Choice Models & Dynamic Choice

Foundational paper deriving multinomial logit from random utility maximization; established modern discrete choice.

Random coefficients demand with market-level data; workhorse for IO and empirical demand.

Proves mixed logit approximates any random utility model; establishes simulation-based estimation.

The definitive textbook covering logit variants and simulation methods; essential practitioner reference.

Mixed logit applied to digital products; estimates substitution between print and online news with cannibalization implications.

Discrete choice demand model incorporating ideological content; demonstrates how consumer preferences drive product differentiation in information markets.

Foundational model of consideration set formation as cost-benefit tradeoff.

Embeds limited information into BLP-style demand; shows advertising affects consideration sets.

Axiomatic foundation for choice with limited attention; introduces 'attention filters'.

Experimental evidence on directed cognition and bounded rationality in search.

Uses Expedia field experiment to causally identify position effects; shows rankings affect search behavior but not purchase conditional on search.

Develops integrated model of simultaneous search and switching costs using consideration set data; methodological template for separating search from switching frictions.

Models search frictions for financially homogeneous products; explains fund proliferation and fee dispersion via information costs.

Introduces NFXP algorithm for dynamic discrete choice; establishes the structural framework.

CCP estimation recovering value functions without solving the full DP problem.

Comprehensive survey of dynamic discrete choice methods and computational issues.

Extends CCP to unobserved heterogeneity; introduces 'finite dependence' for tractability.

DDC with forward-looking consumers, heterogeneity, and repeat purchases; estimated on digital camcorder data—template for tech durable goods.

Dynamic pricing with strategic consumers anticipating price declines; Markov-perfect equilibrium applied to video game consoles.

Structural model showing static demand estimates overstate own-price elasticities by 30% due to stockpiling.

Models optimal pricing when consumers stockpile; sales capture 25-30% of price discrimination profits.

Dynamic structural model with Bayesian learning and forward-looking stockpiling behavior.

Models streaming video consumption dynamics—session continuation, series switching, and inter-session timing; first rigorous treatment of binge-watching economics using Hulu data.

Foundational book introducing 'libertarian paternalism' and choice architecture.

Auto-enrollment increases 401(k) participation from 49% to 86%; canonical default effects evidence.

Commitment device increasing savings from 3.5% to 13.6% in 40 months.

Theoretical framework for optimal default design under heterogeneous preferences.

Organ donation rates dramatically higher in opt-out countries; demonstrates default power beyond savings.

Large-scale RCT deactivating Facebook; finds deactivation increases well-being and reduces news consumption—questions platform welfare measurement.

Economic model of digital addiction with habit formation and self-control; estimates 31% of social media use attributable to self-control problems.

Studies EU privacy directive's effects; shows opt-in defaults reduced ad effectiveness by 65%—demonstrates default effects in digital advertising markets.

The 'Pandora's Box' paper introducing reservation value rule for optimal sequential search—theoretical foundation for all empirical search models.

Develops methodology to estimate search cost distributions from equilibrium price data alone; identification from search model restrictions.

Uses e-commerce clickstream data to test sequential vs. fixed-sample search; estimates search costs in online book market.

Tractable probit search model combining Weitzman's framework with demand estimation; applied to Amazon camcorder data.

Seminal paper introducing Bayesian learning for consumer choice under quality uncertainty; consumers update beliefs from experience and advertising signals.

Clever identification strategy separating informative from prestige advertising effects using experience goods; informative ads affect inexperienced consumers differently.

Integrates multiple quality signals (price, advertising frequency, content, experience) into unified Bayesian framework; shows promotions can erode brand equity.

Comprehensive survey covering identification, estimation, and extensions including forgetting and social learning—essential methodological reference.

Shows collaborative filtering exhibits popularity bias through rich-get-richer effects; recommendations can reduce, not increase, sales diversity.

Contrary to fragmentation fears, finds personalization increases commonality across consumers through volume and mix effects.

Experiments show recommendations serve as anchors influencing users' own preference ratings; creates feedback loops affecting future recommendations.

Using 4+ years of browsing data from 200,000+ U.S. adults, finds differentiated platform effects: Facebook increases diversity but partisan shift; Twitter shows minimal effects.

Field experiments show consumption diversity correlates with retention, yet algorithmic recommendations push toward less diverse listening—identifies core platform design tension.

Seminal survey establishing theoretical framework for how switching costs create market share dependence and affect competitive dynamics.

Demonstrates identification of structural state dependence from spurious heterogeneity using flexible semi-parametric methods; distinguishes true loyalty from heterogeneity.

Quantifies massive switching costs ($2,000+) using natural experiment; shows reducing inertia can backfire by exacerbating adverse selection.

Foundational model distinguishing single-homing from multi-homing; introduces 'competitive bottlenecks' explaining platform lock-in dynamics.

Classic model where rational agents ignore private information and follow predecessors, leading to potentially inefficient herding cascades.

Introduces informational cascades where optimal behavior is to follow predecessors; explains fragility of mass behaviors and sudden social shifts.

Foundational econometrics paper showing why peer effects cannot be identified from behavioral data alone—the 'reflection problem' from simultaneity.

Landmark RCT with 1.4 million Facebook users identifying causal peer influence; passive-broadcast features generate 246% increase in peer adoption.