Features and fixes

[BertvanderVeen]

Features

• Grouped dispersion parameters can now be specified with a formula, with data included as rows in y. Alternatively still takes a separate vector of indices as input
• Added grouped dispersion parameters to the starting value generation for the quadratic response model
• getLV has now three (renamed) site score types: residual (as for num.lv), marginal (fixed-effects with num.RR or num.lv.c), and conditional (including both effects for num.lv.c). Also added some documentation on that
• CMSEPf and sdrandom have a new type argument relating to the chosen site scores. By default returns residual for num.lv, marginal for num.RR and conditional for num.lv.c
• ordiplot has a new type argument, so that it can create plots for all three types of site scores separately with accompanying prediction regions. The type of site scores used is added to the title if that is not specified
• Added the option in ordiplot to determine the size of site labels and to change the line type of species arrows, if drawn
• Added some more defensive coding for grouped dispersion parameters

Bugfixes

• Fixed the bug in residuals.gllvm where the lower threshold for the uniform distribution was occasionally higher than the higher threshold if a model fitted poorly. This resulted in inability to calculate residuals
• Fixed a bug for the Laplace approximation, in the C++ file, where num.RR and num.lv.c in the same model would crash R
• Removed loading messages of the R-package in vignette 5
• Fixed a bug in summary.gllvm where rstruc was not assigned to the object
• Fixed a bug in getResidualCov where num.lv.c should be object$num.lv.c
• Fixed a typo in the documentation of ordiplot
• Minor improvements for ordiplot with constrained model
• Minor textual changes in vignette
• Improved documentation of the gllvm object see issue #58
• Added a more informative error if starting values could not be calculated with ordinal model and predictors
• Fixed a bug in calculating residuals ordinal model (probK[1:y[i,j]] for y[i,j] = 0 wrongly resulted in probK[1]=probK[0])
• Fixed another bug in the calculation of residuals for the ordinal model in gllvm.aux where a cumsum.b was used where it should have been cumsum.a
• Fixed a bug due to which grouped dispersion parameters were not counted properly for the degrees of freedom

Other

• Added a warning relating to the p-values in summary if a (non-balanced, i.e. structured) random-effect is included.
• Added preprint for constrained ordination to the vignette

[BertvanderVeen]

In my last commit I've changed terminology in the getLV function in preparation of a different update where I add a type to the ordiplot function, corresponding to the same types as in getLV. Then, for constrained ordination, users will be able to specify which site scores to plot (with the correct prediction intervals). getLV still defaults to unconstrained LVs for num.lv, conditional LVs for num.lv.c, and marginal for num.RR.