Berkeley bunkum

For the past few days, the die—hard internet conspiracy theorists, desperate to cling to some thread of reasoning that will reassure them that Kerry actually won — that Bush stole the 2004 election —  have been touting a 'study' released by some sociology students at the University of California at Berkeley.  The paper, titled The Effect of Electronic Voting Machines on Change in Support for Bush in the 2004 Florida Election, suggests that Bush received a surprisingly large number of votes in three heavily—Democratic counties in Southeast Florida that used electronic voting machines in this election.   The implication is that something nefarious took place there, and that Bush's vote totals in these three counties were not legitimate. 

The authors of this study should be embarrassed with what they have released.  To put it starkly, their 'study' is garbage, and the methodology and conclusions are ridiculous. 

The study authors wanted to test whether electronic voting in Florida and Ohio was unbiased. In essence, they wanted to determine if the vote count would have been different with a hard copy back—up or other verifiable paper trail for the electronic voting system. To truly test such a hypothesis, one would want to compare the results of the same voters using electronic voting machines, with and without a paper trail, or by each voter once voting with a punch card ballot, and once with an electronic voting machine.  But since that was impossible, instead the study looks at the voting results in prior elections and what this can predict about how today's voters 'should have' voted,  compared with the results as recorded on the electronic voting machines in this election.

The basic problem with this approach is that the pool of voters in 2004 is not the same as in 2000. For one thing, there was a much larger voting pool in 2004. Some voters from 2000 have died and others may have chosen not to vote this time around.  On the other hand, many new voters have moved into the three counties, or came of voting age, and/or were registered to vote for the first time, or decided to vote in this election, but not the prior one. 

Florida's vote total grew by 27% from 2000 to 2004, by 31% in the 64 counties other than the three in Southeast Florida, and by 20% in the three Southeast Florida counties. One can simply not assume, as the study authors do, that all things being equal, the Democratic and Republican share of the vote in 2004 in these three counties should mirror the percentage distribution in 2000. For one thing, the state of Florida voted much more Republican in 2004 than in 2000, and in 2000, it voted more Republican than it did in 1996. Why shouldn't this pattern have also been reflected in these three counties? The study authors do not identify any problem with the results in the other 64 counties in Florida, whether in the other 12 electronic voting counties, (some of which, such as Hillsborough [Tampa] are quite large), or the 52 counties which voted  with an optical scanning system.  Excluding the three counties at issue, (Miami—Dade, Broward, and Palm Beach), the remainder of Florida (the other 64 counties) voted as follows in the last three Presidential elections (Republican candidate first in each case, with the two party share after each candidate)

1996: Dole : 1,761,000 (51.2%), Clinton: 1,681,000 (48.8%)
2000: Bush: 2,290,000 (54.3%), Gore : 1,925,000 (45.7%)
2004: Bush: 3,135,000 (56.8%), Kerry: 2,382,000 (43.2%)

The 2004 results posted above and below reflect the same incomplete statewide numbers from CNN that were used in the Berkeley study, and are about 1% short of the final numbers statewide. I do not know why the authors chose to use an incomplete sample, but sometimes, when you rush to judgment thinking you have a 'gotcha' on your hands, carelessness slips in, compounding the serious problems in this study with its methodology and conclusions.

Here is the voting pattern in the three supposed 'problem' counties during these same three elections:

1996: Dole 482,000 (35.8%), Clinton 865,000 (64.2%)
2000: Bush: 620,000 (38.6%), Gore 985,000 (61.4%)
2004: Bush 777,000 (40.3%), Kerry: 1,153,000 (59.7%).

The results over the three elections in the three counties mirror precisely the same trend that occurred in the other 64 counties in the state. The Republican share grew in each election. Over the 8 year period, the Republican share grew by 5.6% in the other 64 counties, and by 4.5% in the three Southeast Florida counties. Arguably, the Republicans underperformed in these three counties in 2004. One might have expected their share to grow MORE from 2000 to 2004, than it did in Miami —Dade, Broward, and Palm Beach Counties.  One might argue, if one were a Republican conspiracy theorist, that Bush votes disappeared in these counties, and became Kerry votes. 

But the Berkeley authors argue that the Bush vote total in these three counties was 130,000 too large, or in other words, it exceeded their predictive model for these counties by that much. They then stir the pot a bit more, and suggest that maybe these 130,000 extra or 'ghost' Bush votes as they call them, were actually Kerry votes, which were electronically switched to Bush votes, so the real theft was 260,000 net votes, or just over  2/3 of Bush's final statewide margin of 381,000 votes (still not enough in other words, to overturn Bush's winning the state's 27 Electoral votes, even if true).

The study even tells us how such evil deeds might have been perpetrated on the electronic voting machines — first by having the machines programmed to have a maximum number of votes that could be recorded for a candidate (Kerry, of course), and once that level was reached, additional  votes for that candidate were deducted from his total, or added to Bush's total. Another option was that the vote total counter was set to some positive level for Bush before the voting began. One can just imagine the Berkeley students debating whether it was Florida Governor Jeb Bush or Karl Rove who obtained the secret passwords from the executives of the voting machine vendors ES& S and Sequoia, to enable them to carry out these treacherous deeds. It is easy to forget that this study was done by sociology students at what was once considered a prestigeous university, and not creative writing students, or patients suffering from Post Election Distress Syndrome.

Just for fun, let us assume that the Berkeley students have a clue and are onto something (they haven't and they aren't).  Their first theory is that Bush received 130,000 too many votes in these three counties.  Fine, subtract 130,000 votes from Bush, and Kerry then wins in these counties by 1,153,000 to 647,000, or by 506,000 votes. The percentage breakdown would be 64.1% to 35.9%. Does this sound right — that Kerry should have won the counties by 506,000 votes when Gore won them by 365,000? Why would Kerry's share be 2.7 per cent higher than Gore's was in these three counties, when in the rest of the state, Bush did 2.5% better in 2004 than he did in 2000, and increased his margin of victory by 378,000 votes ?  Also, does it make sense that the total vote in these three counties would only rise by 12% in 4 years, when it rose by 31% in the rest of the state? Does it make sense that Kerry would have won 86.2% of the incremental 195,000 votes cast in 2004 in these three counties as compared to 2000?

Look at this in another way. In 2000, the three Southeast Florida counties supplied 21.3% of Bush's statewide vote total. In 2004, they supplied only 20% of his statewide vote total.  In the other 64 counties in Florida, Bush's vote total grew by 36.9% in four years. In the three 'suspicious' counties, his vote total grew by 25.3% from 2000 to 2004.

Now let us double the nonsense. Take 130,000 votes from Bush and give them to Kerry. Now Kerry wins by 1,283,000 to 647,000 or by 636,000 votes, almost 2 to 1, with 66.5% to Bush's 33.5%. Does that sound right — that the Democrats' share would grow by 5% from 2000 to 2004, and their victory margin would grow by  271,000 votes, when the total vote in the three counties only grew by 325,000 votes?  In essence, of the 325,000 additional votes recorded in the three counties in 2004 as compared to 2000, the Berkeley students are suggesting that the more believable result, the one that would have been predicted by their model, would have been for Kerry to gain 298,000 of them and for Bush to pick up only 27,000. 

For the record, that is a 91.7% to 8.3% victory for Kerry in the incremental vote pool. If this is what the model predicts should have occurred, the Berkeley students have a serious credibility problem.  They also have a real world problem— they obviously never tested their ghost or double ghost vote switch totals for reasonableness in terms of the voter share between the candidates that would result. 

There are a lot of reasons why voting patterns change in states or in individual counties between elections.  As mentioned earlier, the pool of voters and their relative preferences between the parties can change quite dramatically, particularly in a rapidly growing state such as Florida.  You also need to account for the quality of the campaigns run by the two candidates, the advertising spending or its effectiveness by either candidate (advertising is bought in local markets, not statewide), and the success of the get out the vote drive by each candidate.

It might also help to consider some of the things that have happened in the world, the nation, Florida, or in these three counties the last few years. Such events might have contributed to the results in these counties, by making one candidate, or one party, more attractive to the different group of voters who are around to vote in 2004 as opposed to those who voted in  2000.  The 9/11 attacks, for one thing, seemed to matter in this election. Was it just an accident that Bush ran much stronger in New Jersey in 2004 than he did in 2000 (losing by 6% instead of 16%), and also similarly improved his vote share in New York State, and that these were the two states which suffered the greatest casualties in those attacks? 

The exit polls, which were off a few points with their final totals, indicated that self described Republicans equaled the number of self—described Democrats nationwide in 2004, 37% each. In 2000, self described Democrats were 39%, and Republicans 35% of the exit poll sample.  Given that the exit polls understated Bush's actual national support, there were probably more Republicans who voted than Democrats this year. A shift seems to have occurred from 2000 to 2004 in the relative size of the two parties' voting constituency. But the Berkeley study seems to assume constant party share among voters in the three counties, and hence any Bush improvement is suspicious, and related to potential electronic voting machine fraud. 

There are also specific issues that might have mattered to the voters in these three counties, and contributed to a change in the party's relative vote share. For instance, consider one factor the Berkeley model ignores. The level of support by Jewish voters for President Bush was substantially higher in 2004 than in 2000 in Florida. Joe Lieberman was not on the ticket this time, and many Jewish voters were pleased with President Bush's strong support for Israel. This increase in Jewish support for Bush is evident from the exit polls that were conducted on Election Day in Florida, from a phone survey conducted by Frank Luntz in Florida, and by examining the precinct votes in heavily Jewish areas in the three counties in 2004 versus 2000. The three counties that the Berkeley students  identified as producing too many Bush votes, happen to be exactly the three counties where most Jews in Florida live (over 80% of the statewide total, with over 400,00 Jewish voters in these counties alone). 

Attending school at Berkeley, the sociology students there probably do not have much of a clue as to the unique local factors that might actually cause a voter to support one candidate versus another in Southeast Florida.  For instance, Bush did much better in Miami— Dade County in 2000 than Bob Dole had done in 1996. Probable reason: the Elian Gonzalez affair (and Cubans' anger at the Clinton administration), which generated over 80% support by Cuban Americans for Bush. This was a unique factor for one segment of the Hispanic voting population that year.

Finally, let us not forget that these three counties are in Florida, a state that consistently votes more Republican in Presidential elections than the nation as a whole. Starting with 1952, there have been 14 Presidential elections, including the 2004 race. In 13 of them, all but 1976, the Republican candidate for President has outperformed in Florida compared to his performance nationally. This has occurred now in 7 consecutive elections. This year, Bush won Florida by 5%, and won the popular vote nationally by 2.9%. In 2000, Bush won Florida by 0.01% (call it a tie if you like), and lost nationally by 0.5%. In 1996, Dole lost Florida by just under 6%, but lost nationally by over 8%. In 1992, Bush's father carried Florida by 2%, while losing nationally by 5%. So it is no surprise that Bush won Florida this year, or that he won it by a bigger percentage than his national margin. 

All in all, the Berkeley study is an embarrassment to that university, and especially to its Sociology Department. One simply cannot predict a candidate's vote total using the methodology they do.  The pollsters would all use linear regressions on past voting behavior to predict this year's results, if it were meaningful. It isn't and they don't. And their study is also not the way to test the accuracy of electronic voting. The students never tested what the shift of 130,000 or 260,000 votes would mean for the results in the three counties.  As shown above, the results would be ridiculous, as are the conclusions of this study.

But the Berkeley sociology students have contributed to the inability of millions of Kerry voters to let go, and concede their guy lost, and move on, so to speak.  And keeping alive the illegitimacy flame is what may sustain them during the bleak winter ahead.

The author would like to thank King Deets, PhD, for his help with statistical analysis for this article.