Censored Regression with Conditional Heteroscedasticity

Different approaches to censored or truncated regression with conditional heteroscedasticity are provided. First, continuous distributions can be used for the (right and/or left censored or truncated) response with separate linear predictors for the mean and variance. Second, cumulative link models for ordinal data (obtained by interval-censoring continuous data) can be employed for heteroscedastic extended logistic regression (HXLR). In the latter type of models, the intercepts depend on the thresholds that define the intervals.


Changes in Version 1.0-1

o Added argument 'type' to crch() which can be set to "crps" for parameter estimation with minimum CRPS instead of maximum likelihood.

o Added S3 method of crps() from scoringRules for "crch" objects.

o Improvements for the predict() method: - new types "parameter", "density", "probability", and "crps" - with type="response" now the expected value and not the location parameter is returned (not equal for censored and truncated distributions). For better backward compatibility, the default type is set to "location".

o Changed argument names mean and sd to location and scale in logistic and student-t distribution functions

o Added new function "crch.stabsel" for stability selection based on
"crch.boost.fit". Small S3 methods for the returned class "stabsel.crch" are also provided.

Changes in Version 1.0-0

o New release accompanying the R-Journal paper: "Heteroscedastic Censored and Truncated Regression with crch" by Messner, Mayr, and Zeileis which appears as R-Journal 8(1). See also citation("crch").

o Added estfun() method for crch objects

Changes in Version 0.9-2

o The crch() function now supports coefficient optimization by boosting to automatically select the most relevant input variables in high dimensional data settings. Extractor and plotting functions for corresponding crch.boost objects are also available.

o Transferred functions to estimate density, distribution, score, and Hessian matrices to C-code to accelerate coefficient optimization.

o Added option to crch() to avoid computation of covariance matrix

o Added left and right arguments to predict.crch() and predict.crch.boost() to allow quantile predictions for non-constant censoring or truncation points.

Changes in Version 0.9-1

o Added model.matrix() and model.frame() methods for crch objects

o Bug Fix in predict.crch(): In previous versions predictions for models with other link functions than the log gave wrong results

Changes in Version 0.9-0

o Added vignette to introduce the crch() function with some theoretical background and an illustrating example: vignette("crch", package = "crch")

o The crch() function now also supports truncated responses. Furthermore added a wrapper function trch() to fit truncated regression models.

o crch(): Analytical gradients and Hessian matrices are provided for most models to speed up maximum likelihood optimization (not available for student-t distribution with degrees of freedom estimation).

o crch(): For the scale model a link function can now be specified (log, identity, or quadratic). In previous version only the log was supported.

o Added functions for probability density, cumulative distribution, random numbers, and quantiles for censored and truncated normal, logistic, and student-t distributions.

o The residuals() method for crch objects now also provides quantile residuals (Dunn and Smyth 1996).

o Added update() method for crch objects.

Changes in Version 0.1-0

o First official release of the package on CRAN. See citation("crch") for the accompanying manuscripts. Note that the interface of both crch() and hxlr() is still under development and might change in future versions of the package.

Reference manual

It appears you don't have a PDF plugin for this browser. You can click here to download the reference manual.


1.0-4 by Jakob Messner, a year ago

Browse source code at https://github.com/cran/crch

Authors: Jakob Messner [aut, cre] , Achim Zeileis [aut] , Reto Stauffer [aut]

Documentation:   PDF Manual  

Task views: Econometrics

GPL-2 | GPL-3 license

Imports stats, Formula, ordinal, sandwich, scoringRules

Suggests glmx, lmtest, memisc

Imported by MortalityGaps.

Suggested by NetSimR, ensemblepp, insight, scoringRules.

Enhanced by prediction.

See at CRAN