Keuzemodellering

Keuzemodellering is a set of statistical and econometric methods for analyzing decisions among discrete alternatives. The approach is grounded in random utility theory: each alternative provides a latent utility to the decision maker, and the observed choice is the one with the highest utility. Keuzemodellering is widely used to predict consumer and individual behavior and to quantify trade-offs among attributes such as price, quality, and travel time.

Common models include multinomial logit (MNL), nested logit and probit models. Data come in two main forms: revealed preference data from real choices, and stated preference data from surveys. Estimation is typically done by maximum likelihood; model fit is assessed with log-likelihood, AIC/BIC, and predictive validity on holdout samples.

Limitations: MNL assumes independence of irrelevant alternatives, which can bias results; alternative-specific effects and preference heterogeneity may not be captured. Solutions include mixed logit (random coefficients), latent class models, and error component models.

Applications: market research, transportation planning, health economics, environmental valuation, policy analysis. By estimating attribute trade-offs, researchers can derive willingness-to-pay and simulate policy changes.

History and tools: keuzemodellering has its roots in the work of McFadden and colleagues in the 1970s; later developments address panel data, scale heterogeneity, and non-linear attribute effects. Software commonly used includes R packages such as mlogit and apollo, and Python packages such as pylogit, enabling estimation, model comparison, and simulation of policy or market scenarios.