Final Forecasts

The table below summarizes various models I've worked with this fall, as well as the projections of two of my undergraduate classes. All of these projections point to a big night for Obama. I have the most confidence in the Electoral College forecast from the October model, and I expect the (two-party) popular vote to come in somewhere between the October model (53.9%) and my PS/APSA model (55.7%).

It's interesting to note that both of my undergraduate classes independently came to the same conclusion: Obama 228 electoral votes, to McCain's 200. Their electoral maps are exactly the same as Karl Rove's map.

Paths to 270

In addition to my post on leading indicators to watch for tomorrow night, check out Jay DeSart's summary of the "paths to 270."

Leading Indicators for Election Night

Several states (Georgia, Indiana, South Carolina, Vermont, and Virginia) will be the first to report election results tomorrow night, coming in at 7:00 p.m.(EST). Although the outcomes in South Carolina and Vermont are foregone conclusions, I will be watching Georgia, Indiana, and Virginia for indicators of an early or late night.

Here are some things that I think could signal a big night for either candidate and give us some hint about how things will play out:

The shorter version: If Obama carries all three states, let the kids stay up--it's going to be a short night; if McCain carries all three, especially by wide margins, put the kids to bed and get ready for a long night.


UPDATE: Florida is not included here because the much of the panhandle votes until 8:00 EST. That said, I'm not sure if the networks will wait until all the polls are closed to make a call. If they call Florida at 7:00, here's my analysis:

A win for Obama is now expected, based on current polls. A win that is significantly bigger than the spread could be a sign of under polling his support, or of the impact of he GOTV machine, and augurs for a bigger than expected night for Obama.

A win here by McCain, especially if by more than a couple of points, spells trouble for Obama. Among other things, a McCain win could be a sign that pre-election polling was overestimating Obama's support. Obama can still win without Florida, but losing here would be a sign of a much tighter race for the rest of the evening.

Feeling Mavericky

With no offense intended to Senator McCain, I remember the real Original Maverick. I also have fond memories a true American (100%) Maverick. I'm kind of surprised that neither side has found a way to use one of the historical Mavericks to their advantage. Of course it could have something to do with the fact that no one under thirty has ever heard of them. I know my students (and younger colleagues) had no idea what I was talking when I made reference to them.

I don't know what the Original Maverick is up to these days, but I know which side he used to be on.

The Jewish Vote

John Sides has a new post over at The Monkey Cage on the levels of support for Obama among Jewish voters. One question that has come up during the campaign is whether Jewish voters will support Obama to the same degree that they have previous Democratic nominees. The data John cited are from Gallup tracking polls taken from October 1-21 and suggest that Jews are pretty firmly in the Obama camp (74%). While I have no reason to doubt that Jewish support for Obama was high in the early part of October, I'd be interested in seeing what the trend is like, especially in the last couple of weeks.

On October 14, a columnist for the New York Post reported that Jesse Jackson, while speaking at a conference in France, said that "Zionists" would lose their control of U.S. foreign policy under an Obama administration. The story was soon picked by CNN and Jesse Jackson was back in the news for a few days. Who better than Jesse Jackson to drive a wedge between Jewish voters and Barack Obama? Is it possible that the coverage of Jackson's comments had an effect on Jewish support for Obama?

One of the difficulties with assessing Jewish support over the course of the campaign is that data are hard to come by (at least until the election is over). Prior to seeing the Gallup poll I had scoured most of the national polls on Pollster.com looking for religious breakdowns and came up empty--almost. The IBD/TIPP tracking poll provides five-day rolling averages across multiple demographic categories, including religious affiliation. IBD/TIPP doesn't provide the sample sizes for subgroups but the number of Jewish respondents in each survey is not doubt very small, given that the Jewish vote accounted for only 3% of the electorate in 2004. With that in mind, and bearing in mind also that these data come from just one source, the graph below tracks the Jewish vote over the last couple of weeks.

Although Jewish voters clearly favor Obama, their level of support averaged over the entire time period (67.8%) is somewhat lower than the 74% given to John Kerry in 2004. Before the Jackson comments got attention, Obama averaged 74% of the vote (the same as the Gallup results); in the middle period, when the polling samples included some days before and some days after the airing of Jackson's comments, Obama averaged 71.4%; and in the last period, when all of the samples were taken after Jackson's comments came out, Obama averaged 61.2%. Interestingly, while McCain seemed to initially benefit from the decline in Obama's support, there was a real up tick in undecideds in the last period.

I don't want to make too much of this, as I only have a few data points, from small samples, taken over a short period of time. If anyone has access to other data that might shed more light on this, please feel free to share.

Electoral College Update

Jay's latest update to our Electoral College forecast (right panel) switches West Virginia from Obama to McCain. To be honest, I was a bit squeamish about having it in the Obama column but that's what the model dictated, until today. As it stands right now, North Carolina is teetering and could switch to Obama with a few more polls showing him ahead, and West Virginia could move back to Obama if the polls there reverse course. Other than that, though, I don't see many sates as likely to switch between now and election day.

Wisconsin Survey

This semester I'm coordinating a small group of seniors in our new Undergraduate laboratory for the Empirical Analysis of Politics (ULEAP), and I'm also teaching freshman seminar titled, Politics by the Numbers. Both groups of students, along with me and my capable T.A., Aaron Weinshenck, have just completed a survey of Wisconsin voters.

The Survey was conducted from October 8 to October 15, using the CATI lab at the Center for Urban Initiatives and Research (CUIR). There were a total of 434 completed interviews (392 of whom answer the vote question), for a margin of error of appr0ximately +/-4.7 percentage points (+/- 5.0 points for the vote question). The sample was weighted to reflect the distribution of sex, age, and party identification in the Wisconsin electorate, as reflected in exit polls from the 2004 and 2006 elections.

The results of the survey indicate a wide lead in the Badger state for Obama:

Obama's margin of 15 points is wider than most polling in Wisconsin this fall, but is very close to the 17 point spread in a recent Quinnipiac poll.

Why is Obama doing so well in Wisconsin? In a word, the economy. When responding to an open-ended "most important problem" question, Wisconsinites overwhelmingly identified the economy:

When asked which candidate would do a better job handling the most important problem, respondents gave Obama a decided advantage:

When asked more directly to evaluate the state of the economy responses were almost uniformly negative:

Respondents were also decidedly gloomy when asked about general satisfaction with "the way things are going" in the country:

As expected given the nature of these findings, Wisconsin residents view President Bush's job performance negatively:

In Wisconsin, as in the rest of the county, the levels of economic pessimism, dissatisfaction with the direction of the country, and negative evaluations of the president create a hostile environment for the incumbent party candidate, John McCain.

Forecasting the Electoral College with October Data

As promised in my last post on the September Model I'm be posting a rolling Electoral College forecast (in the panel to the right), based on past election results and October national and state-level trial-heat polls. As with the September model, these state-level forecasts were developed with Jay DeSart. Clicking on to map to the right will take you to Jay's web page, where more details are available. As new polls come in the forecast will change, so please check in on on a regular basis for changes.

The results for today are:

• Electoral vote: Obama 354, McCain 184
• National popular vote: Obama 52.85%, 47.15%

The biggest change from the September model is that both Virginia and West Virgina are now in the Obama column.

Electoral College Forecast--September Polls

Back in April I wrote a post about using September polls to forecast election outcomes in the states. The general finding from the work I've done with Jay DeSart is that candidates generally win states in which they hold a lead in September polls and always win when they lead outside the margin of error. Jay and I use this information, along with a lagged vote variable and a control for national trial-heat polls in a state-level forecasting model.

Jay has done the heavy lifting on this model and has posted the details of the 2008 forecast on his web page. The results are summarized in this map.


Based on the September Poll Model, we project an Obama Electoral College win with 336 votes to McCain's 202 votes. Projecting the national popular vote based on predicted state margins and assumed state-level contributions to the national vote, we get Obama with 51.8% of the two-party vote.

Coming attraction: We will soon be unveiling an October Model that tracks changes in projected electoral votes as new polls come out.

Update: Missouri but not Virginia? Several readers have asked in comments why we are calling Missouri but not Virginia for Obama. Good question. The short answer is because that's what the model projects. Sometimes when you run a model like this you get results that make you scratch your head and wonder what's going on. But if you look under the hood of the model, it's somewhat easier to understand.

As described above, the model considers two state-level variables, the average poll result from September and a lagged vote variable from previous elections (averaged over four prior elections). On the lagged vote variable, Democrats have done better in Missouri than in Virginia, so this would push it toward the Obama column. On the poll variable, Obama did better in Virginia than in Missouri but not by enough, given Virginia's voting history to call the state in his favor. While Obama was leading in Virginia polls by the end of the month, he trailed McCain earlier in the month, and the month-long average was just about dead even. In the end, the data from September polls and lagged votes together pushed Missouri just into the Obama column and Virginia just into the McCain column.

No doubt part of the surprise here is due to Obama's strong performance in recent polls in Virginia. But remember, his biggest gains have come in the first week and a half of October and are not incorporated in this model. As promised, an October model is coming soon to a blog near you!

Election Forecasts

A new symposium on forecasting the 2008 presidential election is now available online in PS:Political Science and Politics. This is a collection of forecasts from several political scientists, including me. The crucial table (below, click on it to make it easier to read) comes from Jim Campbell's introductory essay. Campbell summarizes each author's model, prediction, and level of certainty.


All of the forecasts are based on the percent of the two-party vote predicted for the incumbent presidential party, so the predictions are for McCain's share of the two-party vote.

A couple of things are worth noting. First, there is quite a range in predicted outcomes, from a low of 41.8% (Brad Lockerbie) to a high of 52.7% (Jim Campbell), with a median forecast of 48.0% for McCain. My own national forecast (my state-level model depends upon September polls and is not yet ready) is for McCain to get 44.3% of the two-party vote, so I'm a bit on the low end. Second, while there is a lot of diversity among the models every one except Norpoth's includes at least one retrospective variable, such as presidential approval or some measure of economic performance.

So, what should we make of these predictions? One interesting finding from the last several years of election forecasts is that while there is a lot of variation in error across models in any given year, almost every forecasting model has predicted correctly the eventual winner of the popular vote. In fact, going back as far as 1996 I calculate that leading forecasters (mostly published in PS, as well as a couple of other prominent models) called the correct popular vote winner in 27 out of 28 cases . In addition--and I'm not sure how to explain this, or what type of bias it might imply--in 25 out of 28 cases, the models overestimated (sometimes slightly, other times by a wide margin) the share of the vote going to the incumbent party. These two findings, combined with the direction of the forecasts posted above, would seem to suggest dim prospects for the McCain campaign.

For what it's worth McCain's share of the two-party vote in current polls is 46.4%.

Brewers!

C.C. gets the last out: Brewers 3, Cubs 1



The Brewers watch the end of the Mets/Marlins game on the Jumbotron. Go Fish!


Marlins win, Brewers celebrate!


It's nice to take a break from the election.

Debate Effects

The first of three presidential debates is scheduled for this Friday, the 26th of September. We can expect to see the two campaigns positioning themselves and trying to manipulate (lower) expectations in the coming days, and we can also expect the media to focus on how pivotal the debates might be to determining the contours of the remaining weeks of the campaign. These things will happen because of the high visibility nature of these one-night events and the relatively high stakes if one side or another does a lousy job.

But do debates really have much of an impact on the campaign? Are elections won or lost because of debate performances? Is there much of a debate bump? In Do Campaigns Matter? I argued that debates have relatively limited effects because they occur relatively late in the campaign, when opinions are less pliable, and because--unlike conventions--the information flow is relatively balanced, with both sides making their case.

As the debate period draws near, it is useful to look at data from past debates to get a sense of what to expect this year. In the figures and table below I use Wlezien and Erickson's "poll of polls" to gauge relative support for the two major party candidates during the last sixty days of the campaign. The poll of polls measures the percent of the two-party vote for the incumbent party candidate in the latest available public opinion polls for any given date. These data allow me to examine the flow of public opinion during the fall campaign (figures) and also to measure candidate support in discrete periods of time before and after debates (table).

In the figure below every diamond-shaped dot represents a daily polling observation from the poll of polls; the smoothed line is a lowess estimate intended to give a better representation of the "bump and wiggle" of candidate support; and the vertical lines represent the dates on which debates occurred.

Theses figures provide a visualization of the debate effects and there are a few debates the "look" like they mattered. However, some of these visuals could be a bit deceptive because most of the x-axes only span about ten percentage points and also because some of the debates are closer together than others. The table below summarizes the information from the graphs in a way that gives us a firmer grip on the magnitude of the debate bumps. Here, I compare the polling average from six days prior to the debate to the day of the debate with the average from the seven days following the debate.


These data suggest that the norm is for very little swing in candidate support following debates. Across all thirteen presidential debates the average absolute change in candidate support was 1 percentage point. There are a few notable exceptions, of course. Two that stand out are the second debate in 1992, following which George H.W. Bush lost 2 points, and first debate of 2004, after which George W. bush lost 2.26 points. Other debates with above average ( but still small) vote shifts are the first debate in 1996 and the second debates in 1988 and 2000. Each of these debates has its own story, and I'm sure we can all think of anecdotes to explain the bumps and wiggles. Although the analysis is terribly outdated by now, the debate model from Do Campaigns Matter? came to the profound conclusion that the candidate viewed as having won the debate generally gets a small bump (I told you it was profound).

Focusing on single debate bumps may be obscuring a more general, cumulative effect of debates. The last column in the table shows the change in candidate support from one week prior to the first debate to one week after the final debate. Here we see that the debate period generated a 2.42 point bump for George H. W. Bush in 1988, cost Al Gore 3.52 points in 2000, and cost George W. Bush almost 2 points in 2000. Of these, the 2000 debates stand out as the most important, especially in the context of the closeness of the election. Part of the explanation for Gore's swoon during the debate period is perhaps related to his performance but another important factor was the media meme that emerged as a result of the debates, including open discussions of whether or not Al Gore was a "serial exaggerator" (see Jamieson and Waldman).

So what does this mean for the three presidential debates this year? I don't expect to see large swings in candidate support following the individual debates, barring something really spectacular happening. However, even relatively small shifts in the same direction over the three debates could make this relatively tight race even tighter (if the shifts favor McCain), or could blow it open (if the shifts favor Obama).

Note: I've made a minor change to the title of the table since the original post.

Media and Momentum

In my last post I mentioned that I thought part of the explanation for Obama's meager convention bump was that the Palin announcement on the day after the convention robbed Obama/Biden of much of the post-convention media coverage that might otherwise have benefited them. The more general thought along these lines is that convention bumps are the product of candidates dominating news cycles for a period of several days. This is what I found when I wrote Do Campaigns Matter? and I suspect the same holds true today.

The data below illustrate the ebb and flow of media coverage from early August through mid-September (the dots represent daily observations and the line is the smoothed--lowess--trend estimate). These data do not capture the tone of coverage but rather the extent to which one candidate or the other dominates news coverage. It's possible, of course, that more coverage could be due to negative stories, but it strikes me that the coverage around the conventions is mostly beneficial to the convening party.


There are a couple of important observations to draw from this figure. First, Obama's media presence peaked during the Democratic convention and then bottomed out immediately after the convention's end. It certainly seemed at the time that Obama and Biden disappeared following the the convention, and these data show that that was the case. On the Republican side McCain enjoyed a media advantage during the Republican convention and generally has managed to maintain a slight edge in coverage in the weeks since the convention. Although McCain's advantage is slight, he was able to extend the level of coverage from the convention for a longer period than was Obama.

The other interesting finding here is the degree to which Obama dominated media coverage in the pre-convention period. During the period from August 1 to August 24 Obama averaged 57.4% of all candidate news mentions; and with the exception of a single day (August 15) Obama led in news mentions during the entire period. This lends some credence to the McCain campaign's refrain from earlier this summer that the media were providing much more coverage of Obama than McCain. However, whereas Obama clearly drew more coverage in the pre-convention period, the post-convention period is more balanced.

Note: the data for media coverage come from perspctv.com, a web-based news and information aggregater.

Convention Bump Recap

Now that we have a little distance from the end of the Republican convention it's possible to compare the actual convention bumps to those I had anticipated. As described in a previous post, the bump is calculated using Campbell, Cherry, and Wink's (1992) method: the candidate's average share of the two-party vote in trial-heat polls conducted six days to two weeks prior to the start of the convention is subtracted from his share of the two-party vote in polls conducted during the seven days following the close of the convention. It is important to bear in mind that this is only one way to gauge convention bumps and it does not say anything about the duration of the bumps.

Based data from polls* collected from August 29 to Septemeber 3, I calculate that Obama received a 1.8 percentage point convention bump. This is a trivial increase, running below both the average historical (1964-2004) bump of 5.9 points, as well as my prediction of a 5.5 point bump. Using polls collected from September 5 to September 11, I put the McCain bump at a slightly higher 2.1 points, still relatively modest by historical standards but slightly higher than my prediction of 1.4 points. Both candidates enjoyed larger bumps in the first few days after the conventions but support averaged out to much smaller bumps by the end of the first week after the conventions.

While the error in predicting the McCain bump is relatively small, the Obama prediction error is appreciably larger. The most likely explanation lies in the fact is that the Republican VP announcement took place the day after the Democratic convention and the Republican convention started (sort of--Gustav) the following Monday. In effect, Obama/Biden disappeared from media coverage the day after the convention, in contrast to a period of post-convention media coverage that typically (I think) tilts toward the convening party. It was also probably a mistake to wait until the Saturday prior to the convention to announce the Biden selection. Had this been announced on Wednesday or Thursday of the week before the convention, the Democrats would have gotten a couple of extra days of positive and abundant coverage heading into the convention.

One final explanation that has to be considered is that the upper limit of Obama's potential support could be lower than anticipated. Recall that my "bump" model controls for how far ahead or behind the predicted outcome the candidate is running prior to the convention. Based on presidential approval, aggregated satisfaction with person finances, and an open seat contest, my model predicts Obama winning with 55.7% of the two-party vote (seems a bit of a stretch at this point, but lets wait until election day to see how close it is). To the extent that this prediction is higher than his real potential vote, the bump model will have over estimated his bump. At the same time, though, the bump model should have underestimated McCain's bump by a similar margin, which it didn't. In the end, I suspect that all three factors may explain Obama's under-performance.

*Polling data are taken from Pollster.com , realclearpolitics.com
, and pollingreport.com

McCain Bump Prediction

Based on the analysis presented here, my prediction for John McCain's convention bump is 1.4 percentage points. Based on the way I measure bumps, McCain's share of the two-party vote in trial-heat polls conducted during the week after the convention should be just 1.4 points higher than it is now. The primary reason for this small prediction is that McCain is currently running ahead of where he "should" be and, hence, is unlikely to gain much more ground.

Right now, there is no sign of a convention bump for Obama, though it is still too early to tell. I'll provide a recap of both candidates' convention bumps in a couple of weeks.

Nominating Conventions--Who's Watching?

This post is brought to you by Joe Cera, a graduate student in political science at UWM.

One of the central functions conventions serve is as important group identity rallying points. They are party-themed affairs, and they accentuate partisan concerns and partisan thinking. And potential voters expose themselves to information disseminated during the conventions based on their pre-existing partisan affiliations. In other words, to a certain extent conventions are “preaching to the choir.”

Consider the pattern of voluntary exposure to the 2004 convention speeches of Bush and Kerry amongst different partisan groups:

These viewing patterns show who the convention audience is—largely the convening party's supporters and political independents. And this viewing audience helps explain the messages we'll be hearing in the next two weeks. So if you hear the candidates or their surrogates talking about reaching out to the other party, you know that the other party is generally not watching.

Obama Bump Prediction

As promised last week, I have a prediction for Obama's bump in the polls from the upcoming Democratic convention.

Based on his current standing in the polls (4.75 points behind his expected--based on my super-duper forecasting model--vote share), the fact that the convention starts seven days before the Republican convention, and accounting for slight decline in convention bumps over time, the convention bump model predicts a 5.5 percentage point bump in Obama's share of the two-party vote in trial-heat polls. This is slightly smaller than the historical average (5.9 percentage points), in part because the two conventions are closer together than any others during the period analyzed (1964-2004).

The prediction is based on the results of a regression analysis (below) of convention bumps, utilizing the variables described above, as well as a dummy variable for the 1968 and 1972 Democratic conventions (I toyed with labeling this the "lousy convention" control variable).


The model makes substantive sense and fits the data fairly well. Of course, there is still a fair degree of unexplained variation in bumps, as illustrated in the graph below. The smallest error was .45 points (Democrats, 1996), the largest was 5.49 points (Democrats, 1984), and the average absolute error was 2.45 points.


What all of this says is that there is a regular pattern to convention bumps and based on that pattern Obama should get a bump of around 5.5 percentage points. However, it is also clear that it is not at all unusual for candidate to over- or under-perform by a few percentage points.

Of course, the fact that McCain's convention begins just a few days after Obama's ends could have an effect on the magnitude of the post-convention bump. While the model controls for the number of days between the beginning of the conventions, there are no cases in the data set where the conventions are this close together. In other words I am predicting outside the range of the data, always a dicey enterprise.

I should have a McCain Bump prediction about this time next week.

V.P. Pick

...no data

...no model

...just a guess

Hillary Clinton

Why? Well, all the party unity stuff, of course. But also, since she's generally not on many of the beltway lists, picking her would be a bold and unexpected move. And despite her long tenure as First Lady and Senator, the ticket would "look" more like a "change" ticket than one with Biden/Kane/Bayh at the number two spot. Of course, Obama/Sebelius would accomplish that as well (no, I'm not hedging--unless of course he picks Sebelius).

I'm very familiar with all the reasons for not picking Clinton, but I think there's a huge upside as well.

I'll also go out on a limb and predict that the pick will be announced just prior to prime time news broadcasts tomorrow (Wednesday). Okay, maybe Thursday (hedge). Waiting any longer would be silly.

Feel free to leave your pick in the comments section.

Convention Bumps

Many years ago, when I was writing Do Campaigns Matter?, I had a lot of fun working on the chapter on conventions. As the 2008 conventions draw near, I thought it would be useful to update some of the earlier analysis to get a sense of how big a boost the candidates can expect from their conventions.

The figure below presents the magnitude of the convention "bumps" from 1964 to 2004. The bumps are calculated using Campbell, Cherry, and Wink's (1992) measure: the candidate's average share of the two-party vote in trial-heat polls conducted six days to two weeks prior to the start of the convention is subtracted from his share of the two-party vote in polls conducted during the seven days following the close of the convention. The change in poll standing is the the convention bump. I'm sure there are a lot of plausible alternative means to capturing convention bumps but I assume they would all tell a very similar story. Anyway, here are the data (Republicans in red, Democrats in Blue):

(Click on image for a clearer picture)

The first thing to note is that there is a lot of variation in convention bumps. Fortunately, as I showed in Do Campaigns Matter?, there is a systematic component to that variation. Two things in particular seem to drive the size of the bumps. First, candidates who are running ahead of where they "should" be (based on the expected election outcome) tend to get smaller bumps, and those running behind their expected level of support get larger bumps. In this way, the conventions help bring the public closer to the expected outcome and help to make elections more predictable. The perfect example of this phenomenon is the 1964 conventions. Goldwater got a huge bump, in part because he was running 16 points behind his expected vote share, and Johnson got no bump, in part because he was running 6 points above his expected vote share. Likewise, Al Gore was running well behind his expected level of support in 2000 and got a substantial bump, while George W. Bush was running ahead of his expected level in the same year and received a rather modest bump.

The other key factor is the timing of the convention. The earliest convention tends to get a bigger bump, and there is some evidence that going appreciably earlier exacerbates this effect. One reason for this is that the first convention is held by the out-party, whose candidate generally less familiar to the general public. Another possible reason is that opinions are generally less well-formed early in the campaign and may be more easily shaped by the first convention. In the data presented above, the first convention bump was the larger than the second in seven election cycles, smaller in three (1988, 200, 2004), and essentially tied in one (1980).

A couple other considerations. It should be clear that the magnitude of the convention bump is not a great predictor of election outcomes. For instance, Goldwater got a huge bump and went on to a miserable defeat, Nixon got a huge bump and narrowly won in 1968, and Reagan got a very small bump and still won the 1984 election by a wide margin. My sense is that the size of the bump doesn't matter so much as whether the candidate gets the bump he is expected to get.

Finally, each of these conventions has it's own story, and it is easy to come up with plausible explanations for why candidates did better or worse than might have been expected. For instance, the 1968 and 1972 Democratic conventions were disasters, and this is reflected in the convention bumps (or dip, in 1972). Also, John Kerry's minuscule bump in 2004 might be attributed to the fact that he announced his vice-presidential choice in early July rather than during the week prior to the convention (which is what most candidates do), or to the fact that the Swift Boat ads were on the air within days of the end of the convention. At the same time, we also have to consider that Kerry was leading Bush in the polls prior to the convention, so he was unlikely to increase his poll standing very much.

Even with these sort of stories, there is a pattern to convention bumps that should be a useful guide to what to expect later this month.

In about one week, I'll have the final pre-convention polling numbers for the Democratic convention. At that time I'll provide a final, precise prediction for the Obama's bump, along with the regression model (which incorporates the data presented above) used to produce the prediction. In the meantime, given where the candidates stand today, assuming the general pattern in the data holds up, all indicators suggest a substantial (though not huge) bump for Obama and a somewhat smaller bump for McCain.

Post-Kaiser-Harvard Poll

The Washington Post is reporting the results of the Post-Kaiser-Harvard Survey of Low-Wage Workers, which shows that working class voters--even white working class voters--are tilting decidedly in favor of Obama over McCain. Besides the fact that the Post is reporting poll results that are almost a month old, the other thing that struck me as odd about this survey is that it is probably a good example of question-order contamination and economic priming. The first thirty-five questions focused on many different aspects of personal and national economic attitudes, and then the thirty-sixth question asked respondents how they would vote if the election were held today. It's hard to imagine that the emphasis on economic questions throughout the survey did not have an impact on how respondents answered the vote question.

Imagine if the vote question had been preceded by thirty-five questions that focused on terrorism and foreign policy. My guess (I have no data) is that responses to the vote question would have reflected greater weight given to foreign policy attitudes and Obama's margin would have been much narrower.

The upshot is that Obama supporters who were excited by these poll results might want to temper that excitement just a bit. Also, poll consumers should always read the questionnaire.

Update: I just saw that Mark Blumenthal has a post at Pollster.com making many of the same points.

Are White Working Class Voters a Problem for Democrats?

One of the prominent story lines during the last several primaries was the relative success of Hillary Clinton among white working-class voters. This, of course, raised questions about whether and how Obama could attract the white working-class vote. Although I'm not sure how serious he was, Obama's chief strategist, David Axelrod, said in an interview on NPR in late April that he wasn't worried about Obama's poor showing among white working-class voters because they typically vote Republican.

Data from the 2004 and 2000 exit polls (below) only provide limited support for this this claim.



There are a couple of take away points from these data. First, the last two Democratic presidential candidates have faired poorly among white voters. Kerry only managed to garner 41% of the white vote, while Gore took in 44%. Indeed, the white electorate is not a natural source of Democratic votes; so, taken as a whole, it is true that whites tend to vote Republican.

Perhaps the most important point, however, is that it is only among low-income whites that Democratic candidates have had even a fighting chance in the last two elections. Both Gore and Kerry won among voters with less the $15,000 annual income, and both candidates narrowly lost among voters with annual incomes between $15,000 and $29,000. It wasn't even close in any of the other income groups.

So, yes, the white vote has not been friendly to Democratic candidates of late--but it is the much-discussed white working-class vote (especially the poorest whites) that is most likely to side with Democratic candidates. It is this group of white voters who are probably most easily persuaded to vote Democratic. Writing them off could be a bad strategy.

Forecasting Update

With almost all of the votes (95%) counted in Oregon it is pretty clear that the yesterday's outcomes did not deviate much from what was expected based on pre-primary polling data. The table below summarizes the accuracy of my pre-primary forecasts, which were based on the polling trends at Pollster.com

	Clinton Percent of the Two-Candidate Vote
	Forecast Vote	Actual Vote	Forecast- Actual
Kentucky	68.2%	68.6%	-.4 points
Oregon	43.5%	41.3%	2.2 points

The model nailed the Kentucky outcome and over-predicted Clinton's Oregon vote by 2.2 percentage points.

KY and OR Forecasts

I've got jury duty in the morning so I'm posting my Kentucky and Oregon forecasts now based on the 5/19 poll trend summary from pollster.com . These forecasts are based on estimates generated from the same forecasting model used before, but updated to include data from North Carolina and Indiana.

The projected percent of the the two-candidate vote going to Clinton is:

Kentucky 68.2%

Oregon 43.5%

The model missed Indiana and North Carolina by 2.1 and 4.0 points, respectively, so it will be interesting to see how it does tomorrow.

Indiana and North Carolina Forecasts

Forecast of Clinton vote: 52.7% in Indiana and 46.7% in North Carolina.

Details of Model:

To reiterate, the the model is based on the relationship between Charles Franklin's final, pre-election estimates of poll standing (in states where there were enough polls to generate estimates, and excluding Florida) and actual votes cast. I also include a control variable for whether the final poll figure included an estimate for John Edwards' share of the vote.

I've updated the model to include data from the Pennsylvania primary, which has since occurred, as was well as from Wisconsin, which was inexplicably and inadvertently excluded from the earlier estimates.

The new estimates, which are very similar to those originally reported, are presented below.

	b	s.e.	sig.(one-tailed)
Hillary%	.964	.099	.000
Edwards (0,1)	-2.85	1.94	.078
Constant	2.329	4.589	.309
N=24 R-squared =.85

Franklin's estimates of Clinton's percent of the two-party vote (as of 11:45 a.m. CST in this case) are 52.26 % in Indiana and 46.03% in North Carolina. Plugging those numbers into the model produces the forecasts of 52.7% and 46.7% of the vote for Clinton in Indiana and North Carolina, respectively.

Forecast for PA

So I'm sitting here wondering how things will turn out tonight in Pennsylvania and it occurs to me that the wealth of data at Pollster.com might provide some insights. To that end, I've developed a simple little forecasting model for the Pennsylvania primary based on the relationship between Charles Franklin's final, pre-election estimates of poll standing (in states where there were enough polls to generate estimates, and excluding Florida) and actual votes cast. As the figure below illustrates, there have been a few hiccups along the way but the actual results have tracked fairly well with the final poll estimates:
Adding a control variable for whether support for whether Edwards was in the final poll estimate (the negative coefficient suggests that Clinton did about four points worse than expected in the polls when Edwards was in the mix), we get the following model for forecasting Clinton's share of the two-candidate vote in Pennsylvania:



Plugging in the final estimate of her share of the two-Candidate vote (53.58%) we get a predicted Clinton vote of 54.8%, with a 95% confidence interval ranging from 50.97% to 58.63%. Anything outside this range would be very surprising.

Reminder: this model does not include state for which pre-election polls were too sparse to create estimates. Also, including Florida in the analysis changes to prediction only slightly (54.6%).

September Polls

As I mentioned last week, Jay DeSart and I have done some work on using September state-wide trial-heat polls to predict presidential election outcomes in the states. We use the Democratic candidate's share of the two-party vote in state-level September polls (averaged across publicly available polls) , as well as a lagged vote variable to predict state-level outcomes. While the lagged vote variable is important to our model, most of the predictive power comes from the September poll average. Some of the data on September polls are presented below.

First, let's look at some scatterplots of the relationship between September polls on November votes from 1992 to 2004:

Just so-so. No doubt one of the problems with September poll accuracy in 1992 was Perot's entry into the race in October. Even with this, however, point estimates from September polls called the correct winner in 39 of 50 states.

Better--just 6 errant calls (one of the two tied poll results was allocated as a "correct" prediction and the other as "incorrect").

Still good--stronger correlation and six errant calls.

Even better--just two errant calls (WI, NH)

The overall accuracy of September polls from 1992 to 2004 (below) is pretty impressive. The September poll average called the wrong winner in only 25 of the 200 election outcomes. And if you toss out 1992 on the basis of Perot's October candidacy, the polls were "wrong" in only 14 of the remaining 150 cases (9.3%).


One interesting thing to note from last week's post, though, is that in 2004 the correlation between earlier (May and June) state polls and the eventual outcomes was almost as strong as the correlation between the September polls the eventual outcome. The big caveat, however, is that the May and June polls only included results for23 and 21 states, respectively.

How Well Do Early Polls Predict Election Outcomes?

SurveyUSA created quite a stir last month when they released the results of their fifty-state general election trial-heat survey. Over at Pollster.com Charles Franklin and Mark Blumenthal provided a nice breakdown of the results, categorizing states "strong" or "leaning" for one or the other candidates, or as a "toss-up," and concluded that these early results gave a slight advantage to both Obama and Clinton in their match ups with McCain. Bob Erikson and Karl Sigman added to the mix with their teched-up, simulation-based analysis of the same data, reaching a very similar conclusion.

This is all great fun, and poll junkies (I include myself) love to see this type of analysis. But all of this assumes that early polls are good indicators of what will eventually happen on election day. Are they?

Jay DeSart and I have done some work on using state polls to predict presidential election outcomes, and it's quite clear that polls taken in September are good predictors of the state outcomes in November. But how well do statewide polls from spring of the election year predict the eventual outcomes? I thought it might be useful to look at data from 2004 to provide some sense of how much stock we should put in early polls.

In the figures below, I use statewide presidential trial-heat polls from March through June (polls averaged by state and month) to calculate John Kerry's expected percent of the two-party vote and then plot it against the actual two-party vote for Kerry in the November election. Note that there were no spring polls in many states in 2004 (no Survey USA fifty-state poll, for instance) so none of the plots include all 50 states.

Let's start with March, since this is closest to the timing of the2008 Survey USA poll (late February).

As expected, there is a strong, positive relationship between March polls and November votes in the states that had polling results. But I wouldn't exactly describe the data points as tightly clustered, and the point estimates called the wrong winner in 4 (Michigan, Wisconsin, New Hampshire, and Pennsylvania) of the 11 states in which one candidate held a polling advantage (two other states, Ohio and West Virgina, were tied). It is worth noting that in each of these misfires both the polling margin and the eventual margin of victory were fairly narrow.

Results for April (N=21), May(N=23), and June (N=21) are posted below.




Two take-away points here. First, the correlation between statewide polls and the eventual election outcome grew stronger as the 2004 campaign progressed. Obvious enough, I suppose.

Second, when the polling margin was fairly narrow the outcome was truly up in the air. In fact, across all four months the poll result called the wrong winner in 17 of the 36 cases in which Kerry's share of the two-party vote in trial-heat polls was between 47% and 53% (this excludes two case in which the poll result was tied). These results suggest that we should take the term "toss-up" very seriously. At the same time, the poll result was wrong in only 3 of the 44 cases in which Kerry's poll margin was outside this range.

So what are the implications of this for how we should view the early Survey USA results for 2008? Assuming the data from 2004 provide a reasonable basis for speculating, I expect most of the "strong" (as categorized by Franklin and Blumenthal) states to stay in their candidate's camp. But I would not assign "toss-up" or "leaning" states to either candidate with much confidence. One exception to this would be states such as South Carolina (lean McCain) and Massachusetts (lean Obama), whose partisan histories argue in favor of greater confidence.

Update: Per Mark Blumenthal's suggestion, here are the graphs again, except with the same y-axis. Visually, this seems to make the most difference in the impression given by the March data.

Where is the Political Middle Voting this Year?

As I pointed out in a recent post, the idea that there has been a surge in independent voters in the Democratic primaries is not supported by exit poll data. I thought it would be interesting, though, to compare the Republican and Democratic primary electorates to see which party is doing the best job of reaching out to the political middle.

Let's start by comparing the partisan complexion of the primary electorates.


The party electorates are mirror images of each other. About 76% of the votes in each of the parties' primaries have come from their own partisans, 20% from independents and around 4% from the other party. The bottom line is that neither party is doing a better job attracting independents or rival partisans. Once again, this finding is a bit at odds with the common perception that the Democrats have been more successful at drawing independent voters.

But party identification is just one way of slicing the electorate. What about the ideological middle? Which party has been most successful at reaching out across ideological lines?

Here, the picture is much less balanced. The Republican primaries are dominated much more by conservatives than the Democratic primaries are by liberals; Democrats have done a better job attracting moderates than Republicans have; and Democrats have had somewhat more successful attracting conservatives than the Republicans have had attracting liberals.

So the general picture is that both parties' primaries have been dominated by their respective partisans, but the Democratic primaries have been more ideologically heterogeneous than the Republican primaries.

Does this pattern auger well for the Democratic nominee, whomever s/he might be, in the general election? I suppose it does if it reflects the breadth of appeal the nominee will have to the November electorate--but that might depend on who the nominee is.

Note: These data are based on averages across those states for which exit poll data are available.

Are White Voters Rallying for Hillary?

Josh Marshall has a post in which he responds to some recent chatter about whether Obama is doing increasingly worse among white voters. It's a busy weekend but I thought I would take a few minutes and bring some additional data to bear on this topic.

The data posted below follow up on my earlier posts on "gaps" in the Democratic primary electorate. First, let's look at the voting patterns among white voters across the states, ordered by date of primary.



Is there anything here to suggest that whites are rallying around Clinton? To paraphrase my man, Jeeves, "it would not appear to present itself as such." There is a bit of jumping around but there is no clear temporal pattern (remember,the states are ordered by date of primary or caucus).

The figure below breaks the data down by broader time periods.



There is still no evidence of a recent change in the voting patterns of white voters. About all you can say based on these data is that Clinton has done somewhat better among white voters in states holding contests on or after super Tuesday than in earlier states. But that's about it.

Those Fickle Female Voters!

(Note: This post has been updated with a data correction)

Joe Klein has a column today in which he lays at least part of the explanation for Hillary Clinton's wins this week on the response of female voters to the now SNL skit about press coverage of the campaign.

According to Klein:

"A feminine fury was abroad in the land; on March 4, women represented a staggering 59% and 57% of the Democratic electorates in Ohio and Texas, respectively."

So once again, similar to the the alleged feminine reaction to the Diner Moment, it was the response of Fickle Female electorate to surface-level issues that turned things around for Senator Clinton. In effect, the gals rallied around the gal.

I have no idea if women responded differently to the SNL skit than men did, but I can say that there was nothing spectacular about female turnout in the March 4 primaries.

The figure below examines the female composition of the March 4 primaries, relative to other primary and caucus states for which exit poll data are available:


Here we see that the female percent of the electorate in March 4 states (pink bars--I couldn't resist) was, on average, well, slightly below average (mean across states = 58.3%). So the "staggering" turnout of women on March 4 was anything but that.

And what about the assumption that Clinton's fortunes are tied to the relative turnout of female voters? The general logic is that if more women than men turn out to vote, Clinton's vote share will increase. This makes sense, given the nature of the gender gap in candidate preference this primary season.

However, the figure below shows that there is a weak and NEGATIVE (r=-.34) relationship between female percent of the electorate and Clinton's percent of the Obama/Clinton vote (Note: tossing out various home states and Florida doesn't change the picture very much).

In fact, a couple of Clinton's strongest showings were in states like Oklahoma and California where the the female share of the electorate was well below average.

Of course, this is just a simple bivariate snapshot, and it's possible that with a series of appropriate control variables, a more theoretically pleasing pattern would emerge. But these data certainly blow a hole in the idea that all Senator Clinton has to do is turn out the female vote.