|
|
During the past 30+ years there has been an accumulating body of empirical research
in the behavioral sciences concerning probabilistic judgments. In particular researchers
have focused on the (mis)calibration , i.e. the relationship between the judgments and
the probability of an event, and the discrimination, i.e. the relationship between the
judgments and the realization of an event of probabilistic judgments. Existing
statistical techniques for analyzing the calibration and discrimination of probability
judgments rely on the analysis of derived variables. I will present a general framework
using mixed-effects regression models to directly model probability judgments and analyze
their calibration and discrimination. In particular I will focus on the use of an ordinal
probit mixed-effects model for the log-odds of probabilisitic judgments. My discussion
will focus largely on practical matters such as model specification and inference.
|
|