Posted by Jed Lewison on Wed Jan 9, 2008 at 1:24 PM Pacific

There wasn't a Bradley effect in the exit polls, plus an intriguing theory from Slate

Update: Here's the Slate article. I also just reviewed the poll data for pre-NH and pre-IA. In Iowa, Clinton led the pre-caucus polls by a point but came in third; in NH, Obama led by 8 and lost by 2.

There's been a lot of commentary on whether Hillary Clinton's victory can be explained by the Bradley Effect, in which whites tell pollsters they will vote for a black candidate, but when they head to the polls, they don't.

It's very hard to determine the extent to which this took place. Intuitively, it would seem that it did take place.

However, it didn't seem to happen with the exit poll.

According to my analysis of the exit poll data, 38.2% of respondents said they would vote for Hillary, and 36.6% said they would vote for Obama. 16.2% said they would vote for Edwards.

The exit poll therefore underpredicted the vote of each of the candidates.

Apparently, Slate has an argument for why the phone polls may have had a Bradley effect while exit polls didn't: exit poll workers in NH tend to be white, while polling firms are located around the country and employ a higher percentage of black workers. Since racial identity isn't just about skin color, sometimes people perceive some voices to be more "black" than others, and perhaps this where the Bradley effect creeped in.

I haven't looked at data to test this, but I do know that some of the polling firms use automated machines. If those firms had more accurate results, it would support Slate's theory.

Note: I haven't read the Slate article, which was referred to me by dnA at Too Sense.

I arrived at this number by going through all 69 different cross-tabs in the exit poll. For each of the cross-tabs, there are segments. The exit poll shows the portion of the Dem electorate comprised for each those segments and then the portion of that segment that voted for each candidate. By multiplying the percentage of the electorate represented by that segment with the percentage of that segment voting for the candidate in question, and then adding up the totals for each category, and then averaging the results of all 69 categories, I can arrive at pretty good estimate of what their actual reported numbers were.

I downloaded the raw data after polls closed but before (or perhaps just after) the race was called for Hillary.

Net/net -- the exit polls appear to have been right, which raises the question -- why were telephone polls less accurate than exit polls?

There wasn't a Bradley effect in the exit polls, plus an intriguing theory from Slate

Update: Here's the Slate article. I also just reviewed the poll data for pre-NH and pre-IA. In Iowa, Clinton led the pre-caucus polls by a point but came in third; in NH, Obama led by 8 and lost by 2.

There's been a lot of commentary on whether Hillary Clinton's victory can be explained by the Bradley Effect, in which whites tell pollsters they will vote for a black candidate, but when they head to the polls, they don't.

It's very hard to determine the extent to which this took place. Intuitively, it would seem that it did take place.

However, it didn't seem to happen with the exit poll.

According to my analysis of the exit poll data, 38.2% of respondents said they would vote for Hillary, and 36.6% said they would vote for Obama. 16.2% said they would vote for Edwards.

The exit poll therefore underpredicted the vote of each of the candidates.

Apparently, Slate has an argument for why the phone polls may have had a Bradley effect while exit polls didn't: exit poll workers in NH tend to be white, while polling firms are located around the country and employ a higher percentage of black workers. Since racial identity isn't just about skin color, sometimes people perceive some voices to be more "black" than others, and perhaps this where the Bradley effect creeped in.

I haven't looked at data to test this, but I do know that some of the polling firms use automated machines. If those firms had more accurate results, it would support Slate's theory.

Note: I haven't read the Slate article, which was referred to me by dnA at Too Sense.

I arrived at this number by going through all 69 different cross-tabs in the exit poll. For each of the cross-tabs, there are segments. The exit poll shows the portion of the Dem electorate comprised for each those segments and then the portion of that segment that voted for each candidate. By multiplying the percentage of the electorate represented by that segment with the percentage of that segment voting for the candidate in question, and then adding up the totals for each category, and then averaging the results of all 69 categories, I can arrive at pretty good estimate of what their actual reported numbers were.

I downloaded the raw data after polls closed but before (or perhaps just after) the race was called for Hillary.

Net/net -- the exit polls appear to have been right, which raises the question -- why were telephone polls less accurate than exit polls?

The Jed Report Home Page

© Jed Lewison