Archive for October 25th, 2008

How to read State Polls

The election of the President is not really a national election, but fifty separate state elections, awarding ‘electoral votes’. A state’s electoral vote equals the number of Senators plus the number of Representatives. Florida has two Senators and 25 Representatives, so they have 27 electoral votes. In most cases it’s winner take all, though in both Nebraska and Maine they award them based on legislative district. So to guess on who will actually win the electoral college, where 270 votes are necessary to become President, one has to look at state contests.  As seen in 2000, one can lose the popular vote but win the election.

To look at the state polls the best tool is the polling page on the website Realclearpolitics, which allows one to click a state and see the polling history of that state during this election cycle. That’s important because state polls are usually built on a smaller sample size, a larger margin of error, and done less frequently – especially for small states. Only Pennsylvania has a tracking poll, the rest are polls done by various polling organizations. Since some people really get into trying to figure out the state of the race with detail (like me), I’ll go into the ‘complex’ way to read state polls by the end of this post. First, though, the basics for people who want to get the most information with the least effort.

Some states have a large number of polls published. On October 23rd, three different polls were published for the state of Minnesota, showing Obama leads of 19, 15 and 10%. That’s quite a range, and each poll had a margin of error of 4, 4.5 and 4.9% respectively (every time I write margin of error I mean plus or minus). Thus eyeballing it one can be reasonably confident that Minnesota is a state where Obama has a double digit lead. Most of the important states can be ‘eyeballed’ and a large number of polls can be compared. That should give a sense of where the state appears to be headed. To be sure, all the provisos about methods, assumptions, and variation discussed yesterday for national polls apply to state polls, which are almost always less exact than the national ones.

Some wonder about the “Bradley effect” as a systemic error, in which it is claimed people lie to pollsters if a candidate is black, suggesting that the black candidate will lose 5% between a poll and the result. I don’t believe the Bradley effect exists — there are other reasons for late swings (in that case a large quantity of absentee ballots returned with a GOP edge).  Moreover, that election was in 1982.

Now let’s get a bit more complex. A poll came out for Georgia from Rasmussen on October 23rd showing McCain with only a five point lead. Given the heavy early voting in Georgia, this could be a sign of an election night surprise – no one expects Georgia to go for Obama! Rasmussen interviewed 500 people, and it has a margin of error of 4.9%. Comparing with more recent polls, and it shows an improvement for McCain, but nothing dramatic.  On the 24th Inside Advantage showed Georgia with a one point Obama lead.  It was a one night poll with over 600 interviews, a margin of error of 3.8%, and shows the Republican with a two point lead in the Senate race (consistent with other polls).  This still could be an outlier, but the signs point to Georgia as a potential early indicator if Obama is going to have a blowout. 

A couple days ago a South Dakota poll showed McCain ahead 48-41. Another thing to note: anything under 50 shows that it isn’t ‘wrapped up.’ The poll was done by Mason Dixon, a reputable firm, and interviewed 800 voters with a 3.5% margin of error. Interestingly, they polled them on Oct. 13-15, before the latest upsurge in support for Obama. There are fewer polls to compare, and Bush won the state with 20% margins the last two elections, but in the case of an Obama landslide, South Dakota could be another surprise. Montana, also a GOP state, shows a poll with Obama leading by 5. But it’s done with a margin of error of 5%, with students in an upper level political science class doing the interviewing. This gives me a modicum of skepticism, especially given how young people are supporting Obama. Did they conduct all the interviews fairly? Montana is interesting, but I’ve got some skepticism of that poll.

Big Ten Battleground polls, also by universities, show large Obama leads for all the Big Ten states, including a 10 point Obama edge in Republican Indiana. The margin of error is 4%, 586 people interviewed, but it’s very different from polls earlier in the month showing 7 and 5 point leads for McCain. This suggests Indiana is in play, but a ten point Obama advantage? I’m skeptical – this could also be one of those outliers.

On the 24th a few polls arrived which one has to be cynical about.  ETV shows Obama with only one point leads in Virginia and North Carolina, suggesting movement towards McCain.  Yet the polling data was gathered in a period of over a month, since last September.  Given the volatility of the electorate, it may not catch the recent break for Obama.  On the other hand, it might show that McCain’s “hard” support is stronger than Obama’s, suggesting that McCain has a chance to win back those who have moved to the Democrat.  Another group of contrarian polls comes from Strategic Vision, showing McCain up 3 in Ohio, up 2 in Florida, up 6 in Georgia and down only 7 in Pennsylvania.  If true, this very good news for McCain.   However, this pollster is a partisan Republican pollster (on “Realclearpolitics” a partisan pollster is denoted with an “R” or a “D”).  That doesn’t mean it’s wrong, but the partisan pollsters on both sides tend to have a bias towards their party.   They could be looking at particular scenarios in their methodology, or perhaps, say, Democrats talking to a Democratic pollster don’t want to admit if they’re voting for the other party.   Still, I tend to expect a three or four point bias, and even with that built in, the states remain in tight races.

OK, let’s say you’ve got a map in front of you and you want to make predictions based on the polls (RPC also has an interactive map where you can play with awarding states to the candidates and make your own scenario).  I have two scenarios I use state polls to test.

Scenario 1: Can McCain win?  I’ve noted before, his strategy seems to be to flip Pennsylvania and then sweep the battleground “red” states that Republicans have won in the last two elections — Ohio, Indiana, Virginia, North Carolina, and Florida.  He’d also need six more votes, with possibilities coming from North Dakota 3, Montana 3 (likely) or Nevada 5 and Missouri 11 (less likely) Looking at the polls now, all of these contests look close enough that one can’t call the election for Obama yet.  Again, the national polls are meaningless at this level.  If McCain can run the table on these states he can win.  Yet Obama has leads by  6 in Ohio, 7 in Virginia, 10 in Pennsylvania, 2 in NC, 2 in Florida, and is dead even in Indiana (according to an average of recent polls).  This means McCain needs a significant uptick in support to pull it off.  How significant is unclear — state polls are generally not as accurate as national ones, and it could be that there is a systemic bias towards Obama (not due to political views, but in the methodological assumptions).   Still, this makes the job easy for Obama — it just takes a state (or in some scenarios 2) to derail any chance McCain has.

Scenario 2: an Obama landslide.  This would occur if Obama picks off states thought to be safe Republican.   NC and VA seem the most likely to flip, but for a true landslide the indicators would be Georgia early, Indiana later, and ultimately states like Montana and the Dakotas going blue.  State polls in each of these states make an Obama landslide possible as well.   In fact, the odds of Obama pulling of an almost unthinkable 400 electoral vote landslide (Republicans have done this, but it’s seemed out of reach for Democrats) seem even a bit better than the odds of a McCain victory.

So in reading the state polls now, an Obama victory seems likely, a McCain victory remains possible, and an Obama landslide is possible.  Even if McCain pulls off his Pennsylvania strategy, he could still stumble by losing the Omaha district in Nebraska and fall a vote short (which would yield a tie).  So, in watching the state polls, look to see trends.  If the polls are, on average, within 5 points, I’d say anything can happen.

For those of us who unabashedly follow and are fascinated by the ‘horserace’ aspect of elections, following this is addictively fun.  This is truly one of the most exciting electoral cycles of my lifetime, there are more new variables and uncertainties than in any contest I recall.  Right before the election I’ll blog about the states I’m going to be most interested in, and how one can tell early whether we’ll be looking at a McCain comeback, an Obama blowout, or a closer win for Obama.