“It is not light that we need, but fire; it is not the gentle shower, but thunder; we need the storm, the whirlwind, and the earthquake.” – Frederick Douglass
The storm has come and gone. The damage is widespread but also fairly shallow. By that I mean that a lot of people are bearing a little bit of burden each – mostly power outages. The price tag will probably be pretty high, not because Sandy was a severe storm but more because Sandy hit in a very expensive area – the northeastern coastline. Most damage appears to have been due to the storm surge, with an honorable mention going to the rain. The wind took down some trees to be sure, but only because the rain had turned the grounds (where the roots hold tight) into mush. Our prayers are certainly out there for those who have suffered injury and loss of life, which are thankfully few given that the storm hit a densely populated region.
The damage is also measured politically. In the old days before going to war generals and armies would scream and cry out to any gods they cared to and asked for signs as to how the battle would go. A bird flying the right way, or a particular pattern of clouds at sunset, or who knows what. Omens were big with these guys. As omens go the storm was awful for president Obama. Just last night on my deck and empty chair was toppled over by the wind … a clear sign that this empty chair presidency will end soon. [Heyooo!!! these are the jokes, people.]
I’m not at work today – we’ve been closed due to a state-of-emergency here in Maryland (and, that state-of-emergency appears to be that it was windy … yesterday). So, as I sit here on the couch, it seems an appropriate time to pick up with a discussion of polling trends and various reports on interesting statistics in the race.
Poll Trends …
Every time I see campaign managers on TV they’re always talking about the polling trend instead of the absolute numbers. The reason is simple: polls tacitly have unknown “biases” due to model errors, such as a voter turnout model (more on this later) – but taking the difference between the same poll (with the same biases) eliminates the bias and gives a read on which candidate is moving up or down.
Now, when we discuss trends it is useful to understand the variability of the estimate. If a poll has a margin-of-error (MOE) of 3 points (plus or minus), then the trend in the poll will have a margin of error of 4.2 / T points, where “T” is the time difference between when the two polls were taken. (The 4.2 is simply sqrt(2)*3.) With this we have counter-balancing problems. If we choose polls that are near each other in time then we have very large uncertainties. But, if we choose polls that are far from each other in time we are trying to estimate a linear trend over a period of time where the voter preferences may be not even close to linear.
We pulled out polls from the RCP averages and the results were impressively underwhelming. Below I give the polls and the trends taking the most recent poll and the oldest poll that tracked after the end of the “Romney Surge” (on or about 10 October). The trend is given in “points per day” that the race is trending toward the given candidate (MOE is given in parenthesis).
- Gallup, Romney +0.176 (0.166)
- Rasmussen, Romney + 0.125 (0.265)
- ABC/Washington Post, Romney +0.2 (0.33)
- Pew Research, Obama +0.19 (0.229)
- IBD/TIPP, Romney +0.188 (0.309)
- Politico, Obama +0.07 (0.3131)
- Monmoth, Romney +0.182 (0.334)
Now, if you look closely you’ll see that all of the trends are within the margin of error of zero, with the exception of Gallup who puts Romney at trending just ahead of the margin of error. If we perform a simple maximum likelihood estimate using these seven data points we get an aggregate of Romney +0.085 points per day, with a MOE of 0.096. So, the trend is Romney, barley, and within the MOE.
What does this mean? Well, first it means that I like statistics and will dig into them if I have time (like a day off). Second, it means that the race has stabilized according to the polls (as the RCP average also shows). If it is trending then it is trending ever so slightly Romney.
Turnout Demographics …
There are two main reasons polls have inherent biases (and thus why trends interest us). The first is the assumption of voter turnout, and the second is the ability accurately sample the population. These are really the same issue but they take on different forms. In general, one simply wants to take a “representative sample” of the population, and by that we mean the population actual voters. If we can do that then the uncertainty in the estimate of the population behavior will be around 0.5/sqrt(N), where N is the sample size. If one wants a 2 point uncertainty (1 sigma) then you simply need 625 respondents – not so bad. But getting a representative sample can be difficult for the two reasons we point to above. The way in which I place phone calls to do a survey may be biased, and I may call quite a few people who won’t actually vote – even if they tell me they will.
In 2008 president Obama won big, largely because turnout was in his favor. What will happen this time? Or, more specifically, what will the turnout demographics be this time? The answer depends a lot on definitions (and is why Gallup & Rasmussen have a significantly different polling result than does IBD [Investors Business Daily]).
Gallup recently released results of a survey (with a massive sample size of 9,000 likely voters) that claims theracial/social demographics will be much the same this time around. Similar numbers of minorities, youth, middle-class, you-name-it. But the turnout will not be the same in term of party identification. The 2008 election saw party identification as 39% Democrat, 29% Republican, and 31% Independent. The 2012 prediction from Gallup, based on the same demographics, is 35% Democrat, 36% Republican, and 29% Independent. That’s downright amazing. A 10-point Democrat edge in 2008 has become a 1-point Republican edge in party identification. When asked which way the independents leaned, 2012 saw 54-42 edge for Democrats turn into a 49-46 edge for Republicans, a 15-point swing. Yikes.
If the general trend really is 10-to-15 points away from Democrats toward Republicans then I think the Democrat turn-out-the-vote machine is going to have to run fast & hard to keep this thing from being a blowout. (And perhaps they will.)
Registered Voters, Likely Voters, and …
There is another way to crack down that “representatives sample” problem. Polls tend to work to differentiate between “Registered Voters” – which are fairly easy to determine – and “Likely Voters” – which matter more but are tricky to figure out. There is another class of voters that matters these days though – Confirmed Voters. That is, people who get called in a telephone survey and have already voted. How are those guys breaking? (Breitbart pulls the numbers down here.)
In 2008 Obama led in early voters by 55-40. In 2012 he is trailing Romney by 52-45. Now, early voters aren’t exactly a “representative sample” of the entire population, but a shift of 22 points away from Obama means something.