Thursday, November 01, 2012

Homers




Heard about the polls?  Sure you have.  Even if you have no interest in the polls, even if you hate them, the news has been full of reports from all sorts of agencies and groups, some of whom you may never had heard of, and some you frankly should ignore.  This is because the news depends on viewers and readers for their revenue, and a tight election – in theory – keeps their attention.  Polls serve the narrative, because you can’t tell if the poll is right or wrong until after the election.  That’s not to say the polling agencies are dishonest, for the most part at least, because polling groups are businesses and need to be reasonably accurate to be able to attract clients.  But polls are not all the same, and even where they share common ground, sometimes you would be wise to be skeptical.

Let’s start with the obvious.  If I put out a poll on the election and asked, say 600 men but only 400 women their opinions, you’d reject my findings as out of balance, because we all know even without checking the history that men and women vote in roughly equal numbers.  So a poll that did not balance men and women would not be reasonable.  Similarly, I could hardly call my poll reasonable for a national election if 80% of the people contacted were in, say, California and New York.  Or if my respondents were all white, or had some mix that was not in-line with census numbers for race. 

You get the idea.  Any poll which is not a reasonable sample of the actual public is not a valid poll.

With that in mind, the fact that a majority of state polls and a fair number of national polls have weighted political affiliation rather heavily in Obama’s favor is odd, to say the least.  Now yes, it is true that in 2008 Democrats represented 37 percent of the voters, seven full points more than the dispirited Republicans of that year.  So, some people will say the polls are simply reflecting the last Presidential election.  However, the polls in 2008 did not weight party affiliation according to the 2004 turnout, nor did the polls in 2004 use 2000 as a baseline for their weighting.  What’s more, the 2010 mid-term elections demonstrated a rather drastic shift in party participation, as Republicans showed up in large numbers and with great energy.  Ignoring that election seems to be, well, short-sighted.  Let’s also not forget the novelty of 2008.  Obama ran as the first black candidate from a major political party, surely worth political capital in 2008 but hardly the same impact now.  The economy is also a problem for Obama, and while his supporters may stand by him in the main, it’s frankly inconceivable that he would gain support on that point.  But what’s even stranger, is that some of these polls are handing Obama an ever greater advantage in party participation than he enjoyed in 2008.  There is absolutely no empirical basis whatsoever for such a weighting.  It’s misleading, it’s deliberate, and frankly the polls doing so risk humiliation if they – as seems likely – are proven wrong.

The question here is just why they would do something so, well, dumb.  The effect of the weighting is obvious; it’s not hard to expect that Democrats will largely vote for Obama while Republicans will largely vote for Romney, so fudging the weights will obviously change the outcome.  The reason it’s dumb is two-fold – first, it’s not hard to observe that if Romney is leading among Independents in a battleground state (which by definition is not automatically Democrat or Republican in the main), then saying Obama is winning is just plain dishonest, an obvious lie, and second, when the election results are known, the lie will not only be obvious, it will be a matter of record. 

Some folks, like our friends over at Pew Research, like to claim that there’s no reason to use party affiliation weighting.  Their explanation is that however folks respond is a fair representation of voter enthusiasm.  Certainly that helps explain the strong pro-Democrat weighting in 2008.  But there’s no evidence that’s always the case, and anyway, we’d laugh at Pew if they ignored other demographic weights on the basis that interest in answering a poll reflected accurate representation in the voter pool.  It’s lazy and intellectually dishonest to pretend so.  Also, Walter Mitofsky, who pioneered political polling as we know it, candidly admitted that Democrats are “chronically” over-sampled in polls.  He suggested it might have to do with the personality of Democrats versus Republicans, but in any case the bias is documented, and every responsible polling group knows it.

Yet, yet, yet.  I still don’t buy into conspiracy theories, not least because it makes no sense for a polling group to pretend someone is winning when they know he is losing.  Especially if doing so produces no financial benefit for the company, and instead would damage their reputation for accuracy and integrity.  And make no mistake, if Romney wins by a big margin, as could very well happen, then those polls which went out of their way to make Obama look stronger than he is, are in for a hard time of it.  They won’t be able to make the excuse that no one saw it coming, or that their hands are clean.  So why be stupid like this?

They’re homers.  That’s why.

I don’t mean the polling groups want Barack Obama to win, exactly.  Rather, these groups are engaging in an unconscious groupthink, based in their own demographics.  Back in 2008, I pointed out a reason for liberal bias in polling.  It has to do with their home ground.  Below, is a list of major polling groups from the 2008 election and their headquarters addresses:

ABC News 77 W 66th St, #13, New York City, New York
CBS News 524 W 57th St, New York City, New York
FOX News 1211 Avenue of the Americas, New York City, New York
Gallup 901 F St NW, Washington DC
Hotline 88 Pine St, 32nd floor, New York City, New York
IBD 12655 Beatrice St. Los Angeles, California
The Los Angeles Times 202 W 1st St, Los Angeles California
Marist Institute 3399 North Rd, Poughkeepsie, New York
Mason-Dixon 1250 Connecticut Ave #200, Washington DC
Newsweek 251 W 57th St, New York City, New York
The New York Times 1 City Hall, New York City, New York
Pew Research Center 1615 L St NW, #700, Washington DC
Quinnipiac 275 Mount Carmel Ave., Hamden Connecticut
Rasmussen 625 Cookman, #2, Asbury Park, New Jersey
Reuters 3 Times Square, New York City, New York
Survey USA 15 Bloomfield Ave., Verona New Jersey
TIPP 690 Kinderkamack Rd, Oradell, New Jersey
Washington Post 1150 15th St NW, Washington DC
Zogby 901 Broad St, Utica, New York
 


See the trend?

Two Left Coast outfits, and the rest well into Blue-State territory.  They live, work, and operate in an environment where balance and critical thinking, at least in political terms, does not exist and is punished when discovered.  Not a single group based in, say, Kansas, Texas, or Wyoming, or in more balanced states like Iowa, Missouri, or Ohio.  The bias is unavoidable, yet these guys never see it, because for them ‘normal’ is what for us means leaning left.

This also explains why many of these polls have not done what they usually do, make a last-week adjustment to get closer to the actual condition: the heavy storm knocked operations for a loop, and they are stuck with their garbage releases.   It’s too late to nudge things back to actual numbers.

So this seems hinky, but it really can be unconscious behavior.  Back when I officiated sports, I noticed how homers think and act, and what struck me early on, was that even the most outrageous homer believed he was reasonable and balanced; he simply did not perceive the other perspective.  Sports fans may not seem like a good model for a media professional, but I ran into the same effect with coaches and school officials.  If you don’t get out and travel, you can end up with a very limited view of what’s really going on. 

The thing that strikes me about polls, and it reinforces this problem, is that the polling is often done by firms hired by these media outlets, but the press release comes from the media guys, and specifically from their head office.  So, it’s very likely to me that these polls are performed professionally at the field level, then massaged at the high level to produce the desired report.  It’s not at all that these people are trying to deceive you.  They just know what results they expect, and make sure the report is in line with those expectations.  Strange as it may sound, they may be just as surprised next Tuesday night as Mr. Axelrod and his crew.


Wednesday, October 31, 2012

What Past Election Results Tell Us to Expect This Time


There’s no shortage of opinions on how the 2012 U.S. Presidential Election will turnout, and a lot of the disagreement  comes from the different landscapes presented by the National and State polls.  Romney supporters are encouraged by clear and substantial leads in the Gallup, Rasmussen, ABC News/Washington Post, and NPR polls.  Obama supporters point to the CBS News/NY Times and Politico polls, but they also point to the state polls, observing published leads for Obama in Iowa, Ohio, Wisconsin, Minnesota, Michigan, Nevada, New Hampshire, New Mexico, and Pennsylvania.  There’s too many states on Obama’s side, they say, for Obama to lose.  The problem is, the math just does not work.  Either most of the national polls are wrong, or most of the state polls are wrong.  And things just get messy from there.

There’s all kinds of anecdotal stories to read and hear, but there’s also a record we can check.  Generally, there is a relationship between how a candidate does nationally and how he does in a given state.  As his national support rises, his support in each state rises to some degree, not at the same rate of course but the better a candidate does nationally, the states have to reflect it.  So we can look at past state results in relation to national results, to give us an idea of what should happen.  To keep it simple, for here I will just address battleground states.

First, to identify the battleground states.  I disagree a bit with RCP, I don’t think anyone can seriously pretend that Arizona, Missouri, or Oregon are in play this year, but I will take the rest:   Colorado, Florida, Iowa, Michigan, Minnesota, Nevada, New Hampshire, North Carolina, Ohio, Pennsylvania, Virginia and Wisconsin.  Here’s how each state performed, by party, relative to the national election results for the last four Presidential elections:

Colorado:  Democrat candidates average 2.81 points lower in Colorado than their national support, while Republican candidates average 2.01 points higher in Colorado than their national support. 

Florida:  Democrat candidates average 0.97 points lower in Florida than their national support, while Republican candidates average 1.61 points higher in Florida than their national support.   Also, Republicans ALWAYS produced slightly higher results in Florida than their national support.

Iowa:  Democrat candidates average 0.80 points higher in Iowa than their national support, while Republican candidates average 0.62 points lower in Iowa than their national support.   Also, Democrats ALWAYS produced slightly higher results in Iowa than their national support.

Michigan:  Democrat candidates average 3.20 points higher in Michigan than their national support, while Republican candidates average 2.90 points lower in Michigan than their national support.   Also, Democrats ALWAYS produced slightly higher results in Michigan than their national support. 

Minnesota:  Democrat candidates average 1.35 points higher in Minnesota than their national support, while Republican candidates average 3.26 points lower in Minnesota than their national support.  

Nevada:  Democrat candidates average 1.45 points lower in Nevada than their national support, while Republican candidates average 0.16 points higher in Nevada than their national support. 

New Hampshire:  Democrat candidates average 0.44 points higher in New Hampshire than their national support, while Republican candidates average 1.02 points lower in New Hampshire than their national support.  

North Carolina:  Democrat candidates average 4.56 points lower in North Carolina than their national support, while Republican candidates average 6.31 points higher in North Carolina than their national support.   Also, Republicans ALWAYS produced slightly higher results in North Carolina than their national support.

Ohio:  Democrat candidates average 1.21 points lower in Ohio than their national support, while Republican candidates average 0.92 points higher in Ohio than their national support.   Also, Republicans ALWAYS produced slightly higher results in Ohio than their national support.

Pennsylvania:  Democrat candidates average 1.60 points higher in Pennsylvania than their national support, while Republican candidates average 1.49 points lower in Pennsylvania than their national support.  

Virginia:  Democrat candidates average 2.76 points lower in Virginia than their national support, while Republican candidates average 3.67 points higher in Virginia than their national support.   Also, Republicans ALWAYS produced slightly higher results in Virginia than their national support.

Wisconsin:  Democrat candidates average 0.95 points higher in Wisconsin than their national support, while Republican candidates average 1.80 points lower in Wisconsin than their national support.  

What this tells us, is the likely result of three different conditions in the national popular vote:

If Barack Obama wins the Popular Vote, he will definitely win Iowa, Michigan, Minnesota, New Hampshire, Pennsylvania, and Wisconsin, and would likely win Nevada and Ohio .  Romney could still claim Colorado, Florida, North Carolina and Virginia but that would not be enough.  Obama would win re-election by 281 EV to 257.

If Romney wins the Popular Vote with 51.5% or more, he could win all of the battleground states but Michigan; that state seems out of reach in any scenario.  That could mean a Romney win by an Electoral  margin of  331 EV to 207.

If the two candidates are essentially tied in the Popular Vote, Obama would claim Iowa, Michigan, Minnesota, New Hampshire, Pennsylvania and Wisconsin, while Romney would claim Colorado, Florida, Nevada, North Carolina, Ohio, and Virginia.  That would produce a 281-257 Electoral win for Romney.

The actual results will be sure to include some changes from the script, of course.  I think, for example, it’s more likely this year that New Hampshire would go to Romney and Nevada would go to Obama than vice versa, but we shall see.  The point is that the last several elections do give us an idea of how things turn out relative to different levels of polls and voting, and that’s something to keep in mind.