Add calculation of predictor effects

nestedLogit 0.3.2

This is a modest enhancement to the package, adding effect displays and smoothing some infelicities in the previously released version.

• new Effect method for "nestedLogit" objects to create effect displays.
• add Effect example to vignette

nestedLogit 0.3.1

• correct buglet in as.data.frame method when newdata has one column
• predict.nestedLogit() now includes the newdata data frame as an additional component (.data) in its result. Consequently, the newdata argument is no longer required in the as.data.frame method.

Add standard errors, CIs and vignettes

This is a major enhancement to the package, adding computations of standard errors and confidence intervals to predicted probabilities and logits.

Enhancements

• The predict() method now computes standard errors for probabilities and logits using the delta method. These can be obtained for either the response probabilities (and equivalent logits) or for the predicted log odds of the individual dichotomies.
• A confint() method for the result of predict() generates the corresponding confidence intervals.
• as.data.frame() methods for predicted values, either for the nested logit model or for the separate dichotomies converts these to a data frame in long format, handy for using ggplot().
• Added a confint() method for predicted probabilities and logits
• The plot() method for "nestedLogit" objects now plots confidence intervals for predicted probabilities.
• A new vignette, "standard-errors", describes the mathematics behind the standard error calculations.
• An old vignette on plotting methods was completely re-written using the new predict() methods and focusing exclusively on `ggplot2().

Other

• Added an example of lobstr::tree() to print nested lists
• Now use fig.show="hold" to keep par(op) with the code.

Minor fix for CRAN

Fixed two remaining \value problems noted by reviewer.

First CRAN release

This version has now advanced from experimental to stable with the recent enhancements, below. Now submitted to CRAN.

• Now allow dichotomies to be specified by a nested (recursive) of binary splits of the categories [suggestion of Achim Zeileis]
• The model object is now of class "nestedLogit" for uniformity.
• A basic plot() method now operational
• Added a linearHypothesis() method to give Wald tests for hypotheses about coefficients or their linear combinations.
• Expanded vignette to illustrate some other methods.
• Added a models() generic and method to extract separate models from the "nestedLogit" object
• Added a logLike() method, and through it, gets AIC() and BIC()
• Reorganized documentation to separate nested hypothesis methods.