Regressing observed rates on self-reported frequencies measured in two stages

[dholstius]

@Bob_Carpenter

This is another measurement error problem that can be handled the same way. You assume there’s a true number of logs, and then you get a report like “10 logs” being more common than “9 logs”. The trick is always figuring out how to estimate this stuff, but I like to follow the advice I got from @andrewgelman, who said he got it from Don Rubin: think about what you would do if you had all the data, then just turn the Bayesian crank to infer the rest

Following this up: Custom function to reallocate probability mass from certain integers to others. Thanks!