Thursday, April 24, 2008

I messed up

Statistics require attention to detail. Early Tuesday morning, with the Pennsylvania primary coming up, the idea of predicting the outcome based on a simple linear formula built from regression data on race of the states having voted up to that point came to me. I pulled my data from my email account since I wasn't at home at the time. Anyway, I didn't save the equation after I plugged in the Pennsylvania numbers.

When I recreated it at home this morning, I came up with a different result--Hillary winning by 4.4 points, 52.2%-47.8%, instead of 8 points as I'd said earlier. After toying around for a bit, I found the mistake. I took the white and black percentages as a total of all voters, and then multiplied it by the white+black total (.96), to come up with Obama's share. The last step isn't needed. So the polls were more accurate.

I'm really sorry for the carelessness. It's why doing things at home without pressing time constraints is a much better way to go (that's where I almost always do my blogging).

Hillary's better performance than the crude equation predicted likely comes from news about Obama over the last several weeks being worse for him than news about Hillary has been for her. Also, most of those sitting on the fence going into the primary are now voting for Hillary. Before March 4, they'd been voting for Obama. The thrust of the post remains the same--demographics are an anchor, while all the other things that go on comprise the wind, shifting the electoral boat back and forth along its anchor-restricted range at the surface.

The equation used:

Barack's total share = ((((46+[black percentage of voters]*-.28436)*[white percentage of voters])+([black percentage of voters]*.84))+(.33*[Hispanic percentage of voters]))/[1-percentage of remaining voters]


Fat Knowledge said...

Hillary's better performance than the crude equation predicted likely comes from news about Obama over the last several weeks being worse for him than news about Hillary has been for her.

Maybe, but I still think that there has been very little momentum in this race and the results can be explained almost all by demographics. I wonder if you added income and age to your regression if it wouldn't have come out closer to the actual result than the polls.

Or maybe Hillary's increase is due to the "Limbaugh Democrats" who are just voting for her to drag this thing out and give the Republicans an advantage in the fall.

Audacious Epigone said...


Pennsylvania is one of the oldest states in the country, so that alone probably accounts for a couple of percentage points. I'm just speculating on the margins, but I agree with you on the larger point.

Limbaugh's dubbing it "Operation Chaos". In open states, Obama has been taking the Republican vote. In Pennsylvania, it split 50-50 (among former Republicans who presumably registered as Democrats a month ago). So it might be having some effect, but at only 5% of the voting pool, not much.

Al Fin said...

I remember all the good New Englanders back in 2000 bragging about how they "crossed over" to vote for John McCain against Bush in the primaries.

Funny how resentful the dhims get when repos do the same back to them!

As a small "l" libertarian, I can't tell you how many times I've wished to blow up the IRS buildings in D.C.

Audacious Epigone said...


If memory serves, didn't the Kos guy encourage Democrats to do the same thing in Minnesota when Romney was still alive?

If states do not close their primaries, some amount of this political subterfuge is going to happen. Complaining about it is silly.

Anonymous said...

Assuming she gets the Dem. nomination (just for fun), do you think Hillary can win the Presidency without the black vote?

I think she can, but I'd like someone who has crunched the numbers to deal with this.

What say you?

Audacious Epigone said...


I doubt it. Hard to see how any Democratic candidate can at this point in time. Bush beat Kerry by 13 million votes among whites. That's more than the entire black electorate (which totalled about 11 million). Kerry beat Bush among blacks by 8.5 million votes. Thus if Hispanics and Asians voted in the same way in November as they did in '04, Hillary would have to affect something like an 11 million vote swing among whites to win without the black vote.

I'm going to compile the demographic profile that selected the GOP and Democratic nominees once the Democratic side is wrapped up. Maybe comparing the parties up to the point that Romney dropped out will give some idea of where most whites are (or were at that point in time).

It's moot though, since I can't imagine Hillary doesn't pick Obama for VP if she somehow pulls off the nomination. Even if she doesn't, she'll still get most of the black vote.