Paper Finds that Bradley/Wilder Effect Has Disappeared With Crime/Poverty issues

[Lunar]

http://people.iq.harvard.edu/~dhopkins/wilder13.pdf

Marc Ambinder's summary:
Hopkins looked at all senatorial and gubernatorial races that featured a woman or an African-American candidate from 1989 to 2006 -- a total of 133 races. For each, he found at least one poll released within a month of Election Day, enabling him to measure the gap between a candidate's polling and performance.

Hopkins finds some evidence that African-American candidates suffered from something resembling a Wilder effect before 1996, but since then, the effect seems to have disappeared.

This becomes the key finding of Hopkins's study: The Wilder effect is not a durable phenomenon. Rather, it is dependent on particular political conditions.

His theory is that when racially charged issues like welfare and crime dominated the political rhetoric, racial factors affected voting behavior and the Wilder effect asserted itself. But once welfare disappeared as a salient issue in 1996, political discourse was deracialized and race was less of a factor in voters' mind.

Hopkins finds that the salience of racial factors depends on the tone of the national environment, not on the tone of local candidates. He explains that black candidates before 1996 were victim of the Wilder effect whether or not they ran a deracialized campaign; after 1996, white candidates were not able to benefit from that effect even when they attempted to exploit racially charged issues. This also applies to the Democratic primaries of 2008, where Hopkins finds that there was no Wilder effect affecting Obama's performance.

To preempt possible concerns about his study's validity, Hopkins takes a look at alternative explanations for the polling-performance gap. First, he considered whether the Wilder effect only affects African-American candidates or whether it hurts other under-represented groups. Analyzing races that featured a female candidate, he finds that women do not suffer from any Wilder effect - quite the contrary, female candidates on average perform better than their polling indicated.

Second, Hopkins considers the possibility that the polling-performance gap can be attributed to what he calls the "front-runner's fall." Hopkins explains that front-runners' support can be overstated because of their higher name recognition and because of classical regression to the mean, making it necessary to account for such an effect before determining what impact racial bias in Wilder or Dinkins' decline. After running additional tests, Hopkins determines that some of the polling-performance gap can be attributed to a front-runners' fall, but that the Wilder effect is still at play.

In other words, the Wilder effect tends to increase in function of an African-American candidate's initial support. Hopkins argues that this leads to the hype that surrounds the Wilder effect. The candidates that are most associated to that effect - Wilder, Dinkins and Bradley - were all favored to win. That is what got their campaigns so much coverage in the first place and it made their performance gap that much more dramatic - creating a somewhat naïve buzz around the Wilder effect.
http://marcambinder.theatlantic.com/archives/2008/09/has_the_wilder_effect_disappea.php

[Lunar]

Oh, by the way, USelectionatlas.org gets a shout out on page 9!

[Lunar]

Actually, we won’t.  The Bradley Effect cannot be “proven” by one election since there are many reasons polls can be off, including ground game and oversampling bias.

[Lunar]

JJ, ONE affirmative action initiative where exit polls didn’t correlate with the actual result isn’t enough to proved the effect’s continued existence.   ONE election cannot prove or disprove the effect, let alone something completely asymmetrical to what we're talking about.

 How do you explain, as in the original article, the absence of any Bradley Effect (in fact, a slight reverse one if anything) in Harold Ford’s election in Tennessee, a state full of working class whites no less. 

Surely you can see that people would lie about an affirmative action initiative in order to not seem racist but tell the truth about not voting for a black candidate.  Not to mention exit-polling is done IN PERSON - completely different, completely different.

In addition, the author of the article ran 10k simulations of selecting 5 random states from the primaries and in every single one of them, Barack Obama was underpolled not overpolled.

[Lunar]

On the other hand, there is Reverand Wright.

There’s no question that a lot of people won’t vote for Obama because he’s black.  The question is whether they lie to the pollsters to hide their racism or if they feel mentally comfortable openly admitting their vote.  Even if undecideds break for the white candidate, that’s not the Bradley Effect.

As I said, in 2006, Harold Ford was underpolled slightly instead of overpolled.  And that race was highly polarized around racial issues, with possibly the most overtly racist ad of the modern era running (“Harold, call me!”).

[Lunar]

Have you looked?

Do you see the infinite number or problems of using:
a single race, an affirmative action initiative, and person-to-person exit polling

to prove a point about

a larger phenomenon, a well-known multifacted politician with MANY non-racist reasons to dislike him (AA is sort of narrow of a topic), phone-interviews

when

All recent evidence, including parallel phone polling in 33 primaries relevant to this particular black candidate and multiple state-wide senate and governor races says the opposite?  You're grasping onto what, at best, is a tangent example, to disregard perhaps 50 ACTUAL examples. 

Asking people, in person, how they feel about a racial issue like affirmative action might get them to lie when asking them on the phone who they're going to vote for might not.  You do understand the difference, right?

[Lunar]

Do you have any examples when a Bradley effect occurred in a Democratic Primary?  Recently?

Do you have any examples where it didn't?  I'd have to go back before 1996 if this paper is correct and I don't know if I can find primary info from back then easily. 

But fine, let's ignore the Democratic primaries, you're still wrong to cite Michigan 2006 as ANYTHING relevant to the topic at hand.  See above.

[Lunar]

Lynn Swann in Pennsylvania, Kenneth Blackwell in Ohio, and Deval Patrick in Massachusetts.

Oops, they all didn't have a noticeable Bradley Effect either if you look at them holistically.

[Lunar]

Can you stop nitpicking the primary subpoint and actually answer my post please. 

I'll repeat it for you:


Do you see the infinite number or problems of using:
a single race, an affirmative action initiative, and person-to-person exit polling

to prove a point about

a larger phenomenon, a well-known multifacted politician with MANY non-racist reasons to dislike him (AA is sort of narrow of a topic), phone-interviews

when

All recent evidence, including parallel phone polling in 33 primaries relevant to this particular black candidate and multiple state-wide senate and governor races says the opposite?  You're grasping onto what, at best, is a tangent example, to disregard perhaps 50 ACTUAL examples.

Asking people, in person, how they feel about a racial issue like affirmative action might get them to lie when asking them on the phone who they're going to vote for might not.  You do understand the difference, right?

[Lunar]

I'm going to post something lengthy on this tomorrow. 

Looking forward to it.

Obviously the biggest critique of the article IS not a single incidence of person-to-person exit polling on affirmative action.

 Rather, the author's method of collecting polling data is suspect (using the last three polls found on lexisnexus).  Furthermore, a second point would be that a lot of those people who lied during Wilder's race are still alive and voting today.  While the author makes an EXTREMELY good point about the de-racialization of politics (drugs/crime/social welfare/poverty/gangs have been replaced by terrorism/economy/social security), these people might not have gone away.  But, from what limited inferences we can make, a lot of the recent state-wide black politician races have no resulted in obvious overpolling.

If JJ is right, and it's only 1% that remains Bradley, we'll never know, because there are dozens of distinct things that can cause a poll to differ with the actual voting results.

I, personally, find the article a compelling suggestion that the days of the Bradley effect being over are over already.  I think one of the larger reasons why we might not see a Bradley effect in this election is the election's saturation into everyday life exceeds far and above a senate or governor race (especially the national media).  Someone can just turn on O'Reilly or CNN for 30 minutes and get bombarded with reasons to not like Obama besides his race.  The act of lying to a pollster over the phone is something of a psychological issue, the liar does not feel comfortable admitting that he's uncomfortable with the black person but has no other reason to vote for the white person, so he lies.  How could someone who's never voted for a Republican, for example, vote for one in an election where the ideologies do not differ from previous ones?

In 2008, racist voters have many allowances to help them tell the truth to mask their racism.  In the case of the primary, some ex-Hillary supporters can simply become PUMAs, railing against Obama's sexism or elitism to justify voting for the white guy.  Otherwise, there are so many other issues where less-than-informed voters might know where Obama stands while they might not know for their local gubernatorial candidate.  They can say to themselves that Obama has no economic plan, wants sex-ed for Kindergarteners, doesn't support drilling, etc.

In summary, racist voters can more easily find non-racist reasons to oppose the black guy int his election, because it's been such a long media-frenzy (+PUMA providing easy way out).  This means that they have legitimate personal reasons to dislike Obama, even if they reached and argued these reasons to themselves because Obama's black.

But it IS really cool that this site got a call out on page 9.

I suspect if racism has an impact, it's in them undecideds.

[Lunar]

Are there modern white flight-states that are the equivalent of a white-flight suburb?  Northern Virginia to get out of Baltimore/DC?  If you don't want your kids to go to a poor school with minorities, you don't usually fly to Ohio, do you?  Honest curious question.

[Lunar]

But what I meant was that this is mostly inter-state migration, right?  How are they more concentrated in one swing state over another?  Just the states within that "magic Obama death range" of African-American population ~9%-16% during the primaries? Obama won the whitest (except NH, SD) and the blackest states easily, as we know.

But a lot of the areas where we have recently tested the Bradley Effect to find no convincing result have been in these types of states: Swan in PA, Patrick in MA, Ford in TN, and Blackwell in Ohio.

[Lunar]

I'm always hesitant to include Steele for analyzing Obama, because I suspect that the Bradley effect might be more pronounced among Republicans.  Not because they are more racist, but they are more unused to minority leadership.

But yeah, that's the recent data.

[Lunar]

Sam: "that's the recent data" = your post.

I understand why I'd be wrong to remove Steele from the analysis, I just said my hesitancy.  You're right about Maryland, I think what I said was wrong upon consideration .  I'm not sure if I can quite put my finger on it, but it's possible that there could be some difference between the two parties, it's certainly an unusually large variable not held constant.

[Lunar]

Survey USA isn't exactly the gold standard in polling (see: McCain +20 in NC, Obama+11 in IA, etc), the margin of error was 5% as well, so no big deal.

The other recent polls were UCinn (1% under Taft's total), CNN (1% under Taft's total), Survey USA again which was 2% OVER Taft's total.  Considering that polls aren't razor-darts, I don't think there's sufficient data here to conclude any sort of Bradley Effect one way or the other. 

If you average the two most recent Survey USA polls, which were only a week off in that race, you get pretty darn close to the actual result, a point and a half under.  I think Survey USA has more than that in systemic error, personally, but whatevs.

[Lunar]

Not going to quibble on the details, all of the info is out there.  It is odd that out of the two polls on Nov. 6th you chose only the SUSA.  It's a bit unethical to ignore the other poll that taken the same day that showed a 22% margin, and a little shaky to ignore the SUSA taken the weak before that showed a 30% margin.  Actual result 24%.

But ignore that, here's my thought I actually want you to address:

One more caveat - one should be inclined to subtract X, where X is the average (with its own MoE) that SUSA gets on election day polls in non-competitive races in areas with similar demographics, to the closest extent that these numbers exist.  Like, I just took a glance at California '06, because I know that Feinstein had a uncompetitive election with similar numbers, and look and behold, SUSA overestimated the Republican by exactly 5% as well!
http://en.wikipedia.org/wiki/United_States_Senate_election_in_California,_2006#Opinion_polls

Why is it more significant and racially charged when it happens in Ohio?  Same year, same firm, same date, same pollster inaccuracy, similar non-competitive margins.

[Lunar]

Not going to quibble on the details, all of the info is out there.  It is odd that out of the two polls on Nov. 6th you chose only the SUSA.  It's a bit unethical to ignore the other poll that taken the same day that showed a 22% margin, and a little shaky to ignore the SUSA taken the weak before that showed a 30% margin.  Actual result 24%.

 

The second poll was a university poll, and judging of how BRTD feels ab
out them, and I agree to great extent, it isn't very good. 

Now, in 2006, we had five races where a black candidate ran statewide.  In four of them seems a very small under polling for the white candidate (arguably larger in MA) in the later polling.  It's not generally as large as it was in in 1980's, but it's hard to say it's gone.

Two problems:

1.  We don't have a large number of black candidates to look at.

2.  We don't have a number of solid polls to look at. 

That isn't enough to prove the Bradley Effect exists, but it isn't enough to say that it's gone away either.

As I was typing my "quibble" about your shaky decision to single out one poll and drop extremely relevant comparable data right next to it, I knew that JJ was going to ignore the rest of my point to defend that, and sure enough.  C'mon man...

Let me repeat it for you:


But ignore that, here's my thought I actually want you to address:

One more caveat - one should be inclined to subtract X, where X is the average (with its own MoE) that SUSA gets on election day polls in non-competitive races in areas with similar demographics, to the closest extent that these numbers exist.  Like, I just took a glance at California '06, because I know that Feinstein had a uncompetitive election with similar numbers, and look and behold, SUSA overestimated the Republican by exactly 5% as well!
http://en.wikipedia.org/wiki/United_States_Senate_election_in_California,_2006#Opinion_polls

Why is it more significant and racially charged when it happens in Ohio?  Same year, same firm, same date, same pollster inaccuracy, similar non-competitive margins.

[Lunar]

What's weird is that in 2006, of the five races I've seen so far as I type this, Survey USA's election day polls are always off pro-Democrat by EXACTLY 5%four times and 6% once.  Update: this was not the case for AZ Gov.  But that is weird and I’m suspicious of using them to prove a Bradley Effect in that direction, since most of the time they got the exact same bias with white candidates.
 
What I find is that polling often is off.  Is being black the most the most relevant variable to explain why the average of five polls is off between those four races?  Surely the candidates positions, campaign, personality, and other factors would be more rational explanations of cause.

I think, in order to determine whether or not these could be outliers, we should first subtract the absolute value of the average that polls are generally off for such elections.  I also think you are too quick to emphasize occasions where Democrats should have done better due to electoral wins when many Democrats were overpolled that year.  Maybe you should apply the same emphasis to the PA and TN results?

Anyway, here’s a random (stuff I felt like clicking on, with no discrimination) glance other other races with no real significance except to see that the African-American 2006 results, which might have an extremely loose correlation, might not have one compared to generic error, possibly less so than if you were to say that states with the letter “E” are likely to underpoll one party or the other

Florida, Senate 2006

Average of 5 results: +26.8%D, Actual result +22%D (-4.8%D)

Pennsylvania, Senate 2006

Average: +12.4%D, Actual result: +17%D (+4.6% D)

Alaska, Governor 2006 (only 3 recent results)

Average: -6.6%D, Actual: -8.6%D (+2%D)

Missouri, Senate 2006 (averaged two SUSA's into one, tossed out Polimex so I could include Mason Dixon)

Average: +2.9%D, Actual: +2.3D% (-0.6%D)

Arizona, Governor 2006 (only 3 recent polls)

Average: +19.7%D, Actual: +27%D (+7.3%D)

Virginia, Senate 2006

Average: +4.6%D, Actual: +0.6%D (-4%D)

Minnesota, Senate 2006 (average of 5 polls is 17.8%, instead using only Rassmussen and Survey USA since the other ones are crap)
 
Average: +15.5%D Average: Actual: 10%D (-5.5%D)

[Lunar]

random musing: I wonder if there are more ex-Hillary supporters that lie about their vote (or being undecided) right now but will end up voting for Obama than Bradleyeffecters.

[Lunar]

Not all of them are racist

Like I said, the Hillary supporters and long primary are one possible mitigating effect as well.  Racist voters no longer have to lie and can openly vote Republican for the first time in ages because of how sexist and elitist and inexperienced Obama is.

[Lunar]

Well, you don't state why you're voting for who you're voting for (at least not right away, after they've already asked the categorical president question), it's more about what you're telling yourself.  I'm not sure if racist Democrats had an easy opt-out for themselves for Wilder/Bradley to make them different from all of the white Democrats they had vote for prior.

[Lunar]

awww shucks

[Lunar]

Vo-dawg, any comments though on the research itself?  I find it moderately persuasive and it's the first piece of academic work I've seen to cite https://uselectionatlas.org.

[Lunar]

Vo-dawg, any comments though on the research itself?  I find it moderately persuasive and it's the first piece of academic work I've seen to cite https://uselectionatlas.org.

That alone gives it a ton of credibility.

I'd say so!

[Lunar]

one last bump?

Waiting for Sam's second post still.