Back in February, on the occasion of Ralph Nader again declaring his intention to run
headfirst into a brick wall for President, Sifu Tweety Fish of The Poor Man Institute decided that, a sufficient amount of something having somethinged, that the time had come around to annoy Nader supporters.
The current post, however, is not intended to annoy anyone and is not really about Nader, but rather about inconsistencies of American politics that make analyzing voting patterns very difficult.
In the comments to the aforementioned thread, fish posted the following:
Gore lost by around 750 votes. Nader got 97K votes in Florida. Bush got 200,000 votes from registered Democrats. 50% of all registered Democrats didn’t even bother to fucking vote in Florida.
Yeah, your right, it was Nader’s fault.
The part of this I want to focus on is “Bush got 200,000 votes from registered Democrats”.
While I have never looked at detailed data from the 2000 election, I did end up toying with data from Florida in 2004.
An early statistical analysis of the data from the 2004 election indicated that there might have been problems with the voting machines in Florida. My opinion at the time appears to have been that various articles had “quot[ed] these statistics indiscriminately” and that people on all sides had failed to recognize that “it was an incomplete analysis and [thus one should] not jump to conclusions”.
In the course of the discussion of that analysis, I took my own look at some of the data to try and figure out what was going on in some of the counties.
For example, the numbers for Taylor County looked like this:
|Voted Republican (Bush)
|Voted Democrat (Kerry)
|Expected Republican Votes
|Expected Democratic Votes
The “expected” numbers just being the registration percentage applied to the actual turnout. Ignoring for the moment anyone not registered D or R, Democrats voted for their own candidate at less than half the rate expected and Bush received votes at over three times the rate expected and amounting to two-and-a-half times the number of registered Republicans.
When one considers that there were almost certainly Kerry voters who were not registered Democrats, the Democratic turnout looks even worse. And even if one assumes that:
- every registered Republican turned out to vote and voted for Bush (2170), and
- every registered voter who was registered neither Republican nor Democrat turned out to vote and voted for Bush (631)
that total (2801) would still require that 2665 registered Democrats voted for Bush. (Call this measure of required Democratic votes for Bush R.)
There are twenty-eight counties in Florida which voted for Bush in 2004 at nearly or over twice the “expected” (that is, percent registered R times total actual votes) rate (for eleven counties it was over 3x). These are all low-population counties: more than a third have fewer than 10,000 registered voters, and only two have over 30,000.
At the time, some people were pointing to results like this and screaming that it was clear evidence that the Republicans were stealing votes. If I recall correctly, I, too, may have believed this at first. In any case, I decided to check out what the voting history was in such counties in prior Presidential elections. For Taylor County, I found the following:
Democrat Republican Third Other
2004 Kerry 35.5% Bush 63.7% 0.8%
2000 Gore 38.9% Bush 59.6% Nader 0.9% 0.6%
1996 Clinton 44.8% Dole 39.9% Perot 14.3% 1.1%
1992 Clinton 35.6% Bush 37.3% Perot 26.7% 0.3%
1988 Dukakis 30.0% Bush 69.1% 0.9%
1984 Mondale 30.0% Reagan 70.0% 0.0%
1980 Carter 50.5% Reagan 47.3% Anderson 1.3% 0.9%
1976 Carter 62.3% Ford 36.7% 1.0%
1972 McGovern 15.5% Nixon 84.5% 0.0%
1968 Humphrey 18.6% Nixon 15.7% Wallace 65.7% 0.0%
1964 Johnson 39.1% Goldwater 60.9% Unpledged 0.0% 0.0%
Note especially the 1968 results.
The conclusion I came to at the time (based, to my recollection, both on this data and some independent confirmatory information) is that the counties in Florida which follow this pattern are rural counties with very conservative populations which are governed on a local level by Democratic establishments of a form little changed since Reconstruction. Thus, many people are registered Democrats because of how local politics works, but this does not correlate meaningfully with how these voters make their choice in a Presidential election.
Applying measure R — which is obviously an underestimate — to the twenty-eight counties in the group I’ve described one finds that there is an absolute floor of 47,314 Democrats who voted for Bush (17% of registered Democrats).
If one uses the number of votes by which Kerry underperformed estimated expectations, something like 85,000 votes for Bush probably came from registered Democrats in these twenty-eight counties. That too may very well be an underestimate as it assumes no one voted for Kerry who is not registered Democrat. These counties are home to around 4% of Florida registered voters. There are, it stands to reason, rural areas in counties which include urban centers where this pattern also holds but where it is not visible in a coarse-grained dataset.
What I think this shows is that an unqualified statement like “Bush got 200,000 votes from registered Democrats” can not be effectively employed in an argument applied to a state where there exists a significant population for whom party affiliation with the Democrats is not predictive of Presidential preference. I have no particular point here to make about Nader or the 2000 election; I just consider this a useful example of the slipperiness of political statistics.