14 Points

If you share my fascination with the Democratic Primary campaign, than you have probably wondered to what extent the pundits are right when they claim that Senator Clinton wins white states, with working class voters, especially if they are older and Hispanic. You may also wonder whether or not Senator Obama really does do better in states with higher shares of college graduates, young people, and blacks. I decided to find out for myself. So I entered Obama’s margin of victory (or defeat) over Clinton for each state that has voted, using primary data from ABC news and demographic data from the Census Bureau.

At first look, the pundits are basically right. Obama’s margin of victory is negatively and significantly associated with states with a high percentage of whites, Hispanics, women, and people older than 60. He does extremely well in states with high shares of whites who have attended college (the marginal effect was 2.8, with a s.d of .08 implying a 44% point swing from low to high). The only thing the pundits are wrong about is the working class part. There was no significant relationship between median earnings or even the share of households earning less than $50,000 (2006 dollars) and Obama’s victories. I also went beyond the usual demographic data. While reading Robert Putnam’s classic Bowling Alone (summarized at link) it struck me that his American map of the distribution of social capital in 2000 shared much in common with the democratic primary map of Obama’s victories, so I decided to test this formally. The Putnam measure (click for survey questions) of social capital is based on trust and the extent to which one participates in civic life in one’s community.

I used both Putnam’s measure of social capital and the state’s homeownership rate, which significantly predicts social capital (see Glaeser, Laibson, and Sacerdote 2002) as proxies for civic-orientation. Homeownership apparently gives people a larger stake in their community and seems to make them more committed or more able to foster cooperative relationships. The results were dramatic if I included all the other variables except the college share (which is highly correlated with social capital and therefore absorbs some of its impact). Moving from a low level of social capital to a high level (or 1 standard deviation below the mean to 1 s.d. above the mean) increases the predicted margin of Obama’s lead over Clinton by 14 points. So, if both measured move from low to high, Obama gains 28 points in the average state. Controlling for the white college share makes the social capital measure insignificant (t-stat of -.76), but doesn’t affect the homeownership effect (t-stat of 2.10). The white share is also now statistically insignificant.

Fig. 1 shows the extent to which white states vote against Obama. Fig. 2, however throws in a twist. I multiplied the share of whites with the social capital measure. This interaction significantly predicts Obama victories. That is Obama wins in white states where trust levels are high. Specifically, in states where Putnam’s social capital measure is above 8.7, Obama is predicted to win -if all other variables are held at their average level. These figures controlled for the following percentages: households above 60, women, Hispanics, whites, social capital, homeownership rates, and median earnings (or household income below 50K). All of those variables remained significant (homeownership had a t-stat of 2.33). These results were also robust to controlling for the white college share, which only lowered the impact of the interaction term slightly (t-stat dropped from 1.99 to 1.84).

Fig 3. shows this another way. The downward curve between white share and Obama’s margin of victory declines until about 77% white and then goes up. This can be explained by the social capital measure. The average state score on social capital was 5.35, but it was much higher in white states where Obama defeated Clinton by more than his average margin (i.e 11 points). It was 7.59 in those states, where the white share was over 77%. There were 17 of them -a third of all states. That is a major coalition of white support, especially since five have yet to vote.

Figure 4 pulls everything together and focuses on homeownership rates, which was the most robust predictor of Obama’s success. As with the other figures, the states closet to the lines best fit the model’s predictions. It suggests that Obama lost NY because, after controlling for social capital, incomes, and demographics, it has a low homeownership rate (that Clinton is senator there was probably very important too); it suggests that he won Minnesota and Delaware because their homeownership rates are quite high, given their other characteristics.

We can also use this model to predict future primaries. The model with all the demographic, economic, and education variables, Putnam’s measure, the homeownership rates, and my social capital interaction yields the following predictions: Hillary will win in West Virginia by 42 points, Kentucky by 22 points. Obama will win Oregon by 16 points, South Dakota by 18, and Montana by 29. But this model is too generous because I used Florida and Michigan. If I take them out, the importance of homeownership becomes enormous (t-stat=4.83) and Clinton’s margins in WV and KY shrink to 36 and 21 respectively, while Obama’s margin’s go up everywhere else. Those may seem like extreme predictions (especially in Clinton’s favor), but according to the latest Real Clear Politics data, they are accurate. Those polls have Clinton up by 29 in WV, 34 in KY, and Obama up 12 in Oregon (MO and ND weren’t surveyed). If these predictions are even close, Clinton simply won’t be able to close the gap. In other words, Clinton should concede.