Social Science Statistics Blog

I have not seen the details of the pre-primary polls, only the Clinton v. Obama spreads. What percentage of the respondents were undecided? CNN mentioned that, based on exit polls, a large share of the women who decided to support HC made that decision the day of the primary. If so, they could not be picked up by the pre-primary surveys.

Also, did these polls report the percentage of potential respondents that declined to be interviewed. I imagine that in NH during primary season the number is quite high.

Perhaps I'm missing something on the "Bradley Effect" scenario, but I don't think it holds water here. If I'm wrong, please correct me.

If an appreciable number of voters lied to pollsters about intending to vote for Obama, then the polls would overstate Obama's vote. However, the polls got Obama's vote exactly correct. Since the polls did not overstate Obama's vote, there is no Bradley Effect.

Bruce

It's hard to say, given the lack of an academic literature defining exactly what the Bradley effect is and if it even existed in Bradley's election, much less this one. One potential objection to the "polls did not overstate Obama's vote" explanation is that it seems to imply all of the undecided voters broke for Clinton in order to lift her above Obama; if that is the case, then there is a problem with the exit polling data, which shows a much closer split between late-deciding voters.

Just to be clear, I'm not saying that there was a "Bradley effect" in New Hampshire, but the idea is out there; Andrew Kohut from Pew has an op-ed in the Times today on precisely this question:
Getting it wrong.

Could you explain to me what you mean with the Bradley effect? Regards

Wouldn't the Bradley affect also apply to Hillary Clinton since she's a woman?

I'm hoping someone can tell me why this analysis is wrong:

Using this data from the NH democratic primaries I regress Clinton's votes on the total number of votes per district, Obama's votes, whether the district's value was a hand- or diebold-count, and the interaction of the latter two.

The interaction is significant and indicates disproportionately higher allocation of votes to Hillary from Diebold machines only in those districts where Obama has a higher proportion of the vote (p=.013).

What does this mean, and if nothing, why?

Very interesting, CHCH. Could you do a similar analysis on the republican results? It would be nice to have a baseline for the variance.

Stuart, that's a great idea.

UPDATE: there was a discrepancy between blackboxvoting.org's files for diebold vs. hand-count NH districts and those maintained by the government. when I redo the analysis with the gov't's numbers there is still a significant interaction, but it goes in the opposite direction. However, the interaction of the quadratic effect of obama votes with the diebold var (controlling for population, age, education, the simple effect of the diebold var, and the simple and quadratic effects of obama votes) goes in the direction I mentioned in my earlier comment.

But this is beyond my comfort-zone for interpreting stats, so I was hoping someone here might be able to help... Let me know if you want a copy of the file.

Would it be possible to doa similar analysis on the republican results?

It is possible - I haven't done it but have heard from others who have that the apparent diebold effect for Romney *can* be eliminated with these covariates, perhaps providing additional confidence in the idea that the Diebold effect is real in the case of clinton/obama.

The candidate with the best Healthcare Reform Policy should win the Presidential election.

Over 46.6 million Americans are uninsured, with 15.9 percent of Americans lacking health coverage.

Census data show that 46.6 million Americans were uninsured in 2005, an increase of 1.3 million from the number of uninsured in 2004 (45.3 million). The percentage who are uninsured rose from 15.6 percent in 2004 to 15.9 percent in 2005. The number of children who are uninsured rose from 7.9 million in 2004 to 8.3 million in 2005.

The term "Bradley effect" comes from a 1982 campaign involving Tom Bradley, the long-time black mayor of L.A. He ran as the Democratic party's candidate for Governor of California against Republican candidate George Deukmejian, who was white. The polls consistently showed Bradley with a lead. However, Bradley narrowly lost the race. Post-election research indicated that a smaller percentage of white voters actually voted for Bradley than polls had predicted, and that voters who had been classified by those polls as "undecided" had gone to Deukmejian in statistically anomalous numbers.

Stuart, that's a great idea.

Very interesting, CHCH.

Stuart, I would also like to see this.

Thanks, this information it was very useful to me.