Model choice and framing effects: Do discrete choice modeling decisions affect loss aversion estimates?

Abstract

This paper examines whether the presence and magnitude of estimated loss aversion (LA) in a discrete choice experiment is a function of modeling choice. The experiment examined preferences for utility-scale solar energy siting based on a series of installation attributes and changes in household electric bill (the payment vehicle, which can increase or decrease relative to the status-quo). We employ multiple discrete choice modeling approaches and show that, across all models, the implications of accounting for loss aversion are qualitatively similar and match theoretical predictions. Despite this similarity, when comparing results across models we find that model choice has substantial impacts on estimated loss aversion. Specifically, different models estimate loss/gain ratios below two and in excess of six for the same data set. Thus, the consequences of framing decisions, which are an important aspect of nonmarket valuation, are not just the provenance of survey and choice experiment design but may also be heavily influenced by empirical model choice.