Updating my posts analyzing the correlates of Obama support at the state level, here are some new graphs using the currently available numbers.

In a linear regression predicting Obama’s proportion of the two-candidate vote (i.e. his vote plus Hillary’s) in contested non-home states (i.e. not FL, MI, AR, NY, or IL), the statistically significant factors are black population (+), Democratic presidential vote (-), having a caucus (+), and levels of state education (+), while Hispanic population, state population (logged), and Southern Baptist population are not, which is largely consistent with my last post. Here are the relevant graphs:

Also, though these factors aren’t statistically significant in a linear regression using the same predictors listed above, the white vote for Obama continues to be significantly negatively correlated with a state’s black population and Southern Baptist population (the latter has been suggested as a proxy for “Southernness”):

(Notes: In the absence of a clear way to handle the hybrid Texas primary-caucus, I’ve counted the primary and caucus there as separate observations. The association between black population and Obama vote is concentrated among primary states and the linear fit above is only for those states. White support for Obama is only available in some states due to missing exit poll data. Finally, the fitted line above for Democratic presidential vote excludes the outlier of Washington, DC.)

Update 3/5 9:30 AM: TNR’s John Judis has a nice analysis of the exit poll data, as does Jay Cost at Real Clear Politics.

Update 3/6 8:48 AM: By request, here is the graph of Obama support by black population with a quadratic fit, which is heavily influenced by his strong showings in caucuses in heavily-white states (by contrast, the plot above is a linear fit for primaries only):