Eliminating the Outside Good Bias in Logit Models of Demand with Aggregate Data

Huang Dongling () and Rojas Christian ()
Additional contact information
Huang Dongling: The Lally School of Management, Rensselaer Polytechnic Institute, Troy, NY 12180-3590, USA
Rojas Christian: Department of Resource Economics, University of Massachusetts Amherst, 219A Stockbridge Hall, Amherst, MA 01003, USA

Review of Marketing Science, 2014, vol. 12, issue 1, 1-36

Abstract: The logit model is the most popular tool in estimating demand for differentiated products. In this model, the outside good plays a crucial role because it allows consumers to stop buying the differentiated good altogether if all brands simultaneously become less attractive (e.g. if a simultaneous price increase occurs). But practitioners lack data on the outside good when only aggregate data are available. The currently accepted procedure is to assume a “market potential” that implicitly defines the size of the outside good (i.e. the number of consumers who considered the product but did not purchase); in practice, this means that an endogenous quantity is approximated by a reasonable guess thereby introducing the possibility of an additional source of error and, most importantly, bias. We provide two contributions in this paper. First, we show that structural parameters can be substantially biased when the assumed market potential does not approximate the outside option correctly. Second, we show how to use panel data techniques to produce unbiased structural estimates by treating the market potential as an unobservable in both the simple and the random coefficients logit demand model. We explore three possible solutions: (a) controlling for the unobservable with market fixed effects, (b) specifying the unobservable to be a linear function of product characteristics, and (c) using a “demeaned regression” approach. Solution (a) is feasible (and preferable) when the number of goods is large relative to the number of markets, whereas (b) and (c) are attractive when the number of markets is too large (as in most applications in Marketing). Importantly, we find that all three solutions are nearly as effective in removing the bias. We demonstrate our two contributions in the simple and random coefficients versions of the logit model via Monte Carlo experiments and with data from the automobile and breakfast cereals markets.

Keywords: logit model; random coefficients; demand estimation; market potential; outside good; differentiated products (search for similar items in EconPapers)
JEL-codes: C15 C82 D12 D43 (search for similar items in EconPapers)
Date: 2014
References: View references in EconPapers View complete reference list from CitEc
Citations:

Downloads: (external link)
https://doi.org/10.1515/roms-2013-0016 (text/html)
For access to full text, subscription to the journal or payment for the individual article is required.

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:bpj:revmkt:v:12:y:2014:i:1:p:36:n:1

Ordering information: This journal article can be ordered from
https://www.degruyter.com/journal/key/roms/html

DOI: 10.1515/roms-2013-0016

Access Statistics for this article

Review of Marketing Science is currently edited by Ram C. Rao

More articles in Review of Marketing Science from De Gruyter
Bibliographic data for series maintained by Peter Golla ().