Monday, July 03, 2006

Lies, damn lies, and statistics

With the July Fourth holiday rolling around, this old chestnut has been brought up again:
In a recent study these dates, ranked in order, were identified as the 10 deadliest days of the year to drive. Mark these on your calendar and stay put on these dates.
The list that follows is notably mostly holidays, particularly summer holidays. The list is usually provided without any explanation of methodology.

Some people conclude you should stay home on those dates. Far too many, in fact. But of course even a moment's thought will reveal the problem: there are more fatalities on those dates for no more reason than that there are more cars on the road. I don't mean "heavier traffic causes collisions" (though that may well be true); I mean simply, if there's more cars out there, there's more collisions.

At a bare minimum, the study should be reporting number of fatalities per thousand cars on the road, or per thousand people in cars on the road, or something. If there are four times as many cars on the road tomorrow as today, and twice as many collisions, is it safer to drive today or tomorrow? Obviously tomorrow.

Fine, easy enough. I've written before about this. But let's push up another level. What really gets my goat is that, if you point this out, the knee-jerk, thoughtless chorus of "oh my, we should stay home that day" idiots are replaced by another knee-jerk, thoughtless chorus of "statistics are all lies" idiots.

Yes, statistics can be deliberately deceptive. And they are often deceptive unintentionally, particularly to those who (a) don't understand them, and (b) aren't in the habit of thinking critically. However, that's not a trait of statistics. Charts can be deceptive to those same people. Essays can. Photographs can. Sentences can. Words can. Inarticulate grunts can.

The problem isn't in statistics. It's in the people who mislead with them, and even more, in the people who do not understand them and then pass blame for that on everyone but themselves. Statistical analysis can be a fine and effective tool for uncovering, describing, and understanding truth, particularly patterns in the world around us. But they depend on the listener to invest a little thought, understanding, and wisdom, as well as the speaker, if they're going to work. Don't blame statistics themselves.

1 comment:

litlfrog said...

There was an argument on RPGnet before the last presidential election over polling validity. One guy flat-out refused to believe in statistical sampling, saying there's no way that 1,200 people can give an approximation of national opinion. He held onto this opinion despite patient explanations from mathematicians.