Desperate to be outraged about something, the Wall Street Journal reports on claims of vote fraud in South Dakota.
The problem the Journal faces is that the evidence of widespread fraud doesn’t exist. There is evidence of small-scale fraud with 4 native-americans being paid to vote. This should be prosecuted.
What is amusing is the use of a post-doc at the Harvard-MIT data center for authority. Most of the folks who get that job are bright, but certainly not authorities on vote fraud. One should focus on the argument made, and I do below, but if one is truly interested in a statistical analysis one should ask a scholar with a background in such issues, like, oh, I don’t know…Gary King at Harvard. This is especially true since ecological inference was developed by King and would be especially well suited to the problem.
But Michael New, a post-doctoral fellow at the Harvard-MIT Data Center, has inspected the South Dakota Secretary of State’s Web site to discover other striking facts: While Democrat Tim Johnson ran statewide about 12 percentage points behind what Mr. Daschle got in his 1998 Senate victory, in Shannon County Mr. Johnson ran about 12 percentage points ahead. He got 92% of the vote compared with Mr. Daschle’s 80%. Nowhere else in the state did Mr. Johnson improve his vote share relative to Mr. Daschle.Senate voter turnout was up 27% statewide for this year’s close contest compared with 1998, but in Shannon County turnout increased by 89%. Again, no other county in the state showed comparable turnout increases. Shannon County is largely Indian country, home to the Oglala Sioux nation, and is heavily Democratic. But Mr. Thune managed to receive only nine more votes there than did Mr. Daschle’s opponent in 1998, notwithstanding the much larger turnout.
Why didn’t New compare registration rates to the final results? Because he wouldn’t like the results. In Shannon County, South Dakota, 7.2% percent of registered voters are Republican. 7.8% of the voters in Shannon County voted for Thune. There is a higher proportion of other registered voters in Shannon which I assume are independents, but given Native American voters often vote almost 90% Democratic unless John McCain is on the ballot, the overall results aren’t that surprising. Indeed, it appears that for a county with 94.2% Native Americans, the numbers are to be expected.
mr. new points out that this is just a 4% increase in gop votes over 1998. in the other three south dakota counties where indians constitute more than two-thirds of the population, mr. thune gained between 23% and 43% more votes than the gop candidate in 1998. the oglala sioux would seem to give new meaning to the phrase “bloc voting.”
But this doesn’t give us the relevant information. Who voted in those places? Were Native Americans turning out? Or were other people in the county? Given there are 505 registered Republicans in Shannon County is it that absurd that half turned out? Not only that, but the turnout virtually mirrors the 1998 Senate race and the 2000 Presidential race. He is assuming the marginals should stay the same if turnout increases. But targeting specific types of turnout certainly alters the marginals.
In fact, the story is more obvious. Democrats put a lot into turning out Shannon County for a reason. They had a lot of potential votes there. Republicans probably didn’t bother because they have not broken 252 votes in the county for statewide offices. The Republican vote has been constant while Democratic vote has been increasing. Given the Democrats were doing voter registration and GOTV in Shannon, unless the Democrats were complete idiots and targeted Republicans, this is as expected.
As a clue to those unfamiliar with how to evaluate stats, if there is a new variable one should expect a change in the behavior. Given there was a massive voter registration drive and significant GOTV efforts by Democrats only in this county, one should expect a very different effect on the relative proportion of the vote.
As Mr. New concedes, "this could all be a coincidence." But "this trifecta of late results, high turnout and unusually strong support for the Democratic nominee should, if nothing else, arouse suspicion." >But Mr. New has a problem, no one in their right mind would call it a coincidence, they would call it a voter drive. Treating a significant effort to turnout voters sympathetic to the Democratic Party as a coincidence is malpractice for one who is trained in statistics.
New wants to look at this issue from the electorate being a constant proportion. In other words, if turnout increases there should be proportional increases in both parties votes. This is not necessarily the case, especially in a county like Shannon. In 1998, 2000, and 2002, about 250 votes were cast for Republican Senator or President in each year. This is almost exactly 50% of the registered Republicans in Shannon County, South Dakota. Given Thune would have been stupid to spend turnout dollars on so few votes, he didn’t. One might argue a voter drive amongst Native Americans should have at least stayed constant in the proportion of Republican votes amongst Native Americans. Such an assumption is unwarranted. Voter drives usually target voters less likely to vote and almost by definition, less likely to be informed on voting. They would look to peers for cues as to how to vote because of their low level of political sophistication and in a one-party reservation, that is likely to result in near unanimity amongst picked up voters. There is a way for Republicans to get around this. Work for Native American voters?like John McCain does.
The calculus is entirely different for Democrats than for Republicans. If one assumes the 917 independent registered voters are heavily Native Americans (a reasonable assumption in a county that is 94.2% Native American) that means total, likely Democratic voters are around 92.6% of registered voters in the county at around 6473 voters. In a state where an election is going to be close those votes are something to concentrate upon and the Democrats did. Shannon County did deliver the election to Tim Johnson, but that isn?t something dark and devious behind it. It was working your base and getting people to the polls. There were 2856 votes for Johnson in Shannon County this year meaning a turnout of around 44% of likely Democrats and I find that low number depressing.
Nothing in the county affected Republican turnout and so it was constant. However, a variable was introduced into the Democratic turnout and this changed the proportion of votes between the parties. This isn’t rocket science, it is obvious.
Now, if widespread fraud occurred, present legal evidence. If not, stop whining because people exercised their right and celebrate democracy in action.
And the cheap shot about the Ashcroft election at the end of the election is stupid. Ashcroft lost by 40,000 votes. Vote fraud didn?t produce that difference.
Substantially edited from the first posting