Yesterday, I wrote a diary examining how accurately polls predicted turnout. In many states such as Alabama, Georgia, and South Carolina, pre-election polling severely underestimated African-American turnout; the Hispanic vote was also underestimated to a lesser degree in many states. I was originally planning to repeat that analysis for age demographics, but that proved difficult because pollsters weren't consistent in how they broke down the different age groups. So instead, I attempted to predict racial turnout on my own by examining exit poll data from 2004.
I reasoned that there should be a linear relationship between the percent of Democratic voters who were of one race and the percent of Democratic primary voters of the same race. I took CNN 2004 exit polling data from each state, and multiplied the African-American vote by the percent who voted for Kerry to get a baseline of the number of Democratic African-American voters; I divided that number by the percentage of Democratic voters who voted for Kerry (I only counted Kerry voters to remove error from the traditionally Democratic southern states, where voters might be registered Democrat but never vote for one.) I repeated the process for Hispanic voters; overall, the exercise required more work than actual analysis.
When I fitted the data to a linear trendline, I was pleasantly surprised to see how accurate it was for the African-American vote; the Hispanic vote was a bit more irregular.
Surprisingly enough, my model for the African-American vote was more accurate than pre-election polls for most states. The obvious outlier there is Virginia, which I think has to do with the fact that it received a massive amount of non-Democratic voters - 30% independent or Republicans; that amount is usually more in the 15-25% range.
I applied my model to Ohio and Texas, and got that of the Texas electorate, 23% would be African-American and 25% Hispanic; Ohio would be 18% African-American and 4.7% Hispanic. In comparison, the latest SurveyUSA poll shows Texas at 18% African-American and 32% Hispanic, while Ohio is 17% African-American and 3% Hispanic.
As shown in the graphs above, Hispanic turnout is quite irregular, but I'll stick with my prediction that it would be 25% in Texas, as that seems to fit in with what the Burnt Orange Report is saying. My model for the African-American vote seems to fit quite neatly for past primaries, so I'll assume that its right as well. I guess we'll see whether my model is more accurate than SurveyUSA in a couple of weeks.
To predict the votes in each state, I used the SurveyUSA poll for the breakdown among white and Hispanic voters, assigning 'other' and 'undecided' votes equally to each candidate, but assumed that Obama would win the African-American vote 85-15, as polls have predicted it inaccurately in the post (see last diary.) This model has Obama winning Texas 51.5-48.5, and losing Ohio 52.2-47.7
One factor left to consider is that the African-American population in Texas has gone up quite a bit since 2004 from Hurricane Katrina refugees. If those people get to the polls, African-American turnout in Texas may be even higher than what I predict.
A few more possible scenarios:
- The Hispanic vote is going to surge up to 32% in Texas, but my projections are right on everything else. If this happens, Obama wins it in a nailbiter by 1%.
- In the week left before the election, Obama gains among the white vote in Texas and Ohio so that he loses 52-48 and 55-45, respectively, while losing 40-60 among the Hispanic vote; my turnout projections remain accurate.
If this happens, Obama wins Texas by 9 points and Ohio by 4. I think that this is the most likely scenario to occur.
For fun, some optimistic scenarios for both candidates:
- Very optimistic Obama scenario (something along the lines of what Kos seems to think for Texas):
Obama crushes Clinton among the African-American vote 90-10 in both states; his effective ground game pushes African-American turnout to 25% in TX and 20% in OH. He ties among the white vote in TX and loses it by 4 points in OH, while losing 45-55 among the Hispanic vote.
If this happens, Obama wins Texas by 17.5% and Ohio by 12.5%.
- Very optimistic Clinton scenario:
Obama's ground game in TX and OH is overhyped and all predictions of increased African-American turnout are wrong - its only 15% in both states. The Hispanic vote does surge to 32% in Texas. Clinton halts Obama's momentum and crushes him among the white vote 60-40 in TX and 65-35 in OH; she wins the Hispanic vote 70-30 and loses the African-American vote 25-75.
If this happens, Clinton wins Texas by 16% and Ohio by 18%.
I did all this work on Excel; if you want to look at the detailed numbers, I took a screenshot of it here.
UPDATE:
I added a graph comparing my model to actual results. There don't seem to be any obvious patterns.
UPDATE II:
SUSA's Texas poll came out
They have Obama AHEAD 49-45. Most of this apparently comes from him narrowing the Latino voter gap to 13 points.
The crosstabs page won't load (probably from way too much traffic), but I'll update this with an analysis when it does.
Update III:
A bit pointless since nobody's reading this anyways, but the crosstabs finally loaded.
It looks like SUSA's backing down from their earlier turnout predictions - they now have African-American turnout at 21% and Hispanic turnout at 28%, in comparison to 18% and 32% from their last poll.
Most significantly, they have Obama losing ground among white voters but gaining among Hispanics. Surprisingly, the Hispanic vote is closer than the white vote.