Subtopic Deep Dive

Discrete Choice Models of Consumer Brand Selection
Research Guide

What is Discrete Choice Models of Consumer Brand Selection?

Discrete choice models of consumer brand selection use multinomial logit (MNL) and probit models with random coefficients to analyze attribute-based brand choices from scanner data.

These models account for consumer heterogeneity through mixed logit specifications. McFadden and Train (2000) established that mixed MNL (MMNL) models approximate any random utility maximization model (3997 citations). Erdem and Keane (1996) extended them to dynamic brand choice under uncertainty (1253 citations).

Curated Papers

Key Challenges

Why It Matters

Firms use these models for demand forecasting and optimal pricing in new product launches. McFadden and Train (2000) enable segmentation by estimating individual-level preferences from aggregate data. Erdem and Keane (1996) quantify learning from advertising and experience, informing marketing budgets. Crawford and Shum (2005) apply them to pharmaceutical demand, revealing uncertainty effects on brand switching (440 citations).

Key Research Challenges

Heterogeneity Identification

Estimating random coefficients requires simulating individual draws, computationally intensive for large scanner datasets. McFadden and Train (2000) prove universal approximation but note simulation bias risks. Advances demand efficient maximum simulated likelihood methods.

Dynamic Uncertainty Modeling

Incorporating learning from usage and advertising creates high-dimensional state spaces. Erdem and Keane (1996) model noisy signals but face inference challenges from sparse purchase data. Bayesian approaches help but increase estimation complexity.

Endogeneity in Attributes

Price endogeneity from unobserved promotions biases utility estimates. Crawford and Shum (2005) address this via dynamic matching but require rich instrumental variables. Scanner data limitations persist for unobserved brand characteristics.

Essential Papers

Mixed MNL models for discrete response

Daniel McFadden, Kenneth Train · 2000 · Journal of Applied Econometrics · 4.0K citations

This paper considers mixed, or random coefficients, multinomial logit (MMNL) models for discrete response, and establishes the following results. Under mild regularity conditions, any discrete choi...

Decision-Making Under Uncertainty: Capturing Dynamic Brand Choice Processes in Turbulent Consumer Goods Markets

Tülin Erdem, Michael P. Keane · 1996 · Marketing Science · 1.3K citations

We construct two models of the behavior of consumers in an environment where there is uncertainty about brand attributes. In our models, both usage experience and advertising exposure give consumer...

The Economics of Two-Sided Markets

Marc Rysman · 2009 · The Journal of Economic Perspectives · 1.2K citations

Broadly speaking, a two-sided market is one in which 1) two sets of agents interact through an intermediary or platform, and 2) the decisions of each set of agents affects the outcomes of the other...

Agency Selling or Reselling? Channel Structures in Electronic Retailing

Vibhanshu Abhishek, Kinshuk Jerath, Z. John Zhang · 2015 · Management Science · 1.0K citations

In recent years, online retailers (also called e-tailers) have started allowing manufacturers direct access to their customers while charging a fee for providing this access, a format commonly refe...

Progress in partial least squares structural equation modeling use in marketing research in the last decade

Marko Sarstedt, Joseph F. Hair, Mandy Pick et al. · 2022 · Psychology and Marketing · 864 citations

Abstract Partial least squares structural equation modeling (PLS‐SEM) is an essential element of marketing researchers' methodological toolbox. During the last decade, the PLS‐SEM field has undergo...

Search, Obfuscation, and Price Elasticities on the Internet

Glenn Ellison, Sara Fisher Ellison · 2009 · Econometrica · 562 citations

We examine the competition between a group of Internet retailers who operate in an environment where a price search engine plays a dominant role. We show that for some products in this environment,...

Coordination and Lock-In: Competition with Switching Costs and Network Effects

Joseph Farrell, Paul Klemperer · 2006 · SSRN Electronic Journal · 486 citations

Reading Guide

Foundational Papers

Start with McFadden and Train (2000) for MMNL theoretical foundations, then Erdem and Keane (1996) for dynamic applications to brand learning.

Recent Advances

Study Crawford and Shum (2005) for pharmaceutical demand uncertainty; Abhishek et al. (2015) for e-retail channel structures (1012 citations).

Core Methods

Core techniques: random coefficients logit, Bayesian updating for learning, instrumental variables for price endogeneity, simulation-based maximum likelihood.

How PapersFlow Helps You Research Discrete Choice Models of Consumer Brand Selection

Discover & Search

Research Agent uses searchPapers and citationGraph to map MMNL literature from McFadden and Train (2000), revealing 3997 citations and forward links to Erdem and Keane (1996). findSimilarPapers expands to dynamic extensions like Crawford and Shum (2005). exaSearch queries 'mixed logit scanner panel brand choice' for 250M+ OpenAlex papers.

Analyze & Verify

Analysis Agent runs readPaperContent on McFadden and Train (2000) to extract simulation algorithms, then verifyResponse with CoVe checks model convergence claims against GRADE evidence grading. runPythonAnalysis replicates MMNL estimation on sample scanner data using NumPy/pandas, verifying heterogeneity distributions statistically.

Synthesize & Write

Synthesis Agent detects gaps in dynamic heterogeneity coverage between Erdem and Keane (1996) and recent works, flagging contradictions in learning rates. Writing Agent uses latexEditText for model equations, latexSyncCitations for 50+ references, and latexCompile for publication-ready appendices; exportMermaid visualizes utility maximization hierarchies.

Use Cases

"Replicate MMNL estimation from McFadden Train 2000 on synthetic scanner data"

Research Agent → searchPapers → Analysis Agent → runPythonAnalysis (NumPy logit simulation) → researcher gets fitted random coefficients plot and elasticity table.

"Write LaTeX appendix comparing mixed logit to nested logit for brand choice"

Synthesis Agent → gap detection → Writing Agent → latexEditText + latexSyncCitations (McFadden 2000) + latexCompile → researcher gets compiled PDF with equations and citations.

"Find GitHub code for dynamic brand choice models like Erdem Keane"

Research Agent → paperExtractUrls (Erdem 1996) → Code Discovery → paperFindGithubRepo → githubRepoInspect → researcher gets runnable Bayesian learning scripts.

Automated Workflows

Deep Research workflow conducts systematic review: searchPapers (mixed logit brand) → citationGraph → readPaperContent on top 50 → structured report with GRADE scores. DeepScan applies 7-step analysis to Erdem and Keane (1996), verifying dynamic assumptions via CoVe and Python replication. Theorizer generates hypotheses on network effects integration from Rysman (2009) and Farrell Klemperer (2006).

Try Doxa for Discrete Choice Models of Consumer Brand Selection Research

Frequently Asked Questions

What defines discrete choice models in brand selection?

Models derived from random utility maximization using logit/probit with random coefficients for attribute-based choices. McFadden and Train (2000) show MMNL approximates any such model.

What are core estimation methods?

Maximum simulated likelihood with Halton draws for random coefficients. Erdem and Keane (1996) use Gibbs sampling for dynamic Bayesian extensions.

What are key foundational papers?

McFadden and Train (2000, 3997 citations) for MMNL theory; Erdem and Keane (1996, 1253 citations) for uncertainty and learning.

What open problems remain?

Scalable estimation for high-dimensional heterogeneity; integrating network effects (Farrell and Klemperer, 2006); endogeneity in online settings (Ellison and Ellison, 2009).