Articles and Analysis


South Carolina: More Theories and Exit Polls

Topics: 2008 , ARG , Barack Obama , Bradley/Wilder , Chuck Todd , CNN , Exit Polls , Hillary Clinton , Mason-Dixon , MSNBC , NBC News , Pollsters , PPP , Rasmussen , SurveyUSA , The 2008 Race , Zogby

Some thoughts about tonight's South Carolina's primary, including some follow-up on yesterday's post on the variation in pre-election poll results:

1) Reverse Bradley/Wilder? Noam Scheiber sees evidence of a potential "reverse Bradley/Wilder effect" in South Carolina. His theory is that live-interviewer surveys may be understating Barack Obama's support among African-Americans. Scheiber's post is worth reading in full, but here is the gist:

If Obama consistently did better among black voters in automated polls, which eliminate the "social discomfort" that might discourage them from telling (presumably white) interviewers they support him, we'd have evidence for this hypothesis.

So what do the polls say? They say I might be onto something:

In the three most recent automated polls in South Carolina (PPP, SurveyUSA, and Rasmussen), Obama takes 67, 73, and 68 percent of the black vote, while Hillary takes 13, 18, and 16. In the three most recent live-interviewer polls (Zogby, Mason-Dixon, and ARG), Obama's takes 55, 59, and 61 percent of black voters, while Hillary takes 18, 25, and 25.

So, among black voters, that's an average lead of 69-16 for Obama in automated polls, but only 58-23 in live-interviewer polls--a huge difference (53-point lead in the former; 35-point lead in the latter). It's not exactly definitive--I'm only using three data points in each case, and there are other methodological differences between the polls--but it does strongly suggest that some black voters are reluctant to tell human pollsters they support Obama, but feel comfortable saying it to a machine.

2) Lying to Robots? Mickey Kaus floats a whole new theory, that voters are just as likely to lie to "robots," perhaps even more so:

I used to think talking to a robotic phone answerer was pretty close to a "secret ballot"--what was the robot going to do to me, anyway? But machines do a whole lot these days--they track your musical tastes, follow your movements, raise or lower your credit ratings. Now a robot can conceivably do a lot to me, at least in the paranoid part of my imagination activated when I get an unsolicited call. At best, it's probably generating a list to sell someone! I don't want it know my real innermost thoughts, including my political thoughts, especially my un-PC political thoughts. These days, I'd be much more paranoid about pushing a button that say "I'm voting against beloved minority candidate X" than telling a live operator the same thing. Sorry, Rasmussen! The traditional truth-revealing advantage of robo-calling may be the artifact of a transitional era in info-technology

This is an interesting theory that, at least for the moment, lacks supportive evidence. Survey methodologists have been studying "interviewer effects" for decades, and have found consistent evidence that "self-administered" surveys (that use paper or a computer rather than an interviewer) produce more reports of "sensitive" behaviors (sexual activity, drinking, drug use). If the growing presence of computers in ours lives has made respondents less truthful when responding to self-administered surveys, so far at least, no one has proven it.

3) Will The SC Exit Poll Resolve These Questions? Noam Scheiber and my colleague Charles Franklin (and many others) will be looking at vote-by-race tabulations in the South Carolina exit poll. But readers of both wonder if exit polls are susceptible to the same effects as telephone polls. After quoting Franklin, Kaus writes:

Of course, people can lie to exit pollsters too! If you're a black South Carolinian and want to help Hillary as much as you can, you'll walk into the booth, vote for her, then walk out and tell the exit poll person you voted for Obama. There may also be non-Machiavellian peer pressure in black precincts to tell the exit pollsters the same thing (which, perversely, might hurt Obama in tomorrow night's press spin by making it look as if he received an ethnic bloc vote). In white areas similar pressure might enocourage voters to falsely tell exit pollsters they voted for Edwards or Clinton.

One of Scheiber's commenters reaches a similar conclusion:

But aren't the exit polls all done by human beings, not machines? How will you know how African-Americans really voted if they tell you on the way out that they voted for Clinton?

The problem with both arguments is that voters don't "tell" exit pollsters anything. Interviewers hand respondents a paper form, which they fill out privately and drop into a "ballot box."

ballot box.jpg (JPEG Image, 750x123 pixels).png

On the other hand, the characteristics of exit poll interviewers (race, gender and age) may have some influence on whether voters agree to participate in the survey. Historically, exit pollsters depend on mostly younger interviewers and have, as a results, had the hardest time gaining cooperation from older voters. While exit pollsters attempt to correct for such "non-response bias" by weighting, some distortions may remain.

So on the question of racial polarization in today's vote, it would be helpful to attempt to verify the exit poll findings with actual results in heavily African-American precincts.

4) More on Race and the Exit Poll - Speaking of validating the exit polls with hard numbers, NBC's Chuck Todd shares a helpful email he received from Mason-Dixon's Brad Coker after Todd noted that the 2004 exit poll estimated South Carolina Democratic primary electorate four years ago as 47% black, as compared to 55% on the most recent Mason-Dixon poll. Coker is skeptical of the exit polls as an "overall demographic indicator." He writes:

To get a real handle on what the African-American vote is likely to be, one only needs to look at real numbers. The South Carolina Secretary of State's office published the following statistics on South Carolina's 2004 and 2006 state Democratic primary elections. These are based on real voters, not a survey sampling.

According to the state's statistics, the '04 Dem primary for president attracted 58% of non-white voters compared to 42% of white voters; In the '06 Dem primary for governor, the ratio was 60-40 in favor of black voters.

These hard numbers show a much higher percentage of African-American voters in South Carolina's state primary races for Governor and U.S. Senate, so I don't think it is a stretch to expect a similar turn-out in a presidential primary that features a major African-American contender. If anything, 55% black might actually end up being a bit on the low side. I will be very surprised if a clear majority of today's Democratic primary voters are not African-American."

5) "Fraudstorm Advisory" - Our friend Mark Lindeman points out in a DailyKos diary that South Carolina "votes on ES&S iVotronics, paperless Direct Recording Electronic (DRE) voting machines" that are "exquisitely, ridiculously vulnerable to attack." As such, he has some helpful information on the shortcomings of those machines and advice on how to interpret the results in light of the inevitable speculation of vote fraud. Most relevant to our topic is his discussion of exit polls:

The exit poll "results" are not what they seem. If form holds, very shortly after the polls close at 7 PM, several networks will post preliminary tabulations based on exit poll estimates. Even before that, once the quarantine is broken, rumors may fly about what the exit polls show. Please be advised: even if you unaccountably believe that exit poll interviews are practically foolproof, these tabulations (or rumors) will not tell you the interview results! The early projections will be based on a combination of interview data and prior expectations. Given the variability in the pre-election polls, who knows what "prior expectations" will be?

Also be advised that in the 2004 general election, the estimated margin of error for the South Carolina exit poll -- assuming that the poll was otherwise unbiased -- was about 8 points on the margin between Kerry and Bush. (This margin of error cannot be figured in advance, and it can't be figured based on the number of respondents alone. It depends on the variability across the precincts in the exit poll sample.)

For those interested, the exit poll tabulations should be available just after the polls close at the following links on MSNBC, CNN and CBS.


Mark Lindeman:

On race, I want to point out a typoid in your source: Coker meant not "58% of non-white voters compared to 42% of white voters" (my emphasis), but that 58% of the voters were non-white.

South Carolina 'gives away' a lot of information that can help people to interpret racial patterns in the results tonight -- especially if they are either diligent or clever with scripting. For registration statistics, try here, choose the demographic "Race," and choose "All" to get a summary by county, or a county to get statistics for each precinct in the county. With just a bit of random fiddling, the most uniformly non-white precincts I've spotted so far are Greenville 14 and several precincts in Richland County: Richland Wards 7, 8, 9 and 19, Ardincaple, Fairwold, Greenview.

Now, go here and click on "Results by County," and you can drill down to the precinct level. For instance, it looks like results for Richland will be here , although I have no idea how quickly.

Meanwhile, we will be celebrating Burns Night, so if I do happen to post, don't place too much stock in anything I say.


Mark Lindeman:

Wow. 53.7% Obama in the first tab. 26.6% Clinton, 18.8% Edwards. (Ish.)



Well, that pretty much closes the books. Looks like the PPP polls were right on the money, and Mason-Dixon and ARG were off.

Reverse Bradley?



Folk wisdom among Democrats in the South tells us that black voters are difficult to poll simply because of widespread distrust for pollsters. This isn't a new phenomenon.


Assuming Makes An A...:

Why are they assuming 58% non-white = 58% black?

They do realize that Latinos, Asians, Arab-Americans and Native Americans live in South Carolina as well, don't they?



I would also point out that the old "likely voter" calculations might be thrown off by hordes of young people turning out to vote. This is a demographic that is usually discounted as they are seen as too apathetic to get up off their butts and get to the polls. Therefore, old algorithms designed to determine a likely voter may not work very well this election cycle.



People are disgusted with polls.
I think they are not even attempting to respond honestly anymore.
And the pollsters did that to themselves.


Nick Kaufman:

I have to say that as a European who has some experience with political campaigns and polls on the other side of the atlantic, I am very irritated with a lot of the ways Americans conduct their polls.

Some observations:

1. CNN exit polls from SC seem to be accurate for no other reason than the fact that if one calculates the final election result based on their cross-tabs, it seems to be accurate enough within the margin of error.

2. Having said that, there's no excuse for not having exit polls being accurate, as they have bigger samples, a great method (secret ballot)and the most accurate gauge of the actual population being sampled. Not knowing who is going to vote is one of the biggest problems polls face, I think.

3. There are too many polls with too small samples that increase the probability of flukes. In my country, sample sizes are usually above 1000 people and that decreases the margin of error while increasing the confidence rate.

4. The tragedy of most polls I am reading is not that they are inaccurate; it's that they fail to inform voters properly about the undecided vote. I followed SC polls as closely as I could and it was obvious to me that Obama was heading towards a 15% plus victory. Why? Because I could see that there was 10-20% undecided vote which i then divided according to the percentages candidates were actually getting.
It turns out that the vast majority of the undecideds went with the frontrunner which in my mind isn't that much of a surprise.

Polls should mention the undecided vote and they should do more work on modelling how the undecided vote behaves under different race dynamics.



Obama won over 2 to 1 but with a high African-American turnout. I am trying to figure out how low % of African-American turnout could have been in SC yesterday with Obama still winning. If I am calculating this right, African-American turnout would have had to have been less than 17% of what it actually was yesterday for Clinton to win; so even with a relatively small African-American turnout (slightly more than 18% of the total electorate), Obama still would have won yesterday. Looking at it that way, potentially his vote totals yesterdays (big assumption obviously that all things still being equal elsewhere) would translate to victories in quite a few other states. Am I calculating that correctly?



First things first, in most parts of the South non-white equals black to a degree of about 98%. While we have a large immigrant population, many hispanics are not legal and many of those who are are not registered/do not vote, same with asian american populations and other minorities.

Second, if the 2004 % was 58% non-white and MSNBC reports that over 150,000 black voters that did not vote last time voted this time, black percentage of the electorate could be 61% or higher. Assuming that 79% of the black electorate chose Obama, that knocks his non-black share down to about 17%, not 25%. In other words, Clinton+Edwards would have gotten a larger share of the black vote (at about 21%) than Obama got of the white vote. That actually puts Edwards at about 43% of the white vote and Clinton at about 40% of the white vote.

Those percentages and results get you pretty close to the actual percentage (particularly with Edwards' vote). Finding some high white precincts in various parts of the state should back that up.




We no longer need to speculate. You can look at the actual exists.

In addition to listing numbers as "white/non-white" it also gives you "black/non-black" numbers.

The final numbers, which you can certainly find yourself, leave Obama with 24% of the white vote.

The whitest county of the state was the only one Clinton won, while Obama won a great deal of very white counties, albeit by a much smaller margin than his overall vote.



Shane, look at individual precinct returns for Horry county (which Clinton won). The 20 precincts with the lowest non-white registration averaged 14 non-white voters and 16 votes for Obama. In one of the precincts, Obama got 0 votes and in a handful he got fewer than 10.

Black share of the primary electorate in 2004 was 58% and 60% in 2006 according to the South Carolina Secretary of State's website. It's not a bad assumption to assume it was closer to 62% this year, which means about 49% of eligible "non-white" voters turned out for this primary and about 12% of eligible white voters did. If you apply those numbers to county and precinct results across the various parts of the state, the exit poll % for Obama doesn't hold up.

With that kind of turnout, a county's overall election could be as low as 19% non-white and still have a majority of actual primary voters be non-white. In fact, once a county is 9% "non-white" Obama would be predicted to defeat Clinton 35-34 in the county.

Horry county is 15% black. Based on the exit polls and the statewide turnout model Obama should have defeated Clinton 43-31 here but actually lost by a reverse margin.

Now, there's actually evidence that whites turned out at a higher share in this county than they did statewide (even though the '04 turnout in Horry county was majority non-white) which still makes Clinton's victory surprising if you follow the 24% white share that the exit polls are reporting. If the county's electorate was 40% black then according to exit polls Obama should have still won the county with around 45%.

What probably actually happened there is that the county's share of black voters was around the high 20's and Hillary got a larger share of the white vote there than she did statewide (at Edwards expense) but that Obama got a very low percentage of the white vote.

Of course, this will all be a moot point soon, they'll either release the South Carolina voting stats so we can see what turnout actually was (and whether the exits add up) or we'll have other states with larger populations and we'll see how all the candidates do among all demos in those places. The bottom line is that Hillary was incredibly strong among white women who only make up about 20% of the South Carolina primary but make up close to 50% in some states and at least 35 or 40% in many of the delegate rich states on Super Tuesday.



The problem with the polls is there is too much churn in the system right now. Many folks are making their minds up at the last minute. New stories - like Bill Clinton's comments - can have a big impact on each race.


Post a comment

Please be patient while your comment posts - sometimes it takes a minute or two. To check your comment, please wait 60 seconds and click your browser's refresh button. Note that comments with three or more hyperlinks will be held for approval.