Multinomial Logit Models for Categorical Responses and Discrete Choices (original) (raw)

This packages provides estimators for multinomial logit models in their conditional logit (for discrete choices) and baseline logit variants (for categorical responses), optionally with overdispersion or random effects. Random effects models are estimated using the PQL technique (based on a Laplace approximation) or the MQL technique (based on a Solomon-Cox approximation). Estimates should be treated with caution if the group sizes are small.

The statistical models

• Conditional logit models
• Baseline-category logit models
• The relation between baseline logit and conditional logit models
• Random effects in baseline logit models and conditional logit models

Technical aspects of model fitting

• The IWLS algorithm used to fit conditional logit models
• Approximate Inference for Multinomial Logit Models with Random Effects
• Bias reduction using Firth’s penalized likelihood technique