When zeroth order approximation turned out to be just reference class forecasting, I expected first order approximation to be reference class forecasting modified in a simple mathematical way (such as multiplication) by an inside-view adjustment factor. E.g. if the last 3 homework assignments all took 12-16 hours, and this assignment seems significantly harder than any of those, but not as hard as any two of them put together, maybe it will take 22 hours. (Implicitly, that might mean I scored the relative time-consumingness of the previous three assignments as 6, 7, and 8, and this one as an 11.)
This seems similar to what first order actually ended up meaning in the article, but the article seemed to focus on cases where the inside-view adjustment factor is very objective. In the COVID example, it was just “what’s the date?” Objective adjustment factors are great when they’re available and related in a simple way to the outcome, but subjective adjustment factors like the “homework time-consumingness factor” still seem like they could be significantly better than either (1) making a prediction without using any reference class information at all, or (2) using only the reference class information and ignoring clear differences from the reference class rather than trying to adjust for them.
When zeroth order approximation turned out to be just reference class forecasting, I expected first order approximation to be reference class forecasting modified in a simple mathematical way (such as multiplication) by an inside-view adjustment factor. E.g. if the last 3 homework assignments all took 12-16 hours, and this assignment seems significantly harder than any of those, but not as hard as any two of them put together, maybe it will take 22 hours. (Implicitly, that might mean I scored the relative time-consumingness of the previous three assignments as 6, 7, and 8, and this one as an 11.)
This seems similar to what first order actually ended up meaning in the article, but the article seemed to focus on cases where the inside-view adjustment factor is very objective. In the COVID example, it was just “what’s the date?” Objective adjustment factors are great when they’re available and related in a simple way to the outcome, but subjective adjustment factors like the “homework time-consumingness factor” still seem like they could be significantly better than either (1) making a prediction without using any reference class information at all, or (2) using only the reference class information and ignoring clear differences from the reference class rather than trying to adjust for them.