Don’t trust polls – but based on the polls, I’m picking…

I presume today will be a day for bloggers everywhere to announce their election predictions.  I also presume that virtually no one will bother to read all of these election predictions.  At least I find the most boring thing in my Saturday paper to be the local sports guy’s picks for the Sunday NFL games.  Who cares?  The games are a heck of a lot more interesting than the predictions, they’re real, and they’re only a day away.

But how can I avoid it?  I’ve been following the polls religiously.  I been following the opinion pieces about why no one should be paying attention to polls.  I’ve been reading predictions of confident victories for the President, verging on 330 electoral votes, and I’ve been reading predictions of confident victories for Mitt Romney, pushing the same magnitude.  I’ve seen the predictions of a razor-thin race, which led me to revisit the history of the other close elections in U.S. history, like 2000, 1876, 1824, and 1800.  I’ve seen predictions of a Romney win in the popular vote but an Obama win in the electoral college, which led me to research the case for and against getting rid of the electoral college.  I’ve even written a post about the real possibility of an electoral college tie, and the craziness that would thereafter ensue.

It’s been a fun and enlightening electoral season, but also a frustrating and confusing one.  At the end of the day one could make a very good argument for dispensing with predictions altogether.  But there’s a chance I’ll get lucky and get it exactly right, and then I’ll get to title the next blog piece “I told you so.”  Who could pass up that chance?

To Trust The Polls?

As I’ve said in an earlier post, I have been relying on RealClearPolitics and electoral-vote.com over the past month to track the polls.  Lately I’ve discovered that the Huffington Post also has an interesting view of the electoral map, which has a helpful slider bar at the top of the page that lines the states up based on confidence that one candidate or the other is truly leading.  (This is essentially a reflection of margin of error and whether the voting preference difference falls inside or outside that margin.  I do not know exactly how that is calculated at the Post, however, because for most states the percentages are based upon aggregates of polls.  For reasons related to my comments below, I doubt their confidence percentage is all that useful.)

All of these projections are based upon polls, and this year, more than ever, the reliability of the polls have been called into question.  There’s at least 4 good reasons for that.

Reason 1:  No reliable method of reaching people

The methodologies that polls use have simply had to change.  Twelve years ago – even eight years ago – almost everyone in America had a land-line telephone that was used as the primary telephone.  Most people – at least most people in particular demographics – now use cellular phones as the primary means of communication.  Some people use cellular phones as the only means of contact.  Twelve years ago – even eight years ago – most people answered their telephones.  But now people have fought back with answering machines and caller IDs.  In this age of robo-calling and telemarketing, more people screen their calls and may be apt to ignore a call when it is from an unknown caller.  This may vary by demographic characteristic as well.

Reason 2:  No reliable model to correct for atypical samples

To adjust for the sampling biases that occur with this previously unseen behavior, pollsters have had to statistically correct their data to better-match the true demographic characteristics of the public.  There is debate among pollsters as to whether one characteristic that should be controlled for is political party affiliation.  Romney supporters have complained that the polls are skewed by over-sampling Democrats.  Pollsters counter (reasonably, in my opinion) that you can’t control for party affiliation since one of the things being sampled – i.e., one of the things that is admittedly unknown – is how much party affiliation may have shifted.  But even controlling other characteristics – for example, age, race, and income level of a voter – relies on applying the demographics of previous elections.  But the electorates of 2004 (a good Republican year), 2006 and 2008 (good Democrat years), and 2010 (good Republican year) have been so volatile, it is difficult to have much faith in these models.

Reason 3:  Unknown characteristics of drop-outs

One admission that a poll is not perfect is its margin of error.  However, what is rarely said about margin of error is that it represents the minimum error of a poll, because it only assesses one kind of error – the kind you make projecting the characteristics of a population (e.g., all 3,750,000 voters in Virginia) from a sample (e.g., a telephone poll of 1000 likely voters in Virginia).  The margin of error reported in polls is a straightforward mathematical calculation – every poll of 1000 voters has a reported margin of error of 2.9%.

But imagine the following.  Poll #1 places 2,000 phone calls, but survey responses are obtained from 1,122 people.  The others either weren’t home to answer the phone, or screened the call, or declined to participate after answering.  Poll #2 places 35,000 phone calls, but survey responses are obtained from 1,122 people.

Which poll is more trustworthy?  Both will report a margin of error of 2.93%.  Poll #1 has a drop-out rate of just under 50%.  Poll #2 has a drop-out rate of just under 97%.  Poll #1, by the way, is complete fantasy.  Poll #2 is an actual recent poll of Michigan voters, and by far reflects what is more common.  This 97% failure rate is quite common.  We know nothing about these people.  If they are just like the 3% who did complete the survey, there’s no problem, and the poll’s true margin of error is 2.93%.  But it is awfully hard to believe they are like the other 3%.  Almost by definition, there is something curiously odd about these 3% of phone call recipients.  It’s odd that they were home.  It’s odd they didn’t screen their calls.  It’s odd they took the time to talk to someone they didn’t know.

If they are also odd in their politics, the polls are untrustworthy.  Understand, I don’t mean “odd” in the sense that their politics are radically different from everyone else’s.  Not as in, this is the 3% of Americans who thought monarchy would be a good idea again.  I mean odd as in 10% more likely to be Democrat, or 10% more likely to be Republican.

I can make up scenarios that favor either, but that would be just guessing without data.  The only data we have is that many polls do seem to be undersampling Republicans and/or independents, which is suggestive that the polls may underestimate Romney’s popularity.  Suggestive, but hardly definitive.

There are a number of other reasons why the margin of error is a best-case number, which have been beautifully illustrated elsewhere.  (I can’t recommend this link highly enough.)

Reason 4:  It’s the turnout, stupid

Even with all this having been said, polls are not bad at predicting a winner.  But they’re not especially great at predicting turnout.  A person who tells you “I choose Barack Obama” is probably naming for you whose lever he or she will pull if they get to the polls on election day.  But a person who tells you “I’m definitely voting” is not nearly as reliably predicting his or her own future behavior.  Some Romney fans making the case that the polls underestimate Romney’s support have floated the vapor-thin theory that people don’t want to tell a pollster they aren’t voting for President Obama (because he’s black or because the person previously voted for the President).  This strikes me as implausible.  But far less implausible is that a person will tell a pollster that they are a “likely voter” when they aren’t.

So how do we predict then?

I don’t think the polls are very far off.  For one thing, they aren’t telling us anything beyond belief.  No one’s reported Romney up in California, for example.  No one’s suggested that Kansas is a battleground state, or that Gary Johnson is gaining ground – anywhere.  For another, I don’t think the 3% who answer the phone and put up with the pollster are enormously odd, just a little odd.  So bottom line, we start our predictions with the polls.

But we don’t stop there, and we don’t get false confidence from reported margins of error or from the number of polls reaching the same conclusions (from the same problematic kinds of data sets).  We also think about the state of the economy, the key issues, the momentum, the effectiveness of each man’s message and campaigns, the historical trends in each of the states, the effectiveness of each party’s “ground game”, the kinds of ballot initiatives in each state that might spur turnout, the enthusiasm of the campaign rallies, and, when all else fails, we turn to a random number generator (sometimes called a ‘coin’).  Some of these things favor Mitt Romney and some favor President Obama.

So what’s your prediction, then?

As you can see, I’m stalling.  But here we go.  I’m picking Mitt Romney in a tight 271-267 election, and with Romney narrowly winning the popular vote by less than 1%.  Here’s the map as I see it:

Electoral projections – RAPsheet

The states I am most-likely to get wrong are Wisconsin and Virginia, both of which I have going for Romney.  Virginia has been surprisingly stubborn for Romney supporters, most recently trending back towards the President.

Of the midwestern states that are in play, I believe Wisconsin is most-likely to switch from its 2008 preference for the President.  Obviously most people believe that Ohio is most-likely to switch, and with good reason – Ohio went Republican as recently as 2004.  If coal were a bigger issue in Ohio and the automotive industry was less of an issue in Ohio, then I would be more optimistic.  Plus, the Democrats have simply embedded themselves in Ohio since 2004 and have put an enormous share of their efforts into that state, while Republicans have had to worry about Virginia, North Carolina, and Colorado, in addition to Ohio.

Wisconsin – whose motto “Forward” is also President Obama’s campaign slogan – hasn’t gone for a Republican President in quite some time, but they did go for a Republican governor – twice – in the past couple of years.  Scott Walker won the governorship in 2010.  While that could be written off as riding the wave of anti-Obama sentiment that defined the 2010 midterms, Walker actually earned more votes in the 2012 recall election, suggesting that the anti-Democrat sentiment among Wisconsin independents has staying power.  Furthermore, the 2012 recall election was really centered on one issue: Wisconsin’s budget, and whether the citizens of Wisconsin were willing to tolerate restrictions on union bargaining power.  If the union get-out-the-vote machine couldn’t take down Walker in a recall election – an election taking place at an odd time of the year when motivated turnout should have ruled the day – then my guess is that the Democratic ground game is not as strong in Wisconsin at it is in other in-play states like Ohio and Nevada.  Given that Walker’s reforms have actually worked to get control of Wisconsin’s budget while improving unemployment and without requiring the loss of teacher jobs, Wisconsin voters may well enthusiastically respond to Romney’s vision over President Obama’s.  Paul Ryan, native son, is just icing on the cake.

On the other hand, if Wisconsin goes for President Obama despite these promising factors, it will count as Mitt Romney’s major missed opportunity.  Arguably, Paul Ryan has spent not enough time in his home state, bragging about Scott Walker’s achievements and encouraging Wisconsinites to export their success to the rest of the nation.

Another missed opportunity may be Pennsylvania.  The Republicans have surged there in the past 3 weeks, and they may be kicking themselves later for having focused efforts on Ohio rather than Pennsylvania.  In Pennsylvania, coal is an issue and the auto industry is not.  Romney seems to be making gains with suburban women and Jews, both of which could have helped him outside Pennsylvania’s “T” which leans Republican.  Pennsylvania may also be the one state in which Romney’s 47% line doesn’t hurt quite as much – because the state’s 47%ers have already been denigrated by the President as people clinging to their guns and their religion.  I’m worried though that the Pennsylvania push will be too little, too late.

I think another surprise for the Romney/Ryan camp could be Nevada, which seems to be the battleground state most have as the least-likely Romney upset.  Apparently the union and Harry Reid machines run very well in Nevada.  But Nevada is really hurting right now.  As a tourism-dependent state, Nevadans are waiting and waiting for that roaring economic recovery the administration keeps promising is just a year away.  Are they really willing to live through 4 more years of the recovery being one year away?  With a good turnout in Nevada, I could see the state turning red more easily than Ohio, Iowa, Minnesota, or Pennsylvania.

On the other hand, Republican prognosticators – I include myself – have shown a real blind spot in two areas: Florida and Virginia.  It is almost painful to imagine those states not returning to Republican red after Barack Obama’s upsets there in 2008.  No scenario makes sense without Virginia and Florida.  No hoping for upsets in Ohio or Pennsylvania, no magic Colorado-Wisconsin-New Hampshire trifecta works without Virginia and Florida.  But the prospect of splitting these states is very real, and if that happens and if Wisconsin or Colorado does not come through, it may be a comfortable electoral win for the President.

I’m sticking with my 271-267 pick.  But if it turns out that way, expect a protracted battle of recounts and legal challenges.  Expect stories of voter intimidation in usually-Democratic precincts, and unexplainable silence on similar stories in Republican-leaning districts.  You can already hear the rumblings.  But don’t believe it: your vote will count.  So get out there tomorrow if you haven’t already.  Let the games begin.

Advertisements

5 comments

  1. I agree. You can’t go by the polls. I think Republicans are much more motivated. And for good reason. Great analysis yet again on those polls though. And I also agree it is not realistic to project a landslide victory for Romney as some have done. It is always a fatal mistake to underestimate the opposition in any conflict. The South did this “We can lick the yankees in 5 days”. It took nearly five years. And it is going to take every patriotic hand casting their vote for Romney today to save America from the scourge of the Obama administration.

    1. Thanks for reading and giving a favorable review. I’ve been shaking my head a lot at prognosticators on both sides. It will be an interesting night (or week, depending on how many legal challenges there are).

  2. Oh yes, I am confident this election will be contested. This could drag on for weeks. New York may be an issue because of Sandy. They have some kind of law that says if they have less than a 25% turnout, they can stall the vote until the 20th of November. And that is just for starters. Between faulty voting machines, and a host of other nasty’s this could take awhile. I hope not. Anyway, your most welcome. I appreciate the factual and realistic analysis you provide, it takes a lot of work, and it is well done. Have a great day, it should prove very interesting before days end.

  3. […] elections was whether or not the polls were providing us a clear picture of the 2012 election.  I argued that they were likely underestimating Romney’s appeal just a little, but just enough to make […]

  4. […] there hadn’t been so many forecasts of a Republican win (only among Republicans, by the way, including me) – the election could rightly be seen as an electorate moving back to the right from as far […]

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: