Multinomial Probit and Logit models

Multinomial, Conditional, and Mixed Models

• Multinomial outcome dependent variable with categorical, unordered alternatives.
• Data recorded in wide or long format with codes for alternatives.
• Examples of multinomial outcomes include insurance contracts, product selection, occupational choice, and fishing mode.
• Two types of independent variables: alternative-invariant and alternative-variant.
• Multinomial logit model used with alternative-invariant regressors, with coefficients and marginal effects.
• Marginal effects do not necessarily correspond in sign to coefficients.
• Conditional logit model used with both alternative-invariant and alternative-variant regressors, with coefficients and marginal effects.
• Mixed logit model allows for random parameters and correlation across alternatives.
• Coefficient interpretation for regressors in both conditional and mixed logit models.
• Multinomial probit model similar to multinomial logit model, but uses standard normal cdf.
• Marginal effects interpretation: each unit increase in independent variable increases/decreases probability of selecting alternative by the marginal effect expressed as a percent.
• Mixed logit model produces random parameters coefficients for both regressor and standard deviation of regressor, indicating heterogeneity in effect of independent variable on alternative chosen.

Multinomial, Conditional, and Mixed Models

• Multinomial outcome dependent variable with categorical, unordered alternatives.
• Data recorded in wide or long format with codes for alternatives.
• Examples of multinomial outcomes include insurance contracts, product selection, occupational choice, and fishing mode.
• Two types of independent variables: alternative-invariant and alternative-variant.
• Multinomial logit model used with alternative-invariant regressors, with coefficients and marginal effects.
• Marginal effects do not necessarily correspond in sign to coefficients.
• Conditional logit model used with both alternative-invariant and alternative-variant regressors, with coefficients and marginal effects.
• Mixed logit model allows for random parameters and correlation across alternatives.
• Coefficient interpretation for regressors in both conditional and mixed logit models.
• Multinomial probit model similar to multinomial logit model, but uses standard normal cdf.
• Marginal effects interpretation: each unit increase in independent variable increases/decreases probability of selecting alternative by the marginal effect expressed as a percent.
• Mixed logit model produces random parameters coefficients for both regressor and standard deviation of regressor, indicating heterogeneity in effect of independent variable on alternative chosen.