If you’re in the Obama campaign–or a numerologist sympathetic to it–you probably console yourself with comparisons to polling results from 2008. Today Barack Obama is running about three points behind where he was at this point in the race two years ago, while Mitt Romney is about four points ahead of John McCain. Since Barack Obama won by 7.3% in 2008, the seven-point swing this year still puts the race within reach.
However this is not 2008. That year was an anomaly in my lifetime: the first election since 1952 when there was neither an incumbent president nor a sitting vice president on the ballot. The 2008 election was a choice; this year is a referendum.
How today’s poll results are usually presented is centered on a largely meaningless statistic. Real Clear Politics and 538 track the spread: the difference between the top-line results for Barack Obama and Mitt Romney. This was a statistic that made sense in 2008 during a choice election.
From the end of the debates until election day 2008, Barack Obama led by about seven points. Last-minute undecideds four-years ago broke uniformly for both sides, meaning that the final difference of 7.3% closely approximated the post-debate difference. Poll aggregators slapped themselves on the back for their prescient predictions that validated their complicated models. But the results this year are likely to surprise. In fact, looking at the chart above there is no similiarity between trends in both years.
The 2012 election returns to the historical norm where you have an incumbent and a challenger. The primary metric is the incumbent’s level of support. He is only safe when he is well above 50%. When his support dips below 50% he is in danger. If it stays below 48% he is in extreme danger. Barack Obama sits at 47% and hasn’t been higher than 48% since the first debate.
The spread is not entirely meaningless, so long as one keeps in mind that it is a “horseshoes” type of metric in referendum elections. All the challenger has to do is to keep it close. Let’s look at a post-Labor Day comparison of 2012′s RCP averages versus 2004, the last time we had a similar election. It is true that Romney trailed Obama all throughout September, but he was well ahead of where challenger John Kerry was four years before. Meanwhile, Barack Obama stayed about a half-point behind George W. Bush’s 2004 numbers. In total, Barack Obama was running about 2.5 points behind Bush’s advantage. Bush ended up winning by 2.4%. The pre-debate conventional wisdom was that Obama’s 3-4 point lead put him in a comfortable position. Based on the 2004 precedent, that confidence was entirely misplaced.
In fact, there is remarkable consistency between the 2004 and 2012 poll results. I’ve broken the post-Labor Day period into four segments. Until the first debate both in 2004 and in 2012 there was a period of stability prior to the first debate.
Then following the first debate there was a switch. The incumbent stumbled and the challenger gained points–probably because for the first time, he is on an equal stage with the President where he can demonstrate his stature. Bush lost about two points in the first debate while Obama lost three. Meanwhile John Kerry and Mitt Romney both gained about three points. The spread, therefore, narrows as a result of the first debate by about five to six points. Because Bush entered the debates with a wider gap than Barack Obama did, that mitigated the effects of the 1st debate switch. Barack Obama didn’t enjoy enough of a pre-debate advantage.
The third period occurs during the later debates. The challenger holds his gains while the incumbent reclaims some of his losses. This year the gains came from Obama’s demonstration that he was not a zombie during the second and third debates. Eight years ago, the incumbent partial recovery came as a result of a John Kerry gaffe when he interjected during the final debate that the vice president’s daughter was a lesbian.
We are now at the end of the third period and if past is prologue, Barack Obama is in deep trouble. That is because after the debates there is no further gain in support for the incumbent, while the challenger climbs two to three points.
Let’s look in greater detail at 2004. Osama Bin Laden released a tape on the 29th of October just four days before election day. There appears to be a sharp spike in John Kerry’s numbers during the last weekend of polling before the election. In spite of that spike, the conventional wisdom is that the effect of the recorded message was to congeal support for the incumbent. One reaches that conclusion by looking at the final result when Bush jumped from 49% in the final RCP average but got nearly 51% at the end. I have a simpler explanation that I’ll share with you later in the post. But let’s look at that spike: was it really a four-day spike as a result of Bin Laden’s ”October surprise” or was it just a continuation of a trend of increasing support for John Kerry that began about 17 October, four days after the last debate? I contend that it was the latter. And since Kerry’s gains came without a corresponding drop in Bush’s support, what we’re observing here is last-minute undecideds breaking for the challenger to the tune of a gain of about two to three points.
But that’s not all. Remember George W. Bush’s inexplicable two-point last minute gain? Where did it come from? I contend that it is as a result of sample bias. Most polls oversample those unlikely to vote. Look at today’s ABC-Washington Post poll of likely voters. “Likely” is defined as those registered voters who classify themselves as “extremely likely to vote,” which is 82% of their sample. Add that to the 4% of the sample that says that they have already voted and this ABC poll indicates a turnout of 86%. America hasn’t seen that level of turnout in generations. More recently, presidential election turnout is about 70%, plus or minus 5 points. High turnout models give disproportionate weight to the unenthused. In 2012 the unenthusiastic voter, even more so than in most years, is a Democrat. This is why when you see polls like Gallup that report different results for registered voters and likely voters or likely voters and extremely likely voters, you tend to a see a three to six point swing in Mitt Romney’s favor when the tighter model is used.
The same thing happened in 2004. Looking at Bush’s post-debate numbers that stayed remarkably stable between 48 and 49 percent, there is absolutely no indication of the beginnings of a last-minute shift. Occam’s Razor applies: polls simply oversampled those unlikely to vote, and that meant that all along, George W. Bush was tracking about two points below where he really was with a realistic voter turnout model. When today’s polls are based on turnout models of over 80% we can expect the same result.
So where does that leave us? Today, Mitt Romney sits at 48% in the RCP average of polls. Add 2 percent because of sample bias and add another 2 or 3 points for breaking last-minute undecideds, and I expect Mitt Romney to finish with between 52 and 53 percent to Barack Obama’s 46 or 47 percent.
But what of the state polls, you ask? Bunk. Most of them–especially the media polls–are crap for reasons I explained four years ago. Good polling is expensive, and that’s why the best state polling there is, is done by the campaigns themselves. You don’t get to read those results for free. Instead of tracking individual states, it is more accurate (and simpler) to take the average national result and compare it with the partisan voting index for each state. If, for example, Mitt Romney gets 52% of the vote, we could expect him to win most states with a PVI of D+2 or less. That would give him every state that John McCain won, plus thirteen more: Colorado, Florida, Indiana, Iowa, Minnesota, Nevada, New Hampshire, New Mexico, North Carolina, Ohio, Pennsylvania, Virginia, and Wisconsin. Sure, Romney might lose a couple from that list (probably New Mexico or Nevada), just as Obama lost less partisan Missouri four years ago. But in 2008 he also scored surprise victories in Indiana and North Carolina. If he is above 52%, I would expect Romney to pick up a state or two from among Maine, Michigan, New Jersey, Oregon, or Washington. If he exceeds 53% nationally, even Connecticut or Illinois might not be out of reach.
The fundamentals of this race have never been in Barack Obama’s favor. For two years he has underperformed George W. Bush. For more than two years he has trailed in enthusiasm by margins of no better than 3:2. For two years he has had trouble breaking above 50% approval–and those are in polls of registered voters or even all adults. That’s not a good place for an incumbent Democrat to be. 2012 was predestined to be a referendum election, unlike the one Obama won in 2008. If that wasn’t clear enough, especially after a disastrous 2010 mid-term result, Democrats have only themselves (and their media partners) to blame for their delusion.
UPDATE: Thanks to Glenn for the link.