tobit1 is an R package which fits and tests models for which the
response is truncated.
A large amount of estimating methods are provided:
- maximum likelihood estimator,
- two-steps estimator,
- symetrically trimmed estimator,
- non-linear least square estimator.
The estimation can be performed either on the censored or the truncated sample.
Full fonctionalities are provided for the most popular case where the response is zero-truncated response. Some estimators are also provided for responses where the left truncation point is not 0, where the response is right truncated and also for two-limits response.
Specification tests are of particular interest for tobit models as the
consistency of most estimators relies on the hypothesis of
homoscedasticity and normality. tobit1 provide an haustest for the
Hausman (1978) test and, for conditional moment tests, the cmtest
can be used as it provides specific methods for tobit1 objects.
The endogeneity test of Smith and Blundell (1986) is provided as the
smbltest function. The estimation of simultaneous-equation Tobit
models can be performed using either a two-steps estimator or maximum
likelihood using the ivtobit function.
tobit1 is not currently available on CRAN. Use:
remotes::install_github("ycroissant/tobit1", build_vignettes = TRUE)