(This one’s been sitting around for a while. It’s in response to a November 6, 2006 editorial, so it’s not very timely at all. But what the heck.)
This op-ed demonstrates why sociology is not considered a “hard” science. The errors abound.
1) 99% is not the standard significance threshold in science, 95% is.
2) It’s easy to count penny jars accurately. It’s called being careful. It’s especially not hard to count penny jars when you have million dollar budgets.
3) He spends some time stating how four different counts produce four different numbers, and suggests averaging them. Three paragraphs later, he ignores all this to say that in Washington’s 2004 race, 1,373,361 votes didn’t beat 1,373,232 votes by enough, so it shouldn’t count.
4) While it’s vaguely true that recounting may get you slightly different numbers, that’s not where the issues lie. The issues are not implementing a counting methodology, but having an agreed upon counting methodology. Do hanging chads count or not? Are absentee ballots with insufficient postage valid or not? If someone waited 12 hours to vote and couldn’t because of broken machines, should they get to vote? If someone was incorrectly purged from the voter polls, does their vote count? If they voted in the wrong precint, does it count? If they were confused by a butterfly ballot, can their vote be re-assigned to who they obviously meant to vote for? If a voter doesn’t have the correct ID, even though they are a legal voter, does it count? etc. Those are the questions that lead to recount after recount.
5) The proposed remedy of do-it-again is just plain insane.
6) I’m a statistics person, but statistics are not the way to hold elections. A 99% certainty level (or a p=0.01) means that about 1 in 100 times, you get the wrong answer and are OK with it. 99% is not enough here.
The answer is agreed upon counting methodologies. Execution may not be easy, but it is staightforward and auditable. Given reasonable time and money, the counting will always be accurate.