How I repeatedly failed to use Tobit modelling on censored data — LessWrong