A BRAND CHOICE MODEL USING MULTINOMIAL LOGISTICS REGRESSION, BAYESIAN INFERENCE AND MARKOV CHAIN MONTE CARLO METHOD

Deshmukh Sachin¹, Manjrekar Pradip²*, Gopal R.^3
1Department of Business Management, Padmashree Dr. D. Y. Patil University, Navi Mumbai, 410614
²Research & Consultancy & Extension Centre, Department of Business Management, Padmashree Dr. D. Y. Patil University, Navi Mumbai, 410614
³Department of Business Management, Padmashree Dr. D. Y. Patil University, Navi Mumbai, 410614
* Corresponding Author : drpradipm@gmail.com

Received : -     Accepted : -     Published : 15-06-2010
Volume : 1     Issue : 1       Pages : 1 - 28
Int J Econ Bus Model 1.1 (2010):1-28
DOI : http://dx.doi.org/10.9735/0976-531X.1.1.1-28

Keywords : Brand Choice Model, Multinomial Logistics Regression(MNL), Bayesian Inference, Markov Chain Monte Carlo (MCMC) Simulation, WinBUGS Software
Conflict of Interest : None declared

This paper explains a Brand Choice model using Multinomial Logistics Regression(MNL), Bayesian Inference and Markov Chain Monte Carlo (MCMC) Simulation method. The model enables a marketer to forecast probability of choosing a particular brand by a consumer at a purchase occasion. Three hundred and twenty households in one tier (Mumbai, Pune), two tier (Kolhapur, Nasik, Nagpur), and three tier (Parbhani, Latur, Amravati) cities/towns of Maharashtra participated in the survey. Data were collected on three leading detergent brands – Surf, Ariel and Rin. The basis for our brand choice model is the multinomial logistics regression which was then converted into a software program. The parameters of the model are then estimated with the help of WinBUGS (Windows Version for Bayesian Inference Using Gibbs Sampling) software that uses Bayesian Analysis and Markov Chain Monte Carlo method. The utility function in the model used five attributes – a dummy term for brand2 and brand3 each, price term giving the price in rupees, a “feature’ term (advertisement) that was “1” when the respondents were exposed to the advertisement between two purchases, otherwise a “0” was assigned, the promotion term taking values “1” or “0” when the product was specially promoted/not promoted. The calculated parameters are further used to estimate the probability for a brand which is priced at Rs. 55. A marketer can thus forecast the probability of purchase of a brand at different price levels.

[1] Bettman James R., Mary Francis Luce, and John Payne (1998) Journal of Consumer Research, 187 – 217.  
» CrossRef   » Google Scholar   » PubMed   » DOAJ   » CAS   » Scopus  

[2] Bettman James R.(1998) Journal of Marketing Research Vol 4, pp 465 – 471  
» CrossRef   » Google Scholar   » PubMed   » DOAJ   » CAS   » Scopus  

[3] Bettman James R. and Michel A. Zins (1977) Journal of Consumer Research, Vol. 4, pp. 75-85.  
» CrossRef   » Google Scholar   » PubMed   » DOAJ   » CAS   » Scopus  

[4] Geman S., and Geman D. (1984) IEEE Transactions of Pattern Analysis and Machine Intelligence 6:721–741  
» CrossRef   » Google Scholar   » PubMed   » DOAJ   » CAS   » Scopus  

[5] Metropolis Nicholas, and Ulam S. (1949) Journal of the American Statistical Association 44:335–341  
» CrossRef   » Google Scholar   » PubMed   » DOAJ   » CAS   » Scopus  

[6] Hastings W. K. (1979) Biometrika 57:97–109.  
» CrossRef   » Google Scholar   » PubMed   » DOAJ   » CAS   » Scopus  

[7] Chib Siddhartha, and Edward Greenberg (1995) The American Statistician 49:327–336  
» CrossRef   » Google Scholar   » PubMed   » DOAJ   » CAS   » Scopus  

[8] Cowles M.K. and Carlin B.P. (1995) Journal of the American Statistical Association 91:883–904  
» CrossRef   » Google Scholar   » PubMed   » DOAJ   » CAS   » Scopus  

[9] Brooks S. P. and Roberts G.O. (1998) Statistics and Computing 8:319–335  
» CrossRef   » Google Scholar   » PubMed   » DOAJ   » CAS   » Scopus  

[10] Jackman Simon.( 2004) Annual Reviews of Political Science 7:483– 505.  
» CrossRef   » Google Scholar   » PubMed   » DOAJ   » CAS   » Scopus  

[11] Payne John W. (1982) Psychological Bulletin, Vol. 92, pp. 382-402  
» CrossRef   » Google Scholar   » PubMed   » DOAJ   » CAS   » Scopus  

[12] Brooks S. P., and Roberts G.O. (1998) Statistics and Computing 8:319–335  
» CrossRef   » Google Scholar   » PubMed   » DOAJ   » CAS   » Scopus  

[13] Ionnis Ntzoufras (2008) Bayesian Modeling Using WinBUGS, John Wiley & Sons Publication U.S.A. pp 1 – 150  
» CrossRef   » Google Scholar   » PubMed   » DOAJ   » CAS   » Scopus