Posted by
Rob on Tuesday, October 07, 2008 9:03:45 PM
The
Obamania has affected not just the MSM but the polls as well. Now
Jay Cost may disagree with me, and for the record, I was completely and utterly wrong in 2006 when I predicted a much better showing for Republicans than actual. I also predicted that Bush would invade Iran by summer 2008 at the latest. So I publish these thoughts at my own peril, jinxing the election with my 100% failure rate. Nevertheless, here are my reasons.
(a) The infamous Democrat slant in polling.
As is well known, there are more registered Dems than Republicans. Accordingly, any poll that picks a random set of phone numbers from the telephone directory will be biassed Democratic by some 4 points. However, more Republicans actually vote than Dems. So the pollsters got in the habit of making two categories "registered voters" and "those likely to vote". Presumably the number crunchers at all the big polling places apply this correction to their numbers. Except when they don't want to. And there's been a lot of that in the past year. If the poll doesn't indicate what category is being represented assume that they are, say, in Denver at the DNCC.
(b) The
Dinkins effect (or Bradley effect, or Wilder effect).
In 1989, the democratic candidate for mayor of NYC was David Dinkins, the first black man to run for the job in a heavily democratic city. When pollsters asked New Yorkers their opinion (the only thing they give freely), they favored Dinkins over Guiliani by 18 points. When the election came, however, the spread had narrowed to 2 points. So approximately 8% of the public lied to pollsters. Wikipedia said it was "white voters" who lied, though I don't know how they proved the color of a voter in a secret ballot, but it would seem reasonable. Consequently, NYT and other mouthpieces of the cultural left have already prejudged the November election, claiming that if Obama doesn't win, it will be racism that is to blame. The inference is clear: white guilt will cling to you like an Albatross if you don't vote Obama. Under those circumstances, who is going to tell a pollster that they are racist? Given the 3 months of nonstop racism training, I would imagine that the 8% might be on the low end of those who lie to pollsters.
(c) The Recursion effect.
I had a fellow graduate school friend who was notoriously late for every meeting. He would set his watch ahead five minutes to prevent this, but then found himself compensating so he would set it ahead again. Consequently he would get to the point where every clock in his apartment had a different time on it, and he had no idea what time it was. (Obviously, this was before the era of cell phones and the diktat of GPS millisecond clocks.) But this is precisely what is going on with the polls. As soon as the public wises up to the slant in the polls, the pollsters add more slant. They know they are doing it, yet they end up believing their own slant, or even worse, unable to know what to believe at all. If you doubt me, look at the spread in polls, say, on
realclearpolitics.com.
This is going to get technical, so bear with me. At the above mentioned site are 10 or so polls conducted in the last week or so. Each one tells you how many people were polled. They try to do this "scientifically" to take out bias beforehand, rather than after the fact. You could do it after the fact, but then you need to poll more people, so it is cheaper to set up your poll properly. (Which a lot of them don't do.) So if the poll is unbiassed, and if N is the number of people polled, then the error in the poll is the squareroot (N) / N. Suppose N=100, then the error is 10/100 = 10%. Not too helpful if the candidates are running neck-and-neck. Suppose we want to get the error down to 1%, then we would need to poll...
(okay class work this out before I give the answer)
Yup, the answer is 10,000 people. But that's 100 times more expensive! Precisely, that's why nobody tries to poll at a 1% level. Usually they settle for about 4%, or about 625 people. So by looking at the various N, we can get a general idea of how accurate these polls are
in a best case scenario. That is, if our scientific polling system really did remove all the systematic bias in the system. There's no telling what the error is if we actually wanted to cheat. But if cheating is somewhat random, we would be able to spot it by noticing how far away this poll was from the others. Let's see if we can spot that. Here's the data from realclearpolitics where RV=registered voters, LV=likely voters.The average has Obama +5.5% lead.
As expected, the RV poll gave more weight to Obama than any of the LV polls. So rather than try to factor out the 4% weighting, we'll ignore it. However, notice that it is not 4% higher than the LV poll right under it. So already we have a peculiarity which might be due to the LV vote being already biassed more than it admits (or more registered voters are voting than usual etc). Since the amount of bias appears *reduced*, it suggests that we are already seeing evidence of recursion-->biassed biassing.
Taking the best of the bunch, which is the Rasmussen poll with 3000 LV and a 1.8% error, we see that the lowest Rasmussen would predict is +8 - 1.8 = +6.2. This is still above the average +5.5 by +0.7 or nearly 1.5 times the expected error. This could happen, of course, since random things are random, but statisticians give it a probability of a 1:10 miss. The other polls are within their errors. So right away we notice that the poll with the most responders has the largest deviation. Not only is it away from the mean, it is away from the median. But if we throw it out and recalculate the average, the average drops down to where the GW poll is now unlikely too.
Well is there a way we can throw out outliers and recover a decent poll? Not self-consistently. It is a problem Bayesian statistics people have been warning us about for decades. When we "cherry pick" the data, even the metadata, we are introducing conclusions into the assumption of randomness. But my goal in showing these problems is merely to point out that the polls are behaving strangely, and clearly not in a random fashion. When we actually read the questions being used in the polls, we begin to see why they are acting strangely. That's why you and I are not paid to deliver polls to the public, its a minefield of biasses and correcting the corrections.
Nevertheless, it would appear to me that the "real" Obama advantage in LV is under 5%, and noting the RV problem, probably under 4%, and then factoring in the Dinkin effect, closer to -2%. Of course, if I can do this so easily, so can the pollsters, which encourages them to cheat some more..., until no one has any idea what the actual poll means.
And now I have one more reason to not worry.
(d) The God effect
The Bush-Gore election had everything stacked for Gore, yet Bush won by such a razor thin majority that no poll could have predicted it. No two elections would give the same result. In fact, there were more valid votes thrown out than those that decided the election. So how does one calculate this putative effect?
One doesn't. It is detected spiritually. When Haman came back to his house to tell his wife of the shameful
farce that had forced him to lead his archenemy Mordecai around the
city,
his wife knew immediately what that portent meant. Likewise, during the Bush-Gore election of 2000, it felt like a great cloud (the Clinton wilderness wandering years) was lifting from America, said my wife. The dark empire could see the victory slipping through their fingers and were powerless to stop it. One senses these momentous events in the little details: the sense of despair in the commentators voice, the whispered hope in the little man, the number of prayer candles being lit, the look in the eyes of the anchorman. Many people have commented on the reaction of the news anchors to Palin's acceptance speech in Minneapolis. They all agreed, it looked as if someone had killed their puppy. The portents have not been in Obama's favor.
There is no doubt that God's judgement will come to USA for its many misdeeds, but His mercy is great. And undeserved.