Robert Kennedy, the election, and statistics
Robert F. Kennedy Jr. gives a nice presentation of all the unusual events that happened on Election Day 2004. His discussion of all the efforts undertaken to suppress the Democratic turnout in Ohio (misallocation of voting machines, making voter registration difficult, disqualification of provisional ballots, etc.) is well-written, and certainly makes the case for unethical (and possibly illegal) behavior on the part of Kenneth Blackwell. However, I wanted to clarify one issue for people. Kennedy writes at length about the difference between the exit polls (which showed Kerry winning) and the vote tally (which Bush won), and he repeatedly says things like “The statistical odds against such a variance are just shy of one in 3 billion”.
This is the probability of what statisticians call the null hypothesis. In this case, it is that “any difference in the proportion of votes for a candidate in the exit poll and the proportion of votes for a candidate in the precinct tally is due solely to the random nature of sampling voters as they exited the polls.” This hypothesis is vanishingly small, as has been known for more than a year, so it can be safely rejected. But this probability says NOTHING about the truth/validity of either of the following two alternative hypotheses being true:
1. The exit polls are biased, and did not accurately measure what voters did in the ballot box.
AND/OR
2. The votes cast in the ballot box were manipulated to give Bush a victory.
These hypotheses are not mutually exclusive, and it’s possible that in a few precincts there was vote fraud, while the exit polls were systematically wrong in many states due to bias in the methodology. What does not seem plausible to me, given the decentralized nature of the American voting system, is the implication that only hypothesis 2 is correct- there’s just no way that vote fraud could be conducted in dozens of states, involving hundreds of people, and have it remain quiet. Pollsters may not have figured out the source of the bias, but that doesn’t mean it doesn’t exist. In any event, showing people huge numbers like “one in 3 billion” only serves to rhetorically confuse people into thinking that’s the odds of vote fraud having occurred. If I’d reviewed this paper for a scientific journal I would’ve edited all such sentences out, as they aren’t germane to the main rhetorical thrust of the piece.
