A different way of looking at poll demographics

We all know that looking at the party breakdown of a state's voters isn't necessarily the best way to figure out how inclined it is to vote Democratic or Republican in a Presidential election. For instance Louisiana and West Virginia are Democratic states as far as party identification goes, but they're not likely to vote for a Democratic Presidential candidate any time soon. Virginia and Alabama are both pretty evenly divided on the party id front, but John McCain won a landslide victory in the latter while Barack Obama took a closer but still solid one in the former.

This chart shows the states we've polled in the last few months from most Democratic to least Democratic, along with the 2008 results:

State	Party ID	2008 Results
West Virginia	D+15	McCain +13
Illinois	D+15	Obama +25
Ohio	D+15	Obama +4
North Carolina	D+14	Obama +1
New Jersey	D+13	Obama +15
Wisconsin	D+10	Obama +14
Louisiana	D+8	McCain +19
Minnesota	D+7	Obama +10
Virginia	D+1	Obama +6
Alabama	R+3	McCain +21

As you can see, not a lot of correlation. So if party ID isn't the best barometer for telling how inclined a state is to vote Democratic or Republican for President, what is?

I think the percentage of voters in a state who identify as conservatives is a pretty good start. Looking at these ten states, there's not an entirely linear relationship between the proportion of voters who describe themselves as conservatives and how they voted last fall, but it certainly comes a lot closer than using the party id measure:

State	% of Conservatives	2008 Results
Louisiana	50	McCain +19
Alabama	49	McCain +21
West Virginia	43	McCain +13
North Carolina	39	Obama +1
Wisconsin	39	Obama +10
Virginia	38	Obama +6
Illinois	35	Obama +15
Minnesota	35	Obama +10
Ohio	34	Obama +4
New Jersey	28	Obama +13

Two closing thoughts:

-People frequently bash poll results they don't like by criticizing the party id breakdown, but it might actually be more relevant to take a closer look at the ideological breakdown.

-What other measures of a state's electorate do you think can better predict Presidential voting patterns than party id breakdowns?