#44: Base Rates of Error
Whenever you have some key measurement that you try to optimize, it is almost always the case that the optimal measurement is not quite 100%. Take the Unemployment Rate: the policy goal is not 100% employment (or 0% unemployment). The policy goal is something like 2-5% unemployment, reflecting some workplace mobility and the usual employment turnover due to the business cycle. If you think about what it would take to reach 0% unemployment, chances are some weird, dystopian thing would be going on that nobody wants.1
Some years ago, I watched an interview with Peter Thiel, wherein he suggested that the US doesn’t have enough corruption. At first, I wrote this off as an ill-advised view, but as I reflect, maybe it is good to have just a tiny bit of corruption that allows people to grease the wheels and bypass the slow-and-ossified regulatory system here and there. It acts as an escape hatch to overwrought legal process. Don’t get me wrong: I’m not suggesting that we need more corruption in the US; I think we probably have too much corruption, but the optimal rate also isn’t zero. If it were zero, again, chances are some weird, dystopian thing would be going on.2
As we’ve all seen the FTX disaster unfold in recent weeks, there has been much discourse about how skilled investors like Sequoia, Lightspeed, etc. could invest in such a thoroughly rotten fraud, with no board governance to boot. Smart commentators point out that if venture capitalists performed tighter scrutiny, they would likely be too conservative and miss great companies — even if they managed to avoid the occasional embarrassment, their overall returns would likely be worse. If Fund X invests aggressively in emerging moonshot bets and 1% of that is lost to fraud and the remaining 99% returns several multiples on the fund, I’d think that’s pretty good. Even in investing, the optimal rate of fraud is not zero.
This stuff is everywhere: when you’re hiring employees, you also won’t get it right 100% of the time. Some small percentage of staff will not perform, you’ll need to let them go, and that’s just life.
People often recriminate about getting it wrong: in the hiring context if someone doesn’t work out, in the investing context if there’s a fraud, and in many other contexts when the bad thing that you’re trying to avoid happens. But it’s important to remember that every one of these metrics has some base rate of the thing you’re trying to avoid simply happening, and it’s unavoidable unless you try unreasonably hard. I think this notion of optimal, non-zero base rates is interesting. Keep it in mind the next time that you see something go awry and ask: what’s the base rate of error that we expect and should treat as unavoidable?
Thank you for reading! Follow me on Twitter, or subscribe for free to my Substack:
For example, a crackdown on employment mobility laws, slavery, or legal requirements that everyone must have a job, even if they don’t need one.
You can imagine some scenario that involves the exhaustive monitoring of all communications and all financial transactions.