01 February 2013

Statistics do not care about propriety

The Consumerist | Mary Beth Quirk | Consumer Group: The Rich May Pay Less For Car Insurance Even If They’re Not Safe Drivers –

Driving safely and avoiding accidents isn’t just common sense — injuries hurt, car wrecks are bad — but also a way to make sure drivers keep their auto insurance premiums down. But according to figures released by a consumer group recently, insurance companies are in the habit of charging higher premium to safe, low- or moderate-income drivers than to richer people who were at fault for an accident. The review by the Consumer Federation of America (via Bloomberg News) says that out of 60 cases it studied, the good drivers were hit with higher prices two-thirds of the time, because of factors like education and occupation. [...]

Using two hypothetical characters the group compared premiums offered to two 30-year-old women. Both had driven for 10 years, lived on the same street in a middle-income Zip code and both wanted the minimum insurance required by whichever state the group was researching.

The imaginary woman who wasn’t married, rented a home, didn’t have coverage for 45 days but has never been in an accident or ticketed with a moving violation was compared to a married executive with a master’s degree who owns her home and has always had continuous insurance coverage. But she’d been in an accident (again, hypothetically) that was her fault and caused $800 in damage within the last three years.
The difference between the "safe driver" and "reckless driver" was $266.67 in damages annually. We're not talking about totalling your car by plowing into someone's living room. We're talking about the difference between a single, very minor, fender bender and nothing. It's a lot easier to see how other factors trump that difference in driving history than the way I've seen it presented in most of the press, which is a simple, categorical distinction between a "safe" and "unsafe" driver.

Many actuaries are saying that the features {income, rent-vs-buy home, marital status, education, ...} are a better predictor of outcome than the feature {history of fender-benders}. Our intuition is that fender-benders should be the best predictor. There are several possibilities:
  1. The actuaries are biased or vengeful or irrational, and are using their power to exploit people they don't like. If this is the case I would expect other insurance companies (State Farm, who regularly offered lower rates to the "safe" driver) to gobble up market share from them. As long as someone in the industry is greedy enough to prefer increased market share over punishing poor people, this problem should sort itself.
  2. The actuaries have made a mistake. This is possible, but we're seeing the same patterns across multiple companies. Perhaps they all have some theory induced blindess, but that they independently reached the same conclusions gives me confidence they aren't wrong. Insurance may also be the situation in which I have most confidence in Large Calculations: the sorts of models they're building for car insurance can make concrete predictions which can be tested rigorously and often.
  3. The actuaries are right. In which case... deal with it. If you really think your intuitions can make better predictions about a complex, high causal density system like socially-derived driving habits than the boffins with their numbers and data and computers and regressions then go for. Start an insurance company and get rich. Just do me a favor and don't call yourself a supporter of Science while you're doing it.
A couple of years ago I was at a conference and the speaker had presented a mildly controversial paper claiming that X was correlated with Y. Members of the audience kept objecting. Finally he got exasperated, threw up his hands, and declaimed "Look, you can keep giving theories about why X and Y shouldn't be correlated until you're blue in the face. But they are. Either give me an actual criticism of what I did wrong in the measurements or calculations, or accept my conclusions. Simply repeated 'but it shouldn't be that way' isn't getting us anywhere." That's what someone needs to tell the CFA.
“State insurance regulators should require auto insurers to explain why they believe factors such as education and income are better predictors of losses than are at-fault accidents,” said J. Robert Hunter, CFA’s director of insurance and former Texas insurance commissioner.
This is not how math works. It doesn't matter why. You don't have to tell ex post stories about why correlations exist; you just have to know that they do.

The fact is that income etc. are better predictors. You can attack the way the insurance companies actually calculated that, and demonstrate that they made a mistake. But you can't attack them for not having a sufficiently convincing story to tell you about why.

(Speaking of which, let's make some stories. People with low incomes are more likely to be one-marshmallow people. People with low incomes are less likely to pay their bills on time. People who didn't finish school are less responsible in their education, and thus may be less responsible on the road. People who are unmarried are more likely to drive home from a night of drinking. People with lower incomes have lower job satisfaction, causing them more stress, causing them to drive more erratically when commuting. I don't know if any of these are true. But that's my point: I don't need to know if these stories are true, because they're just the stories. What I need to know is if the numbers are accurate.)
“Policymakers should ask why auto insurers are permitted to discriminate on the basis of nondriving-related factors such as occupation or education,” he added.
The English language seriously needs two different words to distinguish "discrimination" in the sense that Hunter is attempting to use it from statistical discrimination. Because people would be straight-up horrified if insurance companies stopped practicing statistical discrimination, not to mention the shenanigans that resulted from people conflating the two forms of discrimination during the ObamaCare debates.
[Consumerist is] of the same mind. Rewarding good drivers with higher premiums seems like a backwards way to do it.
This. This right here is the crux of the issue. Prices are not rewards or punishments. Geico, Nationwide, et al. are not passing moral judgement on people by offering them higher rates. It's not about how much they like you or whether they think you're a good person or if you give them warm-and-fuzzies in the cockles of the hearts. They're attempting to predict how expensive you will be as a customer and add a few percent for profit. That's it. Stop treating prices like a popularity contest.
If they’re going to get charged more anyway, where’s the motivation to be a safe driver, beyond insuring your own personal safety?
Wait, what? Does Quirk understand what happened at all? It's not like if you get in a few accidents but leave everything else the same you'll get a discount. The options studied were not "Person A, with no accidents" and "Identical Person A, with one accident." This is just... I'm not even going to bother with this not-even-wrong foolishness.

No comments:

Post a Comment