(No Sunday Screening tonight; if you really need a fix, here is a satire of Greg Mankiw's economics, here is an emergency broadcast system tornado alert from North Carolina in 2004, and here are a dog and a monkey doing sit-ups. That ought to cover my bases, right?)

Ron Brownstein made the cover of Friday's National Journal with an analysis that argues the Democratic electorate is changing. It's a pretty good piece, and I imagine it'll make the usual rounds (Ross Douthat, often an early adopter, has already linked it with enthusiasm). Except there's a severe methodological problem with Brownstein's analysis that renders it basically moot. Here's the argument:

From New Hampshire to California, and from Arizona to Wisconsin, exit polls from this year's contests show the Democratic coalition evolving in clear and consistent ways since the 2004 primaries that nominated John Kerry. The party is growing younger, more affluent, more liberal, and more heavily tilted toward women, Latinos, and African-Americans.

In the 18 states for which exit polls are available from both 2004 and 2008, the share of the Democratic vote cast by young people has risen, often by substantial margins. Voters earning at least $100,000 annually have also increased their representation in every state for which comparisons are available -- again, usually by big margins. Women's share of the vote has grown in 17 of the 18 states (although generally by smaller increments). In 12 of the states, Latinos have cast a larger percentage of votes, as have the voters who consider themselves liberals. African-Americans have boosted their share in 11 of the 18 states... etc.

OK. Brownstein's analysis is based on comparing the electorates of 2004 and 2008; that's completely apples and oranges. 2004 was basically decided after Dean got stomped in Iowa, save John Edwards' late challenge in February -- whereas 2008 is still competitive, and was a pure toss-up until recently. The environments, particularly the turnout levels, are not even remotely similar!

Brownstein seems to think he's accounted for this problem by using relative shares of the vote (rather than absolute numbers). He writes:

The shifts in the Democratic coalition are particularly striking because they are occurring at the same time that party turnout has increased over 2004 in every state that has voted so far. These groups, in other words, are contributing a larger share of a growing pool. Latinos, for instance, increased their share of the Democratic vote in California from 16 percent in 2004 to 30 percent this year, even amid an overall surge in turnout. ... "The context for these shifts is not just a different distribution but a much larger pie, which makes it more impressive," says Geoff Garin, a Democratic pollster unaffiliated in the race.

But this is predicated on the assumption that, in a generic scenario, increased turnout would affect all demographic groups equally. Think for a minute and you'll realize what a silly assumption that is. Some demographics are more likely than others to vote consistently, regardless of the political climate; plus low-information voters (who are more prevalent in particular demographic groups) are likely to turn out in sporadic patterns that have little to do with the political climate or anything else. (I have no idea, BTW, what Geoff Garin is talking about in that quote. It sounds like someone got confused during the interview.)

What are some such groups that don't increase turnout so reliably? Seniors are the most obvious example (they tend to vote with near-perfect regularity); also blue-collar and union workers (who are often turned out to vote by political machines); moderate/"independent" voters, who are usually low-information; and the bottom ends of the education & income scales, which are also heavily low-information. Funny, those are the exact groups that Brownstein claims are losing influence in the party!

Thus the "change in composition" Brownstein's talking about is more illusory than real -- I don't know about the increased share of minority votes, which doesn't track with this model (we can speculate that the identity politics prominent in this race might have to do with it), but the other changes he's pointing out are not functions of realignment. They're functions of turnout inconsistency. And all the implications Brownstein draws from them are consequently hollow.

So when, inevitably, you hear people repeating this canard--

In the Democrats' longtime upscale-downscale divide, these changes are tilting the party away from blue-collar and often gray-haired "beer track" voters toward younger and more affluent "wine track" voters

--remember that, no matter how intuitive it may seem, the evidence does not bear it out and the Democratic Party is not necessarily changing into anything at all.

(Amazingly, this is not the first time I've rapped Ron Brownstein for bad poll analysis -- here's me a year and a half ago accusing him of overvaluing rural voters in the VA and TN senate races, and I believe I was proved at least half right there. He's a gifted and insightful writer, but it seems he really shouldn't be handling polls...)