## An empirical comparison of logit choice models with discrete versus continuous representations of heterogeneity (2002)

Venue: | Journal of Marketing Research |

Citations: | 12 - 0 self |

### BibTeX

@ARTICLE{Andrews02anempirical,

author = {Rick L. Andrews and Andrew Ainslie and Imran S. Currim},

title = {An empirical comparison of logit choice models with discrete versus continuous representations of heterogeneity},

journal = {Journal of Marketing Research},

year = {2002},

pages = {479--487}

}

### OpenURL

### Abstract

Currently, there is an important debate about the relative merits of models with discrete and continuous representations of consumer heterogeneity. In a recent JMR study, Andrews, Ansari, and Currim (2002; hereafter AAC) compared metric conjoint analysis models with discrete and continuous representations of heterogeneity and found no differences between the two models with respect to parameter recovery and prediction of ratings for holdout profiles. Models with continuous representations of heterogeneity fit the data better than models with discrete representations of heterogeneity. The goal of the current study is to compare the relative performance of logit choice models with discrete versus continuous representations of heterogeneity in terms of the accuracy of household-level parameters, fit, and forecasting accuracy. To accomplish this goal, the authors conduct an extensive simulation experiment with logit models in a scanner data context, using an experimental design based on AAC and other recent simulation studies. One of the main findings is that models with continuous and discrete representations of heterogeneity recover household-level parameter estimates and predict holdout choices about equally well except when the number of purchases per household is small, in which case the models with continuous representations perform very poorly. As in the AAC study, models with continuous representations of heterogeneity fit the data better.

### Citations

2311 |
Statistical power analysis for the behavioral sciences (2nd ed.). USA: Lawrence Erlbaum Associates
- Cohen
- 1988
(Show Context)
Citation Context ...is design to detect small differences between model types (a small effect accounts for approximately 1% of the variance of the dependent variable) at a significance level of .05 is approximately 81% (=-=Cohen 1988-=-). For factors with two levels (1, 2, 3, 4, and 7), the power to detect small differences at the .05 level is approximately 92%. For factors with three levels (5 and 6), the power to detect small diff... |

377 |
A Method for Minimizing the Impact of Distributional Assumptions in Econometric Models
- Heckman, Singer
- 1984
(Show Context)
Citation Context ...ifference between components is larger. There is speculation that models with continuous distributions are susceptible to misspecification of the functional form of the distribution of heterogeneity (=-=Heckman and Singer 1984-=-; Wedel et al. 1999). Our third experimental factor produces data sets with normal or gamma distributions of heterogeneity. Whereas normal distributions produce bellshaped curves, gamma distributions ... |

327 |
Estimating the dimension of a model
- Schwartz
- 1978
(Show Context)
Citation Context ... the FM and logit models, BIC was used as an approximation to the marginal likelihood. BIC is based on the Schwartz criterion, which in turn is an approximation to the marginal likelihood of a model (=-=Schwartz 1978-=-). For the HB model, the marginal likelihood was computed using the reweighted importance sampling method outlined in 10Raftery (1996) and Newton and Raftery (1994). In order to make this comparable ... |

129 |
A Probabilistic Choice Model for Market Segmentation and Elasticity Structure
- Kamakura, Russell
- 1989
(Show Context)
Citation Context ...ay not have any advantage over models with less sophisticated representations and may even perform worse if spurious heterogeneity effects or overfitting occurs. In this study, FM logit models (e.g., =-=Kamakura and Russell 1989-=-), which have been applied extensively to scanner panel data, are used to recover discrete distributions of heterogeneity. We use mixed logit models estimated with HB estimation methods to recover con... |

100 | Forecasting new product penetration with flexible substitution patterns
- Brownstone, Train
- 1999
(Show Context)
Citation Context ...ods to recover continuous distributions of heterogeneity. Mixed logit models, also known as random coefficients logit models, have received considerable attention in marketing and econometrics (e.g., =-=Brownstone and Train 1999-=-; Chintagunta, Jain, and Vilcassim 1991; Elrod 1988; Erdem 1996; Gönül and Srinivasan 1993; Revelt and Train 1998). McFadden and Train (2000) establish that, under mild regularity conditions, mixed lo... |

88 |
Marketing models of consumer heterogeneity
- Allenby, Rossi
- 1998
(Show Context)
Citation Context ... discrete choice model derived from random utility maximization. In the past decade, HB-estimated models have emerged with much potential for representing heterogeneity in consumer preferences (e.g., =-=Allenby and Rossi 1999-=-; McCulloch and Rossi 1994; Rossi, McCulloch, and Allenby 1996). Collectively, the FM model and mixed logit model estimated with HB represent the state of the art for representing heterogeneity in log... |

80 | Approximate Bayesian inference by the weighted likelihood bootstrap (with discussion - Newton, Raftery - 1994 |

51 | 2000): “Mixed MNL Models for Discrete Response - McFadden, Train |

44 | Hypothesis testing and model selection - Raftery - 1996 |

38 | Investigating heterogeneity in brand preferences in logit models for panel data - Chintagunta, Jain, et al. - 1992 |

20 |
Choice map: Inferring a product-market map from panel data
- Elrod
- 1988
(Show Context)
Citation Context ...git models, also known as random coefficients logit models, have received considerable attention in marketing and econometrics (e.g., Brownstone and Train 1999; Chintagunta, Jain, and Vilcassim 1991; =-=Elrod 1988-=-; Erdem 1996; Gönül and Srinivasan 1993; Revelt and Train 1998). McFadden and Train (2000) establish that, under mild regularity conditions, mixed logit can approximate as closely as needed the choice... |

15 | Metric Conjoint Segmentation Methods: A Monte Carlo Comparison," The journal of Marketing Research - Vriens, Wedel, et al. - 1996 |

14 |
Modeling Multiple Sources of Heterogeneity
- Gönül, Srinivasan
- 1993
(Show Context)
Citation Context ...s random coefficients logit models, have received considerable attention in marketing and econometrics (e.g., Brownstone and Train 1999; Chintagunta, Jain, and Vilcassim 1991; Elrod 1988; Erdem 1996; =-=Gönül and Srinivasan 1993-=-; Revelt and Train 1998). McFadden and Train (2000) establish that, under mild regularity conditions, mixed logit can approximate as closely as needed the choice probabilities of any discrete choice m... |

8 | Parameter bias from unobserved effects in the multinominal logit model of consumer choice - Abramson, Andrews, et al. - 2000 |

7 | Imran Currim (2002) ”Hierarchical Bayes versus finite mixture conjoint analysis models: a comparison of fit, prediction, and partworth recovery - Andrews, Ansari - 2002 |

7 |
Information-based Validity Functionals for Mixture Analysis
- CUTLER, WINDHAM
- 1993
(Show Context)
Citation Context ...zes of the components are also generated randomly for each data set, and constraints are set up such that no component is smaller than 10% of the sample and none is larger than 90% of the sample (see =-=Cutler and Windham 1994-=-). 6 By generating the parameters independently for each data set, we have more assurance that the simulation results generalize beyond a specific set of parameter values chosen by the researcher. Reg... |

7 |
A Dynamic Analysis of
- Erdem
- 1996
(Show Context)
Citation Context ...also known as random coefficients logit models, have received considerable attention in marketing and econometrics (e.g., Brownstone and Train 1999; Chintagunta, Jain, and Vilcassim 1991; Elrod 1988; =-=Erdem 1996-=-; Gönül and Srinivasan 1993; Revelt and Train 1998). McFadden and Train (2000) establish that, under mild regularity conditions, mixed logit can approximate as closely as needed the choice probabiliti... |

6 |
An exact likelihood approach to analysis of the MNP model
- McCulloch, Rossi
- 1994
(Show Context)
Citation Context ...erived from random utility maximization. In the past decade, HB-estimated models have emerged with much potential for representing heterogeneity in consumer preferences (e.g., Allenby and Rossi 1999; =-=McCulloch and Rossi 1994-=-; Rossi, McCulloch, and Allenby 1996). Collectively, the FM model and mixed logit model estimated with HB represent the state of the art for representing heterogeneity in logit choice models. Although... |

5 |
Mixed Logit with Repeated Choices of Appliance Efficiency Levels
- Train, K
- 1998
(Show Context)
Citation Context ... models, have received considerable attention in marketing and econometrics (e.g., Brownstone and Train 1999; Chintagunta, Jain, and Vilcassim 1991; Elrod 1988; Erdem 1996; Gönül and Srinivasan 1993; =-=Revelt and Train 1998-=-). McFadden and Train (2000) establish that, under mild regularity conditions, mixed logit can approximate as closely as needed the choice probabilities of any discrete choice model derived from rando... |

2 | Currim (2003), “A Comparison of Segment Retention Criteria for Finite Mixture Logit Models - Andrews, Imran |

2 | 2001), “On the Similarity of Classical and - Huber, Train |

1 | Currim (2001), “A Comparison of Segment Retention Criteria for Finite Mixture Models,” working paper - Andrews, L, et al. |

1 | Currim (2002), “Hierarchical Bayes vs. Finite Mixture Conjoint Analysis Models: A Comparison of Fit, Prediction, and Partworth Recovery - Ansari, Imran |

1 | A Dynamic Analysis of - Tülin - 1996 |