Investigating cost non‐attendance as a driver of inflated welfare estimates in mixed‐logit models

Research output: Contribution to journalArticlepeer-review

2 Citations (Scopus)

Abstract

Choice models are used by applied economists for many purposes, such as non-market valuation or estimating willingness to pay for novel food and product attributes. Mixed-logit models allow researchers to account for preference heterogeneity and complex decision-making processes when modelling choices. In mixed-logit models, parameters of monetary attributes such as prices typically are assumed to follow a negative lognormal random distribution to ensure that the marginal utility of a price increase is strictly negative. However, this practice can cause means and standard deviations of welfare estimates to ‘explode’ to unfeasibly large levels, as the model assumes there are some marginal utilities of cost approaching zero. This paper examines whether cost non-attendance, which occurs when respondents ignore costs in stated-preference studies, could be a cause of inflated welfare estimates when a lognormal cost parameter is used. A two-class equality-constrained latent-class model is proposed, in which the cost parameter is fixed at zero for a cost non-attender class and is specified as a random lognormal parameter for cost attenders. This proposed model produces mean welfare estimates that are 17 times lower than a mixed-logit model with a lognormal cost parameter, and 10% lower than a model with a non-random cost parameter. These results suggest that cost non-attendance can result in inflated welfare estimates when employing a lognormal cost parameter, and that accounting for cost non-attendance could be a simple, parsimonious solution to this problem.
Original languageEnglish
Pages (from-to)921-934
Number of pages14
JournalJournal of Agricultural Economics
Volume74
Issue number3
DOIs
Publication statusPublished - Sept 2023

Fingerprint

Dive into the research topics of 'Investigating cost non‐attendance as a driver of inflated welfare estimates in mixed‐logit models'. Together they form a unique fingerprint.

Cite this