As some of you may know, my business is working with poll data. My dissertation tries to uncover the architecture of public opinion in Northeast Asia by using complex methods to disentangle correlations in large surveys across multiple countries. As a bit of a stats and models geek, it’s easy for me to get lost in the malaise of programming R and missing the forest while taking DNA samples from every tree.
There’s a lot of important information you can get from looking at marginal returns (X% vs. Y%), for instance, in a recent NPR poll of likely voters in “battleground states”, the marginal returns for the presidential races are:
Obama: 46%
McCain: 45%
Others: 5%
Undecided: 2%.
Refused: 2%.
What sort of information can we get from this? Well, we can see that in battleground states, McCain and Obama are in a “statistical dead heat”, meaning that their observed percentage is within the margin of error. What else can we glean from these results?
The second headline is the number of undecideds: 2%. In the states where races are expected to be closest, 96% of voters have made up their minds already- before the conventions, before the VP picks, and more than ten weeks out of the election.
As I would ask my undergraduates, what does this mean? First it means that the candidates are going to spending a lot of money in the next two months to woo a very small percentage of the public. Secondly, it means that most of that money will go to waste trying to bring over voters who have no interest in actually going over. This study, which came out this week, lends even more evidence to suggest that trying to convince people is a bit of a fool’s errand.
There is more evidence in the poll to suggest that these voters are stuck on their current candidate. In q13 and q14, the pollsters ask respondents to evaluate their own chances of swithcing. less than a quarter of non-Obama voters say that there is a small or fair chance that they would switch, and roughly 80% say that there is a “very slight chance” or no chance at all of a switch. These numbers are mirrored by the non-McCain voters with a similar distribution. the small number of undecideds and the intransigence of the comitted seems to suggest that in battleground states there isn’t much ground to battle over (I’d like to see the state by state breakdowns, they’re not published in this report and the n’s would likely be too small to be meaningful). Which is why it is funny that they ask the following question:
How important will Senator Barack Obama’s presidential pick be to your decision whether to support his candidacy? Will it be very important, somewhat important, not very important, or not important at all?
Very Important……………………………26%
Somewhat Important……………………26%
Not Very Important……………………..13%
Not at All Important…………………….33% (Mirrored for McCain, similar distribution of results)
Even before they get the response to the vote preference questions above, this should have been flagged as a question that will not meaningfully contribute to the analysis. This is the architypical question a campaign wants to know, but will almost always misinterpret (see George F. Bishop’s excellent The Illusion of Public Opinion for more on deceptive questioning). Let’s think about what you are asking respondents to do here. The pollster is asking the respondent to assess the effect on a near-certain decision of an event in the future, the outcome of which is unknown. In an electorate as interested as this sample (93% say they will almost certainly vote vs. 55% of the eligable population in 2004, 51% in 2000; 91% of this sample places their interest in this election at an 8 or above on a scale of 1-10), this question is more likely than not a measure of interest than of influence. It is preposterous to believe that 52% of the sample is likely to change their vote based on Obama’s pick. In fact, only about 8% of the total sample is at all likely to consider any further information in deciding for whom to vote, and if the undecided voter study linked above is to be believed, then even fewer of them are likely to consider this information.
But first, let’s consider the topline result that is bolded in the report:
Total Important (Very + Somewhat)……52%
Total Not Important (Not Very + Not at all)…. 47%
What is your headline if you are CNN, waiting breathlessly for a text message from your BFF, Barry? To bulk up the importance of your unbelievably content free analysis, you say that “A Majority of Americans Say VP Pick is Important”. What do we know from this poll already? The VP pick in almost completely irrelvant, as only 2% of swing staters are undecided and of those that are decided, they are overwhelmingly decided. The probability based on the previous questions of the VP pick having any perceptable effect whatsoever is near-zero. So thank you media for hyping nothing all week and congratulations to the Obama campaign for letting their own buzz overwhelm their own week of economy message. Had there been less focus on the insipid speculation of who would be the VP nominee, Obama could have spent more time addressing hsi problems that the rest of the poll shows.
The troubling parts of this poll for the Obama campaign come at the end, and more complex analysis of the data would be useful in determining strategy and message to recover). Small pluralities most associate Obama with “being on their side” and “restoring respect for America in the world”. Obama has a 15% lead in “bringing the right kind of change”. Pluralities see McCain as “out of touch”, but not wide enough for comfort (4%). Obama has major deficits on whether he is a “strong leader” (-10%) and whether he “has what it takes to be president” (-12%). Further voters think of him by substantial margins as “too risky”* (+13%) and “saying what people want to hear, not what he believes in” (+16%) and “has flip-flopped on issues (+22%). Further, voters believe by wide margins that McCain is better to handle Iraq (-11%) and Afghanistan (-17%).
Even this late in the game, voters don’t know Obama and they don’t trust him. This may have to do with the fact that he’s black. It may have to do with the ads that McCain has run questioning his depth (and wow, is that balls from a guy whose deepest statement ont eh economy was that he didn’t understand it and that everything would be OK if we let Phill Grahm handle things). In all probability it has a lot to do with the fact that this election should be about George W. Bush and 8 years of insane and failed Republican policy, but instead it’s about whether we trust the smart new guy. Obama is fighting this election on his opponent’s issues, on his opponent’s terms, and on his opponent’s schedule. If Obama can somehow wrest control of the campaign narrative away fron John McCain (if only there were some large, week long event designed to showcase the Democratic Party beginning this evening!), and completely redefine the issues, stakes, and points of reference in this election and then show that in all of those points, it is John McCain that is wrong and dangerous for our country. I’m not saying go negative, I’m saying get a spine and start playing sme offense. In politics, the best defense is a good offense.
When we look at the breakdown of who has seen campaign ads, we see a few more interesting observations. Fewer people have seen ads about McCain than have seen ads about Obama. The good news for Obama is that for those who have seen campaign ads, respondents say those ads make them more likely to vote for Obama. Unfortunately, this is based on self-reporing and is in all probablility completely bogus. This is another question that is better answered by correlation analysis than by asking outright.
Finally, and perhaps most interestingly in this poll, we have measures of how individuals get their campaign news. This is a potentially excellent source of infornation, but somewhat useless on it’s own.
*as a side note, what is this question? a lot of reporters have intepreted it as a measure of discomfort with his race. I can certainly see that, but would need to see the raw data to make that statement. Outside of the race context, what does “too risky” mean? too risky in what situations? too risky for what? This question is teribly inspecific, and a great example of a question that is unreadable on its own. If this came up at my questionairre design meetings, I wold have nixed it unless we had a plan to use it as a dependet variable to explore what it means or had more specific wording. It appears that the sample knows what this means – at least enough to provide a definitive advantage to McCain – but the problem with questions like this is that they have such a wide potential for intepretation that even if we know for certain that the respondents think it means something definitive, we don’t know what they’re thinking without further analysis.
Recent Comments