Monday, November 5, 2012

Punditfulls of Predictions!



My Election Prediction 2012 (270toWin)

During the primaries, I remember telling a friend of mine that if Romney was nominated, this election would look a lot like 2004.Turns out the similarities are striking. However, one of the most comical similarities is the denialism. In 2004, we had sites like the Donkey Rising Blog trying to explain away Bush's favorable numbers in the polls. Today we have sites like UnSkewed Polls doing the same thing to Obama's poll numbers, sometimes even projecting Romney will win in a near-landslide! However, the methods used to come up with these numbers are deeply flawed. But that hasn't stopped a number of pundits from giving similarly absurd projections:
Every one of these pundits is predicting a highly unlikely result. Other than Jim Cramer, each one predicts Romney picking up over 300 Electoral votes, meaning Obama would pick up at most 238 electoral votes. According to the RCP averages, this means the polls would have to be off by up to 4 points in some states! In the FiveThirtyEight model, such a thing would be historically unprecedented to an incredible degree! However, unlike Cramer's prediction, the FiveThirtyEight model at least holds this as a possibility, albeit an incredibly small one. One has to wonder why so many pundits are risking their credibility with such unbelievably unlikely predictions. Of course, one would also have to underestimate the level of cognitive dissonance some of these pundits have, as well as their fans.

I have to confess. Over the last couple of months, I have grown an unhealthy obsession with the polls. I have made trips to RealClearPolitics, Pollster, and FiveThirtyEight a regular habit. When I wake up in the morning to take the puppy out, I check the Rand Corporation's Daily Poll. Throughout the day, I check RCP for updates to the Rasmussen poll, the Gallup Poll, Investor's Business Daily, and ABC/Washington Post, as well as up to date averages of the state polls. I also make regular trips to Pollster to check on polls RCP doesn't factor in. I understand that you cannot cherry pick polls, that you have to account for house effects, and that you have to look at both the the national and state polls to get a clear picture of the race. As a result, I have mostly taken a break from blogging, preferring to post whatever thoughts I may have on Facebook rather than this blog, likely to the utter annoyance of all my friends. So before the election concludes tomorrow (hopefully), I thought I would put my own projection on the table (on the top of this page). Some explanations:
  • Pennsylvania and Michigan: Despite Romney's recent play on Pennsylvania, the chances of him winning the state are extremely low: FiveThirtyEight gives Obama a 99% chance of taking the state. Other than the extremely right-leaning Susquehanna Poll, the most Romney-favorable polls still give Obama a comfortable 3 point lead. As Nate Silver has explained, this state is also incredibly inelastic, meaning there are rarely very many voters who will wait til the last minute to make a choice.  In Michigan, the only poll giving Romney any chance is the Baydoun/Foster poll. Although this poll is classified as a Democratic pollster, its numbers in Michigan have been extremely favorable to Romney, meaning it has a larger than normal house effect. Nate Silver gives Obama a 100% chance of winning the state and Romney has not seriously contested it.
  • Nevada: Although RCP only gives Obama a 2.8 percent lead here, Nevada polls have consistently underrated Democrats for the elections in 2008 and 2010, by large margins. Given the issues among pollsters with the Latino vote, as well as Harry Reid's superior ground game in the state, I doubt 2012 will be any different. Nate Silver also has Obama at a 94% chance of winning Nevada.
  • Wisconsin and Ohio: Both of these states show small but persistent Obama leads. And neither state show any poll with a  Romney lead. RCP has Obama up 2.9 points in Ohio and 4.2 points in Wisconsin. The distribution in Ohio has been very tight, despite the large number of polls conducted in the state. Wisconsin's distribution has been less so but it has also shown better numbers for Obama in its more Obama-favorable polls. However, both the polls showing strong Obama leads have been incredibly volatile over the course of this campaign. Nonetheless, Nate Silver gives Obama a 92% chance of carrying Ohio and a 97% chance of carrying Wisconsin. 
  • Iowa and New Hampshire: Both of these states have had very sporadic polling. But Obama is still favored in both. RCP has Obama up 2.4 points in Iowa and 2.0 points in New Hampshire. In addition, only two polls give Romney a lead in Iowa and none give him a lead in New Hampshire. The two polls that give Romney a lead in Iowa tend to have very Romney-favorable results, and their leads are only a point. At least one Romney-favorable poll also gives Obama a 4 point lead in Iowa. In addition, Iowa's only local poll gives Obama one of his best leads. The situation is similar in New Hampshire. Nate silver gives Obama a 86% chance of winning Iowa and an 86% chance of winning New Hampshire.
  • Virginia and Colorado: Both of these states appear to be toss-ups, although Nate Silver marked them both as "Likely Obama" today. According to RCP, Obama has a 0.3% lead in Virginia and a 1.5% lead in Colorado. However, there are still a few polls giving Romney slight leads in each state. Although early voting in Colorado seems to be favoring Mitt Romney, Democrats have been closing the gap (we should see the final results of early voting tomorrow morning). Some polls that have asked about early voting show that Obama is expected to do better on election day than he has in early voting. Others show he is expected to do worse but may get better results tomorrow morning. But this may not matter. In 2008, Obama barely edged out McCain in early voting, yet still won the state by 8.6 percentage points, meaning historically the early voting trend may not be unprecedented in Colorado given that Obama will likely not win by 8.6 percentage points anyway. For the purposes of my projection, I predict Obama will carry Colorado, mainly because of the polls. Virginia is very close nationally, and it is hard to come by good early voting statistics for the state. Their only local poll has also been extremely volatile. Given Nate Silver's projection, as well as the RCP average, I give Obama the slight edge in the state. Update 11/6/12: Wow, I cannot believe I didn't notice this until now. I fell for the fallacy of equating registration advantage with candidate advantage in Colorado. I took a look at whatever polls I could and noticed that Democrats do not have the identity advantage in Colorado that they do in other states. It appears as though, although Democrats and Republicans seem tied in terms of party identity, Republicans hold a 2-3 point registration advantage over Democrats in these polls, suggesting Obama is leading big with both registered and identified Independents. Indeed PPP's cross-tabs suggest as much. As a result, it is entirely possible, even likely, that Obama is still winning the early vote in Colorado, even though registered Republicans are ahead.
  • Florida: On the surface, this state appears to be even more of a toss-up than other states. RCP has  Romney up 1.5 points, but FiveThirtyEight has Obama up 0.2 percentage points. In addition, this is a state where the inaccuracies in polling the Latino vote should make the final outcome more Obama-favorable than the polls suggest. However, the local polls are Romney's strongest polls, and local polls should be given more weight. As a result, I would buck the FiveThirtyEight model (just barely) and give Florida to Romney.
  • North Carolina: RCP gives Romney a 3.0 point lead in this state and Nate Silver gives Romney a 72% chance of winning the state. Obama does have a significant lead in early voting, but the polls for North Carolina reflect the polls of Ohio, but in Romney's favor. Given the razor thin margin Obama won this state by in 2008, I seriously doubt Obama will win this state again in 2012.
Notice I rarely mentioned ground games. It is extremely hard to predict what kind of an effect a strong ground game will have on the election. Nevada is the obvious exception

So my final electoral vote prediction is Obama at 303 electoral votes and Romney at 235 electoral votes. I would expect the election to swing anywhere from Obama winning 281-257 to Obama winning 332-206. So let's see what happens!

Note: for early voting results, see here.

Thursday, November 1, 2012

Busting The Romney Fauxmentum Myths

http://i2.cdn.turner.com/cnn/dam/assets/121023030536-24-debate-gi-1022-horizontal-gallery.jpg


Those crazy polls are lying to us! At least that's what many pundits would have you believe. The polls tell that at the national level, the race is statistically tied (RCP Obama+0.1, Pollster Romney +0.1, TPM Obama +1.4. Yet state polls suggest a slight but persistent Obama lead. Serious election forecasters all predict an Obama win, although with slightly different levels of uncertainty. But for many pundits, their gut tells them something else. For many of them, Obama is such a terrible president they just cannot fathom the idea that not only is Romney not projected to beat Obama in a landslide, he isn't even projected to win at all! Yet they feel as though Romney has had some kind of momentum coming out of the Denver debate, meaning he should be poised to win. To them, there must be something wrong with both the forecasters and the polls. However, in the last few days HuffPost pollster has done an exceptional job busting a few of the more common myths coming from these pundits:

Myth: Polls are oversampling Democrats, making it look as though Obama is doing better than he really is.

I touched a bit on this claim back in September when it seemed as if pollsters and forecasters left it largely untouched. Since then, we have seen a few tackle this issue. However, it seems to persist even today, along with the observation that those polls which are "oversampling" Democrats also show Romney winning the Independent vote, often by large margins. Nowadays the focus on Ohio, which is no surprise given that it is the state most likely to decide the election. Now it is true there are polls giving Democrats up to an 8 point lead in Ohio. And it is true many of those same polls also show Romney winning the independent vote by large margins. And it is likely going to be the case that exit polls won't show Democrats with an 8 point lead among party identity either. In my last post, I pointed this out, but did not speculate as to why this is the case, and why we should still expect Obama to win by the same margins the polls predict. However, I suspected Romney's popularity among Independents was actually related, suggesting that Romney supporters were more often identifying as Independents than Obama supporters, boosting Romney's support among Independents as well as the Democrat party identification advantage. Today, Pollster blogger Nick Gourevitch confirmed this suspicion:
...in recent Ohio polling, Romney's lead with Independents is inversely correlated with Democratic party identification advantage.
What does this mean? It means that polls with larger Democratic party identification advantages are showing bigger leads for Romney with Independents. And vice versa. The following chart is a slight adjustment of a chart posted by @numbersmuncher last week showing party identification spreads vs. the vote among Independents in recent Ohio polls where this data is published. The chart below is sorted by Democratic party identification advantage -- so the polls with the greatest Democratic margins are up top and the polls with the lowest margins are at the bottom:

2012-11-01-Nov1Chart1.png
A quick glance at the chart above clearly shows this inverse correlation. Romney's biggest leads with Independents (except the one Suffolk poll) all come in polls where the Democratic party identification is largest. And the math backs up what the eye sees -- the correlation coefficient between the Independent vote and the party identification advantage is -0.61. While the correlation is by no means perfect (there are many other variables that impact these numbers), the trend clearly exists. (emphasis mine)
This shouldn't be too hard to understand. The first thing you should note is that there doesn't seem to be practically any correlation between Obama's lead and party ID advantages (with the exception of one Rasmussen Poll, which can be better explained by other factors). This suggests that the proportion of Romney and Obama supporters remains approximately the same throughout each poll, despite party ID advantages. Some polls such as CNN, U of Cincinnati, and Rasmussen are more likely to show Romney supporters identifying as Republicans and/or Obama supporters identifying as Independents than ARG, Time, or CBS/Quinnipac, which show approximately the opposite. As a result, the apparent Democratic ID advantage, as well as the Romney advantage among Independents, can best be explained by how a poll determines party identification. Remember that there is a substantial difference between party registration and party identification. It may also be interesting to note that the Republican-leaning Rasmussen poll gives Obama some of his best numbers with Independents.Yet why do these polls show such different results for party ID? Nick Gourevitch explains:
This can happen for a number of reasons. Pollsters use different questions and methodologies. Live interview polls and automated polls produce different party identification distributions (ones that are also different from in-person Election Day exit polls). Pollsters ask party identification in different ways. Some use three-point scales, others five-point scales and others seven-point scales. Some push "leaners" and some don't. Question ordering also matters - especially for pollsters who put party identification at the end of their survey. If you have a survey with a bunch of questions on social issues, it might cause more people to identify as Democrats in a socially liberal state, but might cause more people to identify as Republicans in a more socially conservative state.
However, why is it that most of these polls show Democrat party identification so much higher than Republican party identification? Pollster Thomas M. Holbrook finds something that should give us a clue:
"One way of assessing the relative value of party for each of the candidates is by looking at rates of party affiliation in the electorate... These data show that the Democrats have held an affiliation advantage throughout the 2012 campaign, one that has ebbed and flowed a little bit and now stands at approximately six percentage points.
...
throughout this campaign period the Democratic Party has been viewed more positively than the Republican Party. In fact, there is not a single poll in this series in which the Republican party registered a net positive rating, and not a single case in which the net Republican rating was higher than the net Democrat rating. The average net rating for the Republican Party in this series is -13, whereas the average for the Democratic Party is +.3." (emphasis mine)
This could explain why Romney supporters are more likely to identify as Independents this election. No doubt many of these self-described Independents identified as Republicans back during the 2010 elections. Yet throughout 2011, Republican party favorability dropped drastically, recovering some but not all in 2012. At the end of 2010, Republicans held a net favorability rating of -2.4. Today that number has fallen to -8.3. Democrats on the other hand, have seen very little change since the end of 2010. Democrats held a net favorability of ~1.4 points (can't see exactly). Today that number has slipped barely to -1.5, barely even statistically significant, and about half the drop seen by Republicans.

Update 11/5/12: I just found this article essentially confirming this suspicion. Actually most polls seem to indicate there are fewer voters identifying as Democrats now than in the 2008 exit polls, instead shifting to Independents. However, the drop is much larger with Republicans. This means that more Republicans than Democrats have shifted to thinking of themselves as Independents, causing Independents as a whole to be more conservative and Democrats to have a large identity advantage. 

All of this should show how wrongheaded attempts to "unskew" polls really are. Unless independents are adjusted along with party ID, you actually end up oversampling Romney supporters!

Myth: Undecideds will break toward Romney because he is the challenger.


While it is true that Obama's lead nationally and in state's like Ohio aren't large enough to put Obama's numbers over 50%, the "Incumbent Rule," which states that undecideds generally break towards the challenger, may be less of a factor than some would have you believe. Pollster blogger Mark Blumenthal explains:
"When The New York Times' Nate Silver examined polls from 1998 to 2009, he found no evidence that "the majority of the undecided vote broke against the incumbents." His advice: "Focus on the margin between the candidates, just as you might instinctively do."
In addition, he found little evidence this year will be an exception. This weekend, we will see the release of Pew's final survey report, which should shed more light on this question.

I will continue to update this posts as new information arrives...