Model selection for the mixed logit with Bayesian estimation
Kelvin Balcombea, Ali Chalakb and Iain Fraserc, ,
aAgricultural and Food Economics, University of Reading, UK
bImperial College, London, UK
cKent Business School, University of Kent, Wye Campus, Wye, Kent TN25 5AH, UK
Received 17 August 2007.
Available online 2 October 2008.
AbstractIn this paper, the mixed logit (ML)using Bayesian methods was employed to examine willingness-to-pay (WTP)to consume bread produced with reduced levels of pesticides so as toameliorate environmental quality, from data generated by a choiceexperiment. Model comparison used the marginal likelihood, which ispreferable for Bayesian model comparison and testing. Models containingconstant and random parameters for a number of distributions wereconsidered, along with models in ‘preference space’ and ‘WTP space’ aswell as those allowing for misreporting. We found: strong support forthe ML estimated in WTP space; little support for fixing the pricecoefficient a common practice advocated and adopted in theenvironmental economics literature; and, weak evidence for misreporting.
Keywords: Mixed logit; Willingness-to-pay; Model comparison
JEL classification codes: C11; C25; C52; L92; Q51
Article Outline1. Introduction2. Issues arising from the literature3. The model3.1. General model specification3.2. Misreporting3.3. Priors3.4. Full-data likelihood, the likelihood and marginal likelihood4. Model estimation4.1. Choice of priors4.2. Calculating the marginal likelihood5. Empirical section5.1. Data5.2. Model results5.3. WTP estimates6. Summary and discussionAcknowledgementsReferences