|Friday, February 24|
|PS2 Poster Session 2 and Refreshments||
Fri, Feb 24, 5:15 PM - 6:30 PM
Conference Center AB
Expanding the Appeal of Model Selection Using Mixture Priors to Incorporate Expert Opinion: A Behavioral Economic Case Study (303451)*Christopher T Franck, Virginia Tech
Keywords: model selection, mixture prior, expert opinion
The applied statistician frequently encounters resistance to model selection. Resistance may occur because off-the-shelf selection approaches treat candidate models indifferently with respect to underlying scientific theory, which is unappealing to non-statisticians. Researchers often prefer domain-specific theoretical justification for model choice instead of selection routines that are perceived as naïve to these considerations. Nonetheless, when statistical metrics reveal data may be less compatible with scientific theory than a researcher imagines, tension can permeate both technical planning and interpersonal communication on the research team.
In this presentation I will compare a historical model of delay discounting with a recent model that forfeits some theoretical elegance to achieve better model fit. Delay discounting describes the subjective rate at which a reward loses its value as a function of delay, and is an important behavioral economic marker of drug addition, alcoholism, excessive gambling, and a variety of other problematic behaviors. The mixture prior approach is shown to incorporate clients’ expertise into the model selection framework.