Well It's My Birthday Too

If you weren't reading carefully, then you might have missed an important nugget in a voter fraud story buried in Tuesday's presidential primary coverage....

April 24, 2008

If you weren't reading carefully, then you might have missed an
important nugget in a voter fraud story buried in Tuesday's
presidential primary coverage. It was in a piece Connecticut's
Hartford Courant ran entitled
"Dead
Voters
?" (Note the punctuation, more on that all-important question mark in a
moment.)

The report concerned a research project to track down dead
voters on the state voting rolls. The researchers first tried to match up
computer lists of dead people to computerized voting records, comparing names
and birthdates from one list to the other.

Most times, this would have been the end of the research—and you'd have seen screaming headlines about rampant voter fraud. So many
matches, so many illegal votes. We saw stories like this, trolling for dead
people or double voters by trying to match records from place to place, in 2000, in 2004, in 2005,
in 2006. And we'll probably see them
again this year.

These stories all share a common problem: trying to identify
individuals by computerized matching of names and birthdates isn't all that
reliable. Even if the underlying information is accurate—and the lists of who voted, or who
is dead
, often have mistakes—the matching exercise itself may get the
answer wrong.

For example, people are often surprised to find out how many
different people have the same name and birthdate. With just 23 people in a
room, it's more likely than not that two will share the same birthday (month
and day). Throw in the year, and—as proven in a new article on the
statistics of double voting
(disclosure:
I'm the co-author
)—the number is about 180.

Which means that if you've got 180 "John Smith"s or "Manuel
Rodriguez"s, at least two of them will probably be namesakes with the same
birthdate.  When you start comparing
millions of voters to millions of other people, you start picking up
doppelgängers everywhere. Finding a few name-and-birthdate matches in lists of
millions of people shows statistical probability at work, not fraud.

Which is why the researchers behind yesterday's Courant article should be commended for
their unusual follow-up. The extent to which their overall match numbers are
inflated by the birthdate problem, or other match errors, is not clear—as
they noted.
But they also recognized that the overall numbers represented the start of an
investigation, not the conclusion. With a list of 100 suspects, the researchers
apparently applied a substantial amount of shoe leather, tracking down the
actual facts behind the voters in question and beyond the match. And lo and
behold: "Although the investigation found no evidence of deliberate fraud, it
uncovered numerous errors in voting and registration records kept by local
registrars."

So, thanks to a little more effort than usual, we know the
answer to the headline's question. Dead voters? No. Responsible research? So it
would seem.