choicer: fast discrete-choice models with a focus on economic applications

choicer provides fast estimation of discrete-choice models for applied economics. Likelihoods, analytical gradients and Hessians are implemented in C++ with OpenMP parallelism, scaling efficiently to specifications with many alternative-specific constants. Post-estimation routines return predicted shares, own- and cross-price elasticities, diversion ratios, and the BLP contraction. Supports multinomial logit (MNL), mixed logit (MXL), and nested logit (NL); more models will be added.

Installation

You can install the development version of choicer with:

pak::pkg_install("fpcordeiro/choicer")

Example

Estimate a multinomial logit model and compute elasticities and diversion ratios:

library(choicer)
library(data.table)

# Estimate
fit <- run_mnlogit(
    data = dt,
    id_col = "id",
    alt_col = "alt",
    choice_col = "choice",
    covariate_col = c("x1", "x2")
)

summary(fit)

# Post-estimation
predict(fit, type = "shares")        # predicted market shares
elasticities(fit, elast_var = "x1")  # own- and cross-price elasticities
diversion_ratios(fit)                # diversion ratio matrix

Supported models

Model	Function	Post-estimation
Multinomial Logit	run_mnlogit()	predict(), elasticities(), diversion_ratios(), blp()
Mixed Logit	run_mxlogit()	predict(), elasticities(), diversion_ratios(), blp()
Nested Logit	run_nestlogit()	—

All fitted models support summary(), coef(), vcov(), logLik(), AIC(), BIC(), and nobs().

Alternative packages

There are multiple R packages that offer similar functionalities:

• mlogit
• logitr
• gmnl
• apollo
• mixl