3 Model specification

Before estimating a mixed logit model, you need to specify the model structure and the variables that affect the choice probabilities. The model structure consists of the utility function, which defines how the observed and unobserved attributes of the alternatives affect the utility of the decision maker, and the error term, which captures the unobserved factors that influence the choice. The variables that affect the choice probabilities include the fixed and random coefficients, which represent the marginal utility and the heterogeneity of the preferences, respectively. You can use theory, data analysis, or prior studies to guide your model specification.

4 Model identification

Model identification is the ability to uniquely estimate the parameters of the mixed logit model from the data. Identification depends on the richness and variation of the data, as well as the functional form and distributional assumptions of the model. To ensure identification, you need to avoid overparameterization, multicollinearity, and degenerate distributions. You can use diagnostic tests such as t-tests, Wald tests, or likelihood ratio tests to check the significance and stability of the parameters. You can also use information criteria such as AIC or BIC to compare alternative models and select the best one.

5 Model interpretation

Model interpretation is the ability to understand and explain the implications of the mixed logit model for the underlying preferences and behavior of the decision makers. Interpretation depends on the scale and sign of the parameters, as well as the marginal effects and elasticities of the variables. To interpret the parameters, you need to consider the units and levels of the variables, as well as the distribution and correlation of the random coefficients. To interpret the marginal effects and elasticities, you need to calculate the changes in the choice probabilities and the market shares due to changes in the variables.

6 Model validation

Model validation is the ability to assess the quality and performance of the mixed logit model in terms of its fit, prediction, and robustness. Fit measures how well the model reproduces the observed choices in the data, while prediction measures how well the model forecasts the choices in new or different scenarios. Robustness measures how sensitive the model is to changes in the data, the specification, or the estimation method. To validate the model, you need to use statistical tests such as goodness-of-fit tests, prediction tests, or bootstrap tests, as well as graphical tools such as histograms, scatter plots, or confidence intervals.

7 Here’s what else to consider

This is a space to share examples, stories, or insights that don’t fit into any of the previous sections. What else would you like to add?