carry through to election day. An alarming number of people make up their minds in the last weekend. For instance, if an incumbent has a five point lead in every poll for months then the very last poll shows a dead heat the challenger will tend to win big. Carter and Reagan were tied in 1980 but the late momentum was to Reagan and it carried through in a landslide. In 1994 many Republicans appeared to be tied late but won by 8%. Humphrey made a late move in 1968. Ford made a late move in 1976. All those late moves were shown as little momentum shifts in late polling that magnified on election day.

So, in very late polls the DIRECTION of the trend is as important as the absolute numbers. The question I was addressing last night was this: We know there was a big dip late last week. Such a dip could either 1) presage a shift in momentum, or 2) roll off the sample as later polls drop the November 2-3 period. A related question was whether the Republican move was "lumpy", meaning big gains concentrated in safe Republican districts, like the ones Bush has been visiting. (If a national generic move was localized to feverish Republican support in uncontested red districts it would have little effect on the election results.)

We have two more polls today that do not include last Thursday and both show a rebound, suggesting that the Kerry-flap buyer's remorse effect was localized in time and that prevailing anti-incumbent trends have reasserted themselves. This is Good News.

If the election were held last Thursday we would have been lucky to win the House. If the election were held Sunday we would have won the House pretty easily. Th election is tomorrow, of course, and things are looking good. The tightening appears to be a natural late campaign effect without a change in the underlying pro-Democratic trend.

SUMMARY: Polling data from last Thurs-Fri raised the question whether late-week damage was temporary or a momentum shift. That was (an is) a real question. If today's polls had not shown a rebound we would have been in serious trouble. But they do show a rebound. (Fox and CNN polls today look pretty good.) The very last polling sets seem to indicate that 1) The Democratic trend remains in place, and 2) That resurgence of the Republican base is localized in areas we were never going to win in any election. So everything is looking good.

[Beetwasher]

1. Well, After Looking At The Internals Of the Pew and ABC Polls

It was clear to me they were purposely skewed to give an illusion of a Repub resurgence. In the ABC poll in particular they oversampled Repubs, especially compared to the their previous polls.

[Hugin]

11. This is a recurrent problem with ABC polling...

It happened after the 2004 debates as well.

Nothing new.

[Birthmark]

2. I agree.

That is my conclusion as well. Things look good. Dems take House. Senate razor close. That's no different than last week or the week before or the week before that.

[DemocratSinceBirth]

3. Just To Add

These polls have a confidence level of .095 so you might not be capturing what you think you're capturing...

And a lot of assumptions are built into polls...

[speedoo]

4. In general, there was no rethug "surge" in individual Senate races.

For example, Webb has an 8 pt. lead over Allen in a new Virginia poll and Tester, Brown, Menendez and McCaskill appear to be improving as well. Only RI shows rethug improvement, unless you want to count a PA poll which appears to be less than independent.

[troublemaker]

12. Agreed. There was, however, a lumpy surge (super core base types in red areas)

[Mass]

5. There are 5 polls including Thursday.

2 show Dems up in the mid teens, 2 show the Dems up by around 5.

USA Today/Gallup 11/2-5/06 44 51 5 7 D

ABC/Washington Post LV 11/1-4/06 45 51 4 6 D
Pew LV 11/1-4/06 43 47 10 4 D
Newsweek LV 11/2-3/06 38 54 8 16 D
Time LV 11/1-3/06 40 55 4 15 D

It seems difficult from these polls to see more than some aberrations in the sampling. which seems confirmed by today's polls.

[DemocratSinceBirth]

8. And Faux's Poll

.
http://www.foxnews.com/story/0,2933,227707,00.html

[jackbourassa]

6. This is nonsense...

There was no decline and there is no rebound.

The reason the Pew Poll had the Democratic lead shrink to 4% was because they overcounted Republicans by more than 10% above their turnout numbers in 2004. The ABC/Washington Poll post overcounted Republicans by 6% above their turnout numbers in 2004.

The Time Poll, which showed Democrats ahead by 15%, was taken on the exact same days as the Pew poll. So why the discrepency?

If you look at the internals...the numbers are actually very similar. The difference are the weighted averages of Republicans in the polls versus the Democrats. In the Pew and ABC polls, they were DISPROPORTIONATELY tilted towards the Republicans - the CNN poll was disproportionately tilted towards the Democrats today. The truth is probably in the middle...around 12% (which is what I have been predicting all year and is close to the net averages of all the polls: +11.3% for the Democrats).

For there to have been a swing of 16% in two or three days would mean that about 24 million Americans decided on Wednesday to Friday that they were going to switch their votes from Democrat to Republican. Then over the weekend, about 40 million people decided that they were going to switch their votes from Republican to Democrat.

That just didn't happen.

[longship]

7. Pollsters tweaking their models.

The sole reason for this so-called Repuke uptick is the pollsters tweaking their analysis models. It's like pollsters are saying, Republicans always finish strong so we have to make them finish strong this time, too.

But there is *zero* evidence other than historical that this is going to happen this time. With the GOP in such dire straits, it could very well be that it won't happen this time.

Now here's the important point. In absense of any actual and current supporting data, the pollsters have changed their models to make the election look more competitive. Why they are doing this in the face of all the global indicators showing a Democratic Party tsunami is beyond me (and many other analysts). The bottom line:

Nothing has changed since last week except the pollster models.

Ignore all recent polls

Vote!!!!

[AndyTiedye]

13. Republicans "Always Finish Strong" Because They STEAL VOTES

[TahitiNut]

9. Pre-election weekend polls are notoriously flawed.

The prevalent reasons for the flaws, some worse than others, are both based in one reason: People Lie. The consequence of lying in the pollster's processes is that all polls are normalized to adjust the sample to match predefined demographics. The primary demographics are (1) party preference and (2) likelihood of voting. It's in the former demographic that we see the greatest inaccuracy when a shift in public opinion confers a sense of 'shame' on many. Fewer and fewer people are willing to be 'loyal fans' (or admit it) of the muscular reich. Thus, the number of self-identified 'Republicans' in the survey is diminished.

When we combine this increased inaccuracy in demographics with the fact that the "Crunch Time" before an election necessitates that pollsters cover shorter periods with a different mix of weekday/weekend calls, where there's a change in the profile of those who respond to their calls, we have a significant overcompensation for the diminution in the proportion of self-identified 'Republicans.'

[april]

10. go to must read little Rickey polls BS

http://www.keystonepolitics.com/Article4761.html