The P.U.M.A. Factor

The results of the polls for President are all over the map. If the polls are supposed to be correct to a 95% degree of certainly give or take 2 or 3 percent, how can the polls be that much different for the same candidates when taken at the same time? The simple answer is that they should not be that far apart. If you take a look at the polls making up the average at www.realclearpolitics.com on October 23rd, you will see a range from Senator Obama being up over McCain from 1 to 14 percent. Now there is such a thing called an outlier poll. That is where that other 5% comes into play and statistics says that the result could be outside of that normal sampling error of 2 to 3 percent. So let’s say that happened with both the 1% and 14% results and throw those away. That still leaves a range of 2 to 11% and that is way outside the margin of error on both polls. How come?

Take a look at the polling results for the 2004 election. Here you do not see the wild differences between the various polling firms. In the same time period in October of 2004, the range was from a 1 to 6 point lead for Bush. That is a five point spread and within the margin of error. That is a big difference between the current 13 point spread that is way outside the margin of error. To have this kind of spread indicates that something unusual is happening that is causing problems with the adjustments polling companies have to make to get the sample of voters to properly represent all of the voters. No random sample of voters perfectly represents all voters and polling firms have to weight their results to force them into being representative of all voters.

I actually supply polling samples for a polling company and hence I know for a fact that the sample has to be 10 times greater than the actual number of households that will be successfully polled. It is not that only one household in ten will actually take the poll but what if they do not answer the phone? Then because of the weighting required, you may need more of a particular subgroup than for others. The group that has come up short could be because of gender, party affiliation, age, urban versus rural, etc. The percentages of these demographic groups are quite well known and the polling sample must reflect the correct percentage of these various groups. Still polling firms are used to having to make all of these various adjustments and even know how to adjust for differences between people who will take the poll and those who will not by comparing poll results to actual election results. That being the case, what is the explanation for this huge variation of polling results in this election?

The only answer that seems to fit is that suddenly in this election, these adjustments are being made quite differently by the various polling firms. If their weighting algorithms are not correct, this would skew the results. Still, major national polling firms have a lot of experience and they have learned how to fine tune these algorithms pretty well. What is far more likely is that the sample simply does not accurately represent the voters in the area being polled as a whole due to some new factor. Obviously, when the polling companies look at the results from other polling firms and they see results different from their own outside the margin of error; they know something is amiss just like I do. This has caused them to try and tweak their weighting algorithms during the election season. Departing from tried and true weighting algorithms is a risky thing to do, but what choice do they have?

The fact of the matter is that instead of properly correcting for this unknown factor, they have only made things worse. I know of nothing else that could cause such wildly different results. This problem in turn has resulted in totally different headlines about what is happening in this election. Some say that the race is tightening considerably and others saying that the gap is widening to the point of being a landslide. The Drudge Report for October 22nd showed both of these claims. If you believe the Zogby results, then we are heading for a blowout. If you believe the Associated Press result, we are headed for another squeaker election. Obviously, they both cannot be right so which one is correct? That is precisely why the folks at www.realclearpolitics.com prefer to average all the polls hoping that the various sampling errors will balance each other out.

If you look at the graph of the RCP averages, you will clearly see that Obama is slowly but surely pulling away. At first glance, it would appear the Zogby poll has it right and the Associated Press poll has it wrong. If you look at the previous Zogby poll taken two weeks earlier, you see Obama up by 5. You cannot blame Zogby for looking at this 7 point jump and predicting a landslide of Reagan proportions. Likewise Pew Research shows a jump from 7 to 14 points. Their weighting algorithms show a blowout in the making. The Associated Press poll shows a change from a 7 point lead to just 1 point. If the Associated Press was alone in showing this, we could say their weighting algorithm is just plain wrong, but they are not alone. The CNN poll shows a drop from 8 to 5 points. The Democracy Corps, a democratic polling firm, shows a drop from 9 to 4 points. Surely no one would accuse the polling firms doing the polls for these three outlets of being biased towards John McCain.

This makes even the RCP averages suspect because they are using all these contradictory results. If the tightening group is correct, then the averages are wrong to show a widening gap. The increasing gap in the RCP averages is caused by more polling firms using a weighting algorithm resulting in a widening gap than a narrowing one. The same thing is happening in the battleground states where the RCP averages show that McCain is in a world of hurt. He has to hold the red states and Colorado, Iowa, New Mexico, Ohio and Virginia are showing up as blue in the RCP electoral map and none of the blue states are even toss up states in this election. If that is true, there is no way McCain can possibly win in the Electoral College. But what if the firms using a narrowing gap weighting algorithm are correct? Then we have a whole new ballgame.

Meantime, back at the McCain camp, their strategy makes no sense. They have practically conceded Iowa, Colorado and New Mexico and are still pursuing some blue states. They are not overly worried about Ohio and Virginia either. Neither do they seem very worried about all of those toss up states. So what is going on here? Clearly, the McCain campaign thinks the narrowing algorithms are correct, but isn't that just wishful thinking? They pulled out of Michigan when the Democratic Rossman Group/MIRS poll had them down by only 5 points. Yet they persist in Pennsylvania when the RCP average has them down by 10.5 points. No one has them closer than 8 points right now and no one has had them closer than a tie in the last six months. They sent Palin to New Hampshire where they are behind from 8 to 13 points. Why would they do this? They cannot be that stupid unless they think they are on to something.

Remember that the McCain camp has their own polling firm and are running their own private polls. The only answer that makes sense is that the McCain folks are convinced that their own polling firm has figured out what factor is causing all those wildly differing results from the other polling firms. Whatever this factor is, it is something that the tried and true weighting algorithms are not handling properly. In a previous article, I speculated that it could be that the Bradley Factor is alive and well in this election. Since this is the first presidential election with a black candidate, the weighting algorithms are just not equipped to handle the race factor. The fact that the race card has been repeatedly played in this election could be causing the Bradley Factor to be much more prominent than it ordinarily would be.

Yet Gallup Polling claims that the Bradley Factor is a wash at best and could actually be adding an extra 3 points to Obama's total in a kind of reverse Bradley Effect. Looking at where the McCain folks are competing when they would seem to have no chance, they all have one thing in common. They were carried by Hillary Clinton and in some cases even after it was obvious that Obama had it locked up. McCain was bound to pick up some of Hillary's supporters anyway after the way Hillary was treated by the Obama campaign. These are the so-called PUMA (Party Unity My Ass) folks. Clearly the McCain folks think that there is a P.U.M.A. Factor in play which is a variation of the Bradley Factor. The results that Hillary got simply cannot be explained by race alone. It was the positions that she took that were different than Obama's that resonated with these voters. Because Obama has such an extremely liberal voting record, there were some issues where even Hillary agreed with McCain more than she did with Obama.

This effect was heightened by the selection of Sarah Palin which gave these folks an even better reason to vote for McCain and Palin. These supporters are primarily female and given the political differences between Sarah and Hillary, one would think that the PUMA people would be few and far between. What is missing in all of this is the males who voted for Hillary over Obama. They are far less inclined to be upset about how Hillary was treated. On the other hand, Sarah Palin is more their kind of woman than even Hillary is. They are blue collar Democrats who are into hunting and fishing just like Sarah is. They were not buying what Obama was offering then and they still feel the same way.

The key point here is that these people are actually Democrats who plan on voting for McCain. Because such people are being called racists for this by people like John Murtha of Pennsylvania, they are not telling the pollsters that they are planning to vote for McCain. Most of the pollsters are weighting their samples based on the usual voting splits between Democrats and Republicans. If the P.U.M.A. Factor is upsetting this voting pattern, the pollsters are giving too much weight to the Democrats in their samples. If the polling firm for the McCain campaign has figured out how to adjust for the P.U.M.A. Factor, then they are getting radically different results and that is especially so in some blue states that Hillary carried by wide margins. This same adjustment has caused them to realize which red states are not good bets any more and which blue states are up for grabs in spite of what other pollsters are showing.

This conclusion is reinforced by the fact that the McCain folks are not alone in thinking that Pennsylvania is not a lost cause for McCain. This list includes Ed Rendell, the Democratic governor of Pennsylvania. Todd Beeton at MyDD has an article making his very point. The difference is that while an Obama internal poll may show him only ahead by two points, it might be even closer if the McCain polling firm has correctly identified how to adjust their weighting algorithm for the P.U.M.A. Factor. The bottom line is that we haven't heard the fat lady (oops, pleasingly plump lady) sing yet. Perhaps Senator Obama is wee bit premature in planning his victory celebration and putting his transition team together.