The various regression analyses now on the Web all seem to try to explain the number of Buchanan voters by other variables. It seems more natural to me to try to explain the percentage of Buchanan voters, since it is obvious that he should do better in the more populous counties. We do not puzzle over why Bush got so many more votes in Texas than in Wyoming when both are populous states.
So I translated the data into percentages first. (You can see a map at < A HREF= "http://www.research.att.com/~volinsky/florida.html"> Chris Volinsky's site.) Here are some notable results:
PALM BEACH. Gore 62, Bush 35, Buchanan-2000 0.79 ,Buchanan--96Primary 15
DADE. Gore 53, Bush 46 , Buchanan-2000 0.09 ,Buchanan--96Primary 8
CALHOUN. Gore 42, Bush 56, Buchanan-2000 1.74 ,Buchanan--96Primary 38
LIBERTY. Gore 42, Bush 55 , Buchanan-2000 1.62 ,Buchanan--96Primary 45
BROWARD. Gore 67, Bush 31, Buchanan-2000 0.14 , Buchanan--96Primary 19.
Or, take a look at a scatterplot graph of how Buchanan did in 2000 compared with in 1996. You'll see that Buchanan did somewhat better in Palm Beach County in 2000 than in 1996, but it is is not a dramatic outlier and various other counties seem equally unusual.
I tried a regression analysis too. I used linear OLS, tobit (with bounds of 0 and 100 for percentages) and log-log OLS, and all had similar results, so I'll just report the linear OLS-- the line through the middle of the points in the scatterplot, roughly speaking. This regression excludes Palm Beach County.
Variable to explain: Buchanan's 2000 percentage
Explanatory variables (with t-statistic):
Constant: -.52 (2.95)
Buchanan's 1996 primary percentage: .026 (7.61)
Bush's 2000 percentage: .003 (1.85)
The R-Squared was .53. Thus, we see that Buchanan generally did better in counties where he did well in 1996 and where Bush did well in 2000, but there are a lot of other things going on also.
To be sure, this yields a predicted vote percentage for Palm Beach County for Buchanan of -.52 + .026 (15) + .003(35) = -.52+.39+ .11 = -.02, which means that Buchanan did much better in Palm Beach than this regression would predict. That is what we would expect from eyeballing the data. Conservatism of a county is important to how well Buchanan did, explaining much of the variance, but by no means all of it, either in Palm Beach County or the rest of Florida. Regression analysis has the limitation that it cannot take into account factors special to individual counties, factors such as especially good campaigners or appeal to small interest groups.
Econometric caveats: (1) I did not correct for heteroskedasticity, which probably exists since the counties are such different sizes. This, by the way, is probably a big problem in the studies now on the web-- when Glades County has less than 5,000 votes and Palm Beach County has more than 400, 000, the variances in the two counties are nowhere near identically distributed. It is still a problem in my regressions that use percentages, but less of a problem, I hope. (2) Buchanan's percentage is bounded below by 0 and above by 100, so tobit is more proper to use. I tried tobit, though, and no censoring occurred-- that is, no predicted values would have been beyond 0 or 100. (3) t- statistics and other statistical tests people use assume normal distributions of disturbance terms. Those are unlikely here,it seems to me, since I don't think county-specific causes of Buchanan votes would necessarily be distributed according to a bell-shaped curve.
Data note: You can get the Florida state data by county from the Florida Division of Elections.
The data I used was actually from Adams and Fastnow, " A note on the voting irregularities in Palm Beach, Florida." They have a very good website, with links to lots of other regression studies.
(2) Buchanan and the other Reform Party Candidates
In Palm Beach County the Reform County candidate for representative, John McGuire, got 2,651 votes. Those votes can't be explained by the butterfly ballot, as far as I know.
We cannot explain the Buchanan vote in that county just as Reform party voting, however, because McGuire did well in different parts of the county than Buchanan did. Buchanan received only 883 of his votes in the part of Palm Beach County in McGuire's 16th Congressional District, so McGuire ran 1,768 votes ahead of him there. In other counties in the 16th Congressional District, Buchanan also ran behind McGuire
This was the *only* U.S. House district in Florida contested by a Reform candidate as well as a Democrat and a Republican. The only other Reform House candidate was in the 9th District; he received 18 percent of the vote because the Democrats did not even bother to contest the race against the incumbent Republican.
The Reform Party candidate for Senator, Joel Deckard, received 1,282 votes in Palm Beach County. Also, some counties were like Palm Beach in that Buchanan ran ahead of Deckard (e.g., Marion (563 to 289), Polk (532 to 400)), in others Buchanan ran behind Deckard (e.g., Miami-Dade (560 to 1,395), Pinellas (1,013 to 1,895)). Thus, we see that individual differences in counties and candidates make a lot of difference.
I advise being slow to jump to the conclusion that the butterfly ballot is responsible for the Buchanan vote in Palm Beach County. The regression analysis tells us that Buchanan got unusually many votes there, and that this is not explained by its votes for Buchanan in past elections. It does not tell us *why* Buchanan did well there-- or in any other county where he did better than average.
The butterfly ballot explanation might be correct, but in general, minor third party candidates get so few votes that any small thing can drastically change their vote totals. Starting from zero, it's hard to go anywhere but up. If something added 1,000 votes to what would otherwise be Buchanan's total in Palm Beach County, that was a 42 percent increase. To increase Gore's vote by 42 percent, on the other hand, would require 113, 000 extra votes-- an impossibility even if Gore spent his entire budget on that one county. Thus, if the one competent campaign organizer Buchanan had in Florida was in Palm Beach County, that could explain the result.
(3) Like all these web studies, this one was done in haste. I don't even have a co-author or research assistant to check my work. So don't base any federal cases on it unless you verify that I got it right. I'll be happy to help out anybody who has questions about the analysis.
See Robert Shimer's "Election 2000", November 16, 2000, at www.princeton.edu/~shimer/election.html for another explanation of some of these points, and other points too.
