10 December 2012

Call It

This is another post left over from the days immediately following the election.

A lot of people have spilled a lot of ink in the last couple of weeks praising Nate Silver's numerical thoroughness. (Myself included.) This makes is all the more annoying to see things like this...
twitter | @asymmetricinfo

South Carolina projected for Romney with 1% of precincts in and Obama at 77% of counted vote. #statistics
and this...

I'm almost positive these kind of anomalies are the results of something like poorly synchronized database queries and not nefarious media plots to throw the election. But they're still really annoying.

More generally, it's frustrating to sit through an evening of TV coverage with anchors hemming and hawing about "ooooooh, too close to call right now, razor's edge, but it's looking good for Smith, but the margin is X, with Y percent of precincts counted, but it's still too early to know, but we're pretty confident, but stay tuned, but we think it's going this way, but..." No one ever bothers to explain what criteria are being used to "call" a state for one candidate or another.

What lead are they looking for given a certain percent of votes cast? What confidence interval are they waiting for? What statistical tests are they doing? Surely after all the Silvergate brouhaha they aren't "calling" states when some editor or producers just scratched his chain and says "yup, that's it! I feel good about 'calling' this state now."

Media: please give me something — anything — about the quantitative method you're using to make these decisions.

If not for me, then for the millions of people out there who are convinced you're doing it completely subjectively in order to discourage Red Team voters from staying in line. Because I don't think it's a conspiracy, but it sure would be nice to have you try to demonstrate that.

Now maybe that's hard to do on live TV. (Although with all the blathering they were doing, it's not like they couldn't squeeze some actually content-filled explanation in there. So what if it's hard? If we believed 1% of the mythology journalists tell about themselves they ought to be willing to try something hard like explaining some statistics to their audience.) It wouldn't be that confusing or difficult to list the number of votes a candidate is in the lead together with a confidence interval for what the "true" lead will end up being. At the very least give me a sidebar somewhere on all the election results websites explaining the method they use to "call" each race.

No comments:

Post a Comment