By John Carr, NMA Massachusetts Activist
Often in science you want to measure something you can’t look at directly.
You might read in the paper that scientists have measured the temperature thousands or millions of years ago. They haven’t. They have measured something in the present day, say the ratio of two isotopes of oxygen or the thickness of layers of rock under an ancient lake, and used a model to draw conclusions about past climate. This does not make the conclusions inherently unreliable. The conclusions are as good as the model. How good are climate models? A newspaper can’t tell you.
In the 1960s scientists showed that the sun was going out — deep inside it was only producing a third as much energy as it should. Was earth doomed? No. In fact, scientists never announced our doom; that was speculation by some science writers. What scientists measured was the number of argon atoms in a tank of dry cleaning fluid a mile underground in South Dakota. Oddly enough, that can tell you how bright the sun is. But in the 1960s they didn’t know you have to multiply by three at one step in the calculations. It took 30 more years of experiments to find a small error in our theory of subatomic particles. (The late John Bahcall has an article from a physicist’s point of view. Wikipedia has an article from Wikipedia’s point of view.)
We can’t build a time machine (yet). We can’t stick a thermometer inside the sun (yet). We have to use indirect methods to estimate paleoclimate or solar fusion. Luckily, we can look at traffic as it happens.
I asked a few weeks ago, why is it I don’t care about red light running and red light running crashes? The answer is, they aren’t important on their own. Some people want to use them to estimate total crashes. We don’t need to. We already know the total number of crashes and injuries. Further, red light running violations and red light running crashes are not accurate predictors of that total.
The theory behind red light cameras is that enforcement reduces violations and reducing violations reduces crashes.
At a typical intersection with cameras, red light violations go down. The first half of the theory is right, but the second half is wrong. Total crashes, total injuries, and total property damage will go up. So the theory as a whole is wrong. I’ll explain another time why I think the theory is wrong. We have enough evidence to say that it is wrong even if we aren’t sure why.
Faced with facts, camera companies quickly invented a new statistic: red light running crashes. Say, right angle collisions where neither car was turning, one car entered on red, and both cars were past the crosswalk when they collided. They took complete crash records, ignored most of the crashes, and said the rest went down.
Like I said, we know the total number of crashes. You should rely on indirect measurements, or “proxy variables,” only when you don’t know the real quantity of interest. Reporting only a decrease in right angle collisions with both cars over the stop line, and pretending that represents an overall safety improvement, is fraud.