The Bradley Effect – Selective Memory

[Nym90]

The Bradley Effect – Selective Memory
By V. Lance Tarrance, Jr.

Now that polls indicate Senator Barack Obama is the favorite to win, some analysts predict a racially biased “Bradley Effect” could prevent Obama from winning a majority on November 4th. That is a pernicious canard and is unworthy of 21st century political narratives. I should know. I was there in 1982 at “ground zero” in California when I served George Deukmejian as his general election pollster and as a member of his strategy team when he defeated African-American Democratic California gubernatorial candidate Tom Bradley, not once but twice, in 1982 and again in 1986.

Bradley Effect believers assume that there is an undetectable tendency in the behavior of some white voters who tell pollsters that they are “undecided” when in fact their true preference is to vote against the black candidate. This so-called effect suggests the power or advantage to alter an outcome – a pretty serious charge. This would render poll projections inaccurate (overstating both the number of undecided voters and the African-American candidate’s margin over a white opponent) and create an unaccounted for different outcome. However, it is indeed a “theory in search of data.”

The hype surrounding the Bradley Effect has evolved to where some political pundits believe in 2008 that Obama must win in the national pre-election polls by 6-9 points before he can be assured a victory. That’s absurd. There won’t be a 6-9 point Bradley Effect –- there can’t be, since few national polls show a large enough amount of undecided voters and it's in the undecided column where racism supposedly hides.

The other reason I reject the Bradley Effect in 2008 is because there was not a Bradley Effect in the 1982 California Governor’s race, either. Even though Tom Bradley had been slightly ahead in the polls in 1982, due to sampling error, it was statistically too close to call. For example, the daily Tarrance and Associates tracking polls for the Deukmejian campaign showed the following weekly summations (N=1000 each) during the month of October:


Week of:
Oct.7th Oct. 14th Oct. 21st Oct. 28 Nov. 1

Bradley 49 45 46 45 45

Deukmejian 37 41 41 42 44


It is obvious that this election was closing fast. Yet, Bradley's win was projected by the most prominent public pollster in the state, Mervin Field, who boasted on Election Day that Tom Bradley would defeat George Deukmejian, “making the Los Angeles mayor the first elected black governor in American history” (UPI 11-3-82). The reason for Field's enthusiasm was that his last weekend polling showed a 7-point margin for Bradley, but this was totally at variance from the Tarrance and Associates internal tracking results. Field's own exit polls, on Election Day itself, where voters were questioned after they left the polling places, also predicted a Bradley win. This caused the San Fransisco Chronicle, ignoring the closeness of the election and mixed polling results, to print 170,000 copies of its early morning Wednesday edition under the headline “Bradley Win Projected.”
Also at variance with the Mervin Field exit polls were the NBC and the CBS networks, using both exit polls and actual returns from key precincts, when they declared George Deukmejian the winner and not Tom Bradley the winner. In an AP report, a KNBC newscaster told viewers on Election Night “...half of the polls are wrong and I don't know who's right.” The only thing we know for sure is the election was too close to call, and some of the Election Day projections were right and others (notably Mervin Fields’ projections) were wrong and, unfortunately, most of this explanation because of selective memory has not been carried forward to this day.

The Field Poll inaugurated the speculation that led to the baseless Bradley Effect theory when, after the 1982 election, Field said “race was a factor in the Bradley loss” (AP 11-4-82). Mervin Field cited no data, but only speculated that white conservative voters of both parties were more undecided and that he may have over-represented minority voters in his polling. Thus, the Bradley Effect was born amidst some major polling errors and a confusing array of mixed predictions, hardly a firm foundation to construct a theory.

Even later analysis of the 1982 election revealed the weakness in the Bradley Effect theory as Bradley actually won on election day turnout, but lost the absentee vote so badly that Deukmejian pulled ahead to win. That Bradley won the vote on Election Day would hardly seem to suggest a hidden or last minute anti-black backlash—on the contrary, it suggests how easy it would have been for weekend polls and Election Day exit polls to get it wrong, since the decisive group of voters had largely already voted before the final weekend and never showed up at the polls to answer the questions of exit pollsters.

When Barack Obama lost the 2008 New Hampshire primary after all seven pre-election polls had Obama projected as the winner, the Bradley Effect got a second wind, blown along by a lot of misinformed press speculation asserting that our nation was still suffering from latent racism. A few weeks later, after much analysis of election demographics, and with a more thoughtful examination, it is clear that race was not the determinant that gave Hillary Clinton a surprising victory. In fact, it was a combination of an older brand of feminism, the open party system that encouraged independents to vote in the primary and some Obama campaign hubris that caused the result.

The New Hampshire polling debacle was also eerily familiar to those of us who witnessed first-hand the 1982 California election day errors. A lesson learned from 1982 campaign, but not remembered in 2008, was what a San Francisco Chronicle editor said the day after the 1982 election, “It seemed logical...to project a continued gain for Bradley.” There was never a consensus of data to support this logic. The 2008 New Hampshire update on the so-called Bradley Effect also falls short of proving this false theory of latent racism. Instead, the New Hampshire debacle should be labeled for what it is, the worst polling disaster since “Dewey Beats Truman.”

The Deukmejian campaign tracking polls did not confirm any Bradley Effect and to interject this type of speculation into the 2008 presidential election is not only folly, but insulting to the political maturity of our nation's voters. To allow this theory to continue to persist anymore than 25 years is to damage our democracy, no matter who wins.

[Lunar]

Great post Nym!

It has always been a fact that Bradley might have overpolled in CA for a myriad of reasons outside of people lying to pollsters based on racial sensitivity.  I'm not completely convinced that race is divorced from Wilder and Bradley overpolling, but this was all 22-26 years ago....before I was born actually.

I might do a long thread someday soon about why we won't see a substantive Bradley effect in the November election.  Even if it's 1%, we won't "see" it because of all of the other reasons polls can be wrong, including organization and methodological error.

[Nym90]

Great post Nym!

It has always been a fact that Bradley might have overpolled in CA for a myriad of reasons outside of people lying to pollsters based on racial sensitivity.  I'm not completely convinced that race is divorced from Wilder and Bradley overpolling, but this was all 22-26 years ago....before I was born actually.

I might do a long thread someday soon about why we won't see a substantive Bradley effect in the November election.  Even if it's 1%, we won't "see" it because of all of the other reasons polls can be wrong, including organization and methodological error.

Very good points.

The key here is that the polls were narrowing significantly prior to Election day. Yes, Bradley did have a big lead and ended up losing, but the race was essentially tied going into the election. The polls were not wrong outside the margin of error; they did pick up on the late Deukmeijan surge.

The bottom line is that the sample size of black candidates has been small enough historically that we cannot even prove the effect ever existed, since the overpolling has been on average within the margin of error.

[J. J.]

Nym:

1.  I know of no documented Bradley Effect in any primary, nor, except for a few posters here (and I was not of them) any claim that NH had one in 2008.  NH primaries have been notoriously hard to poll and have been since at least 1984.

2.  The question is not was there a Bradley Effect in 1982 and 1989, but in 2006 statewide races.  The answer appears to be yes, in 3 out 4 races (I can't find data for the fifth race, PA).

As a corollary to the second point, how much?  In those three races it appears to be between 2-4 points (though arguably, one poll was seven points off in MA).  Several of those, IIRC were outside of the MOE.

There is no question that black candidates overpolled in 1982 and 1989 statewide by fairly large numbers 9-10 points.  That does not exist today in any polling, but it does look like it does in some states, to a lesser extent.

[CARLHAYDEN]

Nym,

Good article which omits the mention of one key event in the 1982 California election, i.e. the impact of Proposition 15.

Prop 15 signicantly altered the electorate, with the result of Bradley losing (he supported Prop 15).

[J. J.]

The only seclective memory is forgetting Wilder in 1989 (I think it was only 9 points).

[Nym90]

Nym:

1.  I know of no documented Bradley Effect in any primary, nor, except for a few posters here (and I was not of them) any claim that NH had one in 2008.  NH primaries have been notoriously hard to poll and have been since at least 1984.

2.  The question is not was there a Bradley Effect in 1982 and 1989, but in 2006 statewide races.  The answer appears to be yes, in 3 out 4 races (I can't find data for the fifth race, PA).

As a corollary to the second point, how much?  In those three races it appears to be between 2-4 points (though arguably, one poll was seven points off in MA).  Several of those, IIRC were outside of the MOE.

There is no question that black candidates overpolled in 1982 and 1989 statewide by fairly large numbers 9-10 points.  That does not exist today in any polling, but it does look like it does in some states, to a lesser extent.

3 out of the 4 black candidates were also Republicans, and Republicans overpolled in 2006. It was a Democratic wave year, and in wave elections the party that has the wind at its back usually overpolls.

Not to mention 3 out of 4 races overpolling within the MOE proves nothing anyway, due to having such a small sample size. If I could show that in 3 out of 4 races right handed candidates overpolled, would you assert that proves a polling bias against right handers?

And yes there is a question as to whether Tom Bradley overpolled in 1982. Deukmeijan's own internal poll had him down by 1. Other polls that were taken further away from the election (there weren't as many polls back then, nor were they taken as frequently, so they were more likely to miss late shifts in the race), including Deukmeijan's own internal, did show Bradley further ahead, prior to the Deukmeijan surge.

The bottom line is that if the effect is as small as you say (2-4 points) it will be impossible to prove since there are lots of effects that can cause polls to be 2-4 points off, including plain old MOE. You'd have to have a sample size of many hundreds of races with an average 2-4 point error before you could prove anything statistically as to a Bradley effect.

[J. J.]

Nym:

1.  I know of no documented Bradley Effect in any primary, nor, except for a few posters here (and I was not of them) any claim that NH had one in 2008.  NH primaries have been notoriously hard to poll and have been since at least 1984.

2.  The question is not was there a Bradley Effect in 1982 and 1989, but in 2006 statewide races.  The answer appears to be yes, in 3 out 4 races (I can't find data for the fifth race, PA).

As a corollary to the second point, how much?  In those three races it appears to be between 2-4 points (though arguably, one poll was seven points off in MA).  Several of those, IIRC were outside of the MOE.

There is no question that black candidates overpolled in 1982 and 1989 statewide by fairly large numbers 9-10 points.  That does not exist today in any polling, but it does look like it does in some states, to a lesser extent.

3 out of the 4 black candidates were also Republicans, and Republicans overpolled in 2006. It was a Democratic wave year, and in wave elections the party that has the wind at its back usually overpolls.

Not to mention 3 out of 4 races overpolling within the MOE proves nothing anyway, due to having such a small sample size. If I could show that in 3 out of 4 races right handed candidates overpolled, would you assert that proves a polling bias against right handers?

Only two of the were Republicans and I think that all three were out of the MOE.  It was one of things that I looked at.

You must be logged in to read this quote.

That still doesn't take into account Wilder.

You must be logged in to read this quote.

We'll probably have a number of polls to look at, in this race.  That's one reason I'm bringing it up.

I've actually figured in 1-2 points, but in OH, MA, and MD, it lkooks like it was between 2-4 points.

[Lunar]

Why isn't the range 0-4% instead of 2-4% for your estimate of the B.E.'s range in 2006?

[J. J.]

Why isn't the range 0-4% instead of 2-4% for your estimate of the B.E.'s range in 2006?

In the three, no.  Putting in Ford, yes.  Swann, I can't find late polling.  There was for Patrick that was 7 off.

[Lunar]

Why isn't the range 0-4% instead of 2-4% for your estimate of the B.E.'s range in 2006?

In the three, no.  Putting in Ford, yes.  Swann, I can't find late polling.  There was for Patrick that was 7 off.

But you've stated before that 2006 overall indicated a Bradley Effect of 2-4%.  I just don't get why the minimum wouldn't be -0.5%.

Also, SUSA was uniformly +5% pro-Dem in every race I saw them in, black guy or not, that year.  They are responsible for a huge chunk of that late-minute voting....

But I shouldn't argue with you about this, I know how passionately you believe in this effect materializing.   I'm just quibbling over your 2-4% representation of that year.

[J. J.]

Why isn't the range 0-4% instead of 2-4% for your estimate of the B.E.'s range in 2006?

In the three, no.  Putting in Ford, yes.  Swann, I can't find late polling.  There was for Patrick that was 7 off.

But you've stated before that 2006 overall indicated a Bradley Effect of 2-4%.  I just don't get why the minimum wouldn't be -0.5%.

Also, SUSA was uniformly +5% pro-Dem in every race I saw them in, black guy or not, that year.  They are responsible for a huge chunk of that late-minute voting....

But I shouldn't argue with you about this, I know how passionately you believe in this effect materializing.   I'm just quibbling over your 2-4% representation of that year.

I said 2-4 point measured in those three races (and discounted the 7 point one).  I've been clear over time that it didn't occur in Ford's race.  I could say 0-7 points, but I prefer the middle.

It also wasn't the black candidate over polling as much as the Republican underpolling.

I've also used a 1-2 point model; that might be an underestimate. 

I wish I could get a measure of PA.