Monday, July 02, 2007

Atlantic Pundit-Blogger Math Skills Lacking

I want pundits who can look at a very basic statistical report and say something reasonably intelligent about it. Unfortunately, I have two good examples of national political pundits who need a basic reminder. The first is Mark Ambinder as he writes about the Obama polling/fundraising memo. It is a good piece, but there are a couple of significant weaknesses:

That's a record -- 258,000 Americans have never contributed to a campaign this early. No data yet on the average contribution size. That'll be key to determining whether his support comes from working class Dems or elite Dems.

We actually can figure out the averages real easily using previous open source information:

Furthermore we can look back at the 1st Quarter reporting and see what the donor distribution pattern looked like. ABC News did all of the work for me with this report:

Obama received donations from more than 100,000 donors, far surpassing any other candidate, including Clinton (50,000), McCain (45,000), Edwards (40,000) or Romney (32,000)

$6.9 million of Obama's donations came from more than 50,000 donors via the Internet, far surpassing any of his rivals

90 percent of Obama's donations were small donations of $100 or less

The second piece that I have issues with is the Andrew Sullivan argument that the under thirty voting bloc despite massive Democratic self-identification, is actually a conservative friendly voting bloc, and my issue is the margin of error fallacy:

38% believe that going to war in Iraq was the right thing to do, compared with 35% of all adults. If you want to find the most antiwar part of the population, you need to look at senior citizens, not the young.

Does this make the next generation more conservative? On some issues, the answer seems to be yes. Take abortion. The underthirties are marginally more opposed to abortion than their parents: One in four want it made illegal altogether. Thirty-seven per cent of the underthirties are pro-choice, compared with 39% of the population;...

The differences he is citing as an argument that under-30s are more conservative than the general population are miniscule differences (3% in one, 2% in another). The poll was conducted by MTV, New York Times and CBS News. The margin of error was 4%.

This means that any single result has an eight point band (4 on each side) where we have to accept any overlap as being the same between the subgroup (U-30s) and the general population. We can legitimately say that a 3% difference has a higher probability of being different than a 2% difference, but we can not say with a straight face that there is anything statistically valid different between u-30 views and general population views.

No comments: