Unless you've been under a rock, you've heard about the surpise win by political nowhere Alvin Greene, victorious on Tuesday in the Democratic primary in South Carolina, winning the right to go up against incumbent Senator Jim DeMint (R-SC). Greene was on nobody's radar. The NYTimes asked in response "Who is Alvin Greene?"
Greene's opponent was not a political unknown: Vic Rawl was elected to the South Carolina State House of Representatives 4 times, serving 1977-78 and 1980-86. With the law degree he earned from University of South Carolina School of Law in 1973, he also served as a circuit court judge from 1991-2003.
It's been suggested that Greene won, because black voters recognized his name as belonging to a black person, and voted overwhelmingly for him. I show here that that's not possible.
Greene won soundly, widely and deeply: he had a majority in 42 of 47 counties of South Carolina. His average percentage of votes across all 47 counties was 61% (that is, in a typical county, he had 1.5 votes for every 1 vote his opponent had). For all Democratic votes cast in the contest, Greene won 59% of all votes between him and Rawl.
So how did this come about? Was it his top ballot position? The fact that he’s African-American, which some have suggested might have swayed the sizable numbers of black primary voters? Or was it something else?
Here’s what the Post and Courier reported from its interviews:
State Representative Bakari Sellers, D-Denmark, suggested Greene might have benefited from being listed first on Tuesday’s ballot, but Fowler said the party’s two relatively little-known Senate candidates in 2008 polled at nearly 50-50.
State Senator Robert Ford, D-Charleston, who lost his gubernatorial bid Tuesday, said race could have played a role. The Democratic primary electorate is majority black, as is Greene, but not Rawl. "Vic Rawl had money, but he didn’t have enough. He wasn’t able to identify himself with black voters," Ford said. "No white folks have an ‘e’ on the end of Green. The blacks after they left the plantation couldn’t spell, and they threw an ‘e’ on the end."
I don't know how to quantify the advantage of having your name first on the ballot after the election is already over. Clearly, the technique to use to mitigate this effect is to randomize the order across all ballots - but that wasn't done.
However, we can do something to investigate if Greene's race played a role: we can use the census figures, and make a few reasonable assumptions.
To do this, I used the 2000 census figures, grabbing the percentage of residents by county who reported their race to be black. In South Carolina, this percentage varies from 6.8% (in the county of Pickens) all the way up to 71% (Allendale).
I assumed that the racial makeup of Democratic voters reflected that of the County.
Now, it doesn't matter that the likelihood that a black resident is a Democrat (and so voted in the Democratic primary) may be greater than another resident being a Democrat, as long as this relative likelihood is unrelated to the percentage of residents who are African American in the county in which they vote -- a reasonable assumption.
Just to give you a preview of what's to come. In the five counties with the lowest percentage of African-Americans: Pickens (6.8% population is black) Greene won with 57%; Oconee (8.4% black) it was 58%; in Lexington (12.6% black) was one of the four counties Greene lost, garnering 47.7% of the vote; county Horry (15% black) Greene had 62.3% of the vote; and county Anderson (16.6% black) Greene had 64.7% of the vote. In the five least black counties in South Carolina, Greene did almost as well in the election than he did across the state, garnering 57% of the vote. That seems highly improbable if it was the black vote which was responsible for Greene's victory.
I ran a simple statistical indicator. I looked for a correlation between the percentage of votes Greene won with in each county and the percentage of the county population which is black (according to the 2000 census). The answer: there is no correlation between the percentage of the votes for Greene with the percentage of the black population reported in the county in the 2000 census. Using the Pearson correlation test, I determined that the likelihood of seeing the trend between these two values is such that it would be observed about 16% of the time if one just randomly associated the two numbers. This is what statisticians call "no significant correlation" -- about 1 in 7 elections could have a result like this even if black voters were no more likely to vote for Greene than for Rawl. It's about as likely as like calling "six" at the roll of a dice and getting it right. This number is about the same (16%) if I remove Charleston from consideration, which is Vic Rawl's home county (and a populus one, and one which Rawl won).
Thus, there is no evidence here that counties with larger black populations turned in greater victories for Alvin Greene. But, if there were, would we be certain to detect such a correlation? I did a simple check, making the following assumptions:
- Assume blacks exhibited enough preference for Greene in order to win him
the vote total he got (>100,000 votes).
- Assume all other voters exhibited no preference (split their votes between Greene and Rawl evenly).
- Otherwise, assume the same number of ballots cast in each county; and the percentage of black voters was equal to the percentage of blacks in each county according to the 2000 census.
- Exclude Charleston County from consideration (Vic Rawl's home city) and use only the results of the other 45 counties.
I simulated this, assuming simple poisson statistical distributions of votes, with a Monte Carlo simulation. First, Greene breaks about 100,000 votes when blacks prefer him about 85% vs. 15% against Rawl statewide. At that point, he's winning about 44 of the 45 (remaining) counties. However, if that's the case, blacks are voting for Greene in such high numbers that suddenly a strong correlation appears between the percentage of votes Greene receives in a county, and the percentage of population who are black in that county.
How strong is the correlation? Again, the way to express the strength of the correlation is to give the probability of the observed correlation occuring (via the Pearson test) if there were no actual correlation between the two values. And, I find the correlation is improbable the level of 1e-22. That's about one billionth of the probability of South Carolina winning Lotto 6/49 if they bought a single ticket.
In other words, if the reason that Alvin Greene had won the nomination was that blacks voted for him in disproportionate numbers, then the correlation between the percentage of the population of each county who are black, and the percentage of the vote Greene received in each county would be so strong, we couldn't possibly miss it. But we don't see any such correlation in the actual voting and census records. It's just not there.
Is there some way to mask such a correlation? The only possible scenario is that, in counties with small black populations, the other voters were disproportionately and overwhelmingly more likely to vote for Greene. To repeat: voters in racially segregated counties with very few blacks would have to have overwhelmingly preferred Greene to Rawl. There is no reason to believe that's the case.
I conclude from this: it's not possible that Alvin Greene won the primary because of overwhelming support from black voters. We'd simply see the booming correlation with the percentage population of each county who are black, and we don't.
That leaves only one non-illegal possibility: that voters picked Greene preferentially, at random -- regardless of the voter's race. It's been put forward that perhaps it's because he was at the top of the ballot, and people didn't recognize either candidate, so they randomly chose the top name. I have no evidence against this interpretation at this time.
There is, of course, the illegal possibility: that the election was fixed. For that to have occurred, it would have to have been fixed by someone who had at least the ability to change voting outcomes in every single county, but also the ability not to change the outcome in Rawl's home county of Charleston, where Rawl won.
After I completed this analysis, I saw the similar analysis by jeffmd at the swingstateproject. I'll digest it and write more later.