The Generalized Multinomial Logit Model: Accounting for Scale and Coefficient Heterogeneity

By Denzil G. Fiebig, Michael P. Keane, Jordan J. Louviere, Nada Wasi. Marketing science (Providence, R.I.), 2010

888 citations recorded by the source corpus.

Subfields: Economics and Econometrics, General Decision Sciences, Marketing

View Paper: https://www.semanticscholar.org/paper/517602de2e5292e094726cd4fab7b6fa1a61f4f7

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Economics and Econometrics

In General Decision Sciences

In Marketing

Citations in the corpus

Visual Attention and Attribute Attendance in Multi-Attribute Choice Experiments (2015, background): “There has been considerable interest in scale heteroscedasticity (e.g. Fiebig et al., 2010).”
Can scale and coefficient heterogeneity be separated in random coefficients models? (2012, methodology): “…relative sensitivities (which would arise if all heterogeneity was caused by scale differences) may be more common than previously thought and that differences in estimated utility parameters are primarily the result of scale differences (see e.g. Swait and Bernardino, 2000; Fiebig et al., 2010).”
Latent class structures: taste heterogeneity and beyond (2014, methodology): “Correlation is rarely introduced in models not based on the Normal distribution, one exception being given in Walker (2001), while flexible correlation structures in continuous mixtures are also a benefit of the GMNL specification of mixed logit (Fiebig et al., 2010).”
The value of supply security: The costs of power outages to Austrian households, firms and the public sector (2013, methodology): “The poll participants' decisions were then econometrically assessed by means of a censored random coefficients model (Reichl and Frühwirth-Schnatter, 2012) accounting for scale heterogeneity (Fiebig et al. 2010) and willingness to pay was inferred following McFadden (1996).”
Discrete Choice Experiments Are Not Conjoint Analysis (2010, background): “, Fiebig et al. 2010; Swait and Louviere 1993). The parameter estimate-scale confound has profound implications for modeling many choices, and starkly differentiates DCEs from traditional CA. For example, BenAkiva and Morikawa (1990) reasoned that if the underlying preference process was the same fo”
The maximum approximate composite marginal likelihood (MACML) estimation of multinomial probit-based unordered response choice models (2011, methodology): “Keywords: multinomial probit, mixed models, composite marginal likelihood, discrete choice models, spatial econometrics, panel data.”
We want to sort! – assessing households’ preferences for sorting waste (2014, positive): “In the G-MNL model each respondent’s utility was assumed to be a linear function of the same alternative-specific attribute levels, however unlike in the MNL model, we now allowed each of the utility function coefficients to follow a normal distribution, to account for unobserved preference heteroge”
Correlation and scale in mixed logit models (2017, methodology): “The most prominent model seeking to capture scale heterogeneity is the “generalized multinomial logit model”, or G-MNL, introduced by (Fiebig et al., 2010; Greene and Hensher, 2010; Keane, 2006).”
Consumer willingness to pay for traditional food products (2016, methodology): “As b and r cannot be separately identified, Fiebig et al. (2010) propose the following general specification of the scale parameter: rn 1⁄4 expð rþ hzn þ se0nÞ, where e0n Nð0;1Þ and zn is a vector of individual characteristics, with r 1⁄4 s2=2 so that EðrnÞ 1⁄4 1, when h 1⁄4 0.”
The Composite Marginal Likelihood (CML) Inference Approach with Applications to Discrete and Mixed Dependent Variable Models (2014, methodology): “…coefficients, almost to the point that the terms mixed logit or mixed GEV or mixed probit are oftentimes used synonymously with normal mixing (see Fiebig et al., 2010, Dube et al., 2002).12 In the context of the normal error distributions just discussed, the use of a GEV kernel structure leads…”
Importance Sampling Squared for Bayesian Inference in Latent Variable Models (2013, methodology): “We consider an empirical application to the Pap smear data set used for simulated maximum likelihood estimation in Fiebig et al. (2010).”
Discrete Choice Experiments: A Guide to Model Specification, Estimation and Software (2017, background): “There are error component versions of MXL [25] where the motivation comes from the need to induce correlation and heteroscedasticity across alternatives rather than from preference heterogeneity.”
The impact of attribute preferences on adoption timing: The case of photo-voltaic (PV) solar cells for household electricity generation (2013, methodology): “We follow the model proposed by Fiebig et al. (2010), the Generalized Multinomial Logit Model (G-MNL).”