There’s a kind of thought process which I enjoy, which I think is pretty core to EA. It’s a particular approach to answering questions of the form “should I do X”. It involves counterfactual reasoning, a bunch of basic microeconomics, and careful attention to your values.

For example, I recently saw someone stealing a bike outside a BART in Oakland. Should I have called the police on the guy? Here are some of the considerations I can think of. I kind of got carried away thinking about this, feel free to skip to the end if you get bored.

What will happen differently if I call the police? Well, the police might come try to stop the guy. They may or may not get here in time. If they get here in time, the guy will probably try to run away. The police might successfully arrest him. (Note that someone else might call the police if I don’t, so I should discount the value of calling the police somewhat if it turns out to be positive.)

If the police arrest him, they’ll have to do all the usual things they do when they arrest people, which seem pretty expensive. Maybe they’ll injure him while he’s being arrested—I know that police killing people is rare, but police injuring people might be relatively common. Maybe they’ll make him go to court. I think you can go to jail for a while for stealing a bike, especially if you’ve stolen bikes before.

I think the outcomes where the police arrest the guy are worse than the outcomes where he steals a bike. All that legal stuff sounds pretty expensive, especially jail.

However, if we don’t arrest this guy now, the counterfactual isn’t just him stealing this bike. It’s him stealing all the bikes he’s going to steal in the future that he’d be prevented from stealing by being arrested.

How bad is stealing bikes? When X steals a bike from Y, then Y loses the original sale price of the bike, plus extra inconvenience from having to replace a bike or deal with arriving at a train station and realizing they don’t have a bike. And X gains the second-hand sale price of the bike, minus the value of their time stealing the bike and then selling it. I think that when bike thieves sell stolen bikes, they probably lose a bunch of the value of the bike. So there are losses to society involved in the theft—the price of the destroyed lock, the difference between the value that the bike owner places on the bike and the value that the thief gets for the bike.

Bike thieves are probably poorer than bike owners though. If we think utility varies as log income, then we just need to compare the percentage of income that the theft affects for both the thief and the owner. If half the value of a bike is destroyed while stealing it, then this is a net positive “transaction” if the bike thief makes less than half as much money as the original owner did.

If this guy gets arrested or has a close call, maybe him and his bike-thief friends will decide to steal fewer bikes. What will they do instead? Will they just live with the reduced income, or get jobs at McDonalds, or use their new spare time to volunteer in museums, or sell heroin to children? How likely are those outcomes and how bad are they?

What are the externalities of stealing bikes? Maybe more people will start to steal bikes. This would result in cyclists having to spend more money on anti-theft devices, or maybe people cycling less. Maybe cyclists will have less desire to live in Oakland. Maybe the crime statistics for the region will rise and house prices will be depressed. Crime is pretty bad for a neighborhood, I hear? Maybe reducing crime is really valuable and I should hope to reduce theft for that reason.

Are there any other reasons to want to jail bike thieves? Maybe they’re also inclined to other crimes which have higher costs to society, like murder or robbery?

According to the internet, it costs about $30k per year to keep someone in jail. If the thief goes to jail and doesn’t steal any bikes while in jail, then society gets the benefit of not as many bikes being stolen. If the thief stole one $400 bicycle per day and destroyed half its value, that would be $73k saved by them being in jail. That seems like suspiciously high income (and tax-free!), so maybe stealing bikes isn’t actually that lucrative. Maybe bike thieves try to steal as few bikes as they can? If they steal only as many bikes as they have to (and stealing bikes destroys half the value), then it’s only good to jail them if their income is greater than $30k a year.

… And so on.

I think that thinking like this is a really useful skill. I don’t quite remember how I acquired it. I think LessWrong was really helpful, as was reading and participating in discussions about EA various places on the internet, and also going to the Stanford EA meetups. Other great sources of this kind of reasoning include Slate Star Codex, Overcoming Bias, and Meteuphoric.

I wish that my 17 year old self had taken a class on this kind of reasoning. I think you could teach a class like that. It would involve a little bit of economics, and an explanation of counterfactual reasoning and expected values, and most of the assessment would involve writing down reasoning along the lines of the stuff I wrote above.

This kind of reasoning is very useful for cause prioritization. It also seems to be a kind of reasoning that GiveWell does a lot.

Maybe someone should offer to help train people in this kind of reasoning. You could give people questions along the lines of the one I discussed above, and then get them to write up their thoughts, and then comment on them. Maybe this would be valuable for helping EAs get better at thinking about EA topics.

Is there a better name for this kind of reasoning? Are there any other good resources for learning it? What are the best examples of this kind of reasoning being laid out extremely explicitly in an EA context?

Turns out that this example of this kind of reasoning really pisses people off. You can view the comment shitstorm here.