Here is a list of known bugs. 2020-10 rrvglm() does not do half stepping so one iteration can be worse than the previous. It seems the latest estimate of A and/or C does not translate into an improvement in the next IRLS iteration. 2016-06 lrtest(zipoissonff.object, poissonff.object) fails. 2016-05 rcim() with alaplace2() may fail. 2014-02 The subset argument of vgam() may not work, especially with multiple responses. To get around this, use subset() to create a smaller data frame and then feed that into vgam(). 2013-11 vgam() can only handle constraint matrices cmat, say, such that t(cmat) %*% cmat is diagonal. 2013-07 quasipoisson()'s scale parameter estimate does not handle prior weights correctly. 2012-09 loge('a', short = FALSE, inverse = FALSE) loge('a', short = FALSE, inverse = TRUE) give the same answer. Coef(vglm.dirmultinomial.fit) fails. Evidently, multiple "mlogit"s saved on vglm.dirmultinomial.fit@misc do not suffice. 2011-12 VGAM version 0.8-4 said it needed R version 2-11.1 or later. But really, R version 2-13.0 or later is needed. This is because the generic nobs() was not defined properly. Another fix is to install the (latest) prerelease version at http://www.stat.auckland.ac.nz/~yee/VGAM/prerelease 2010-04-12 cqo() should be working now. It uses new C code. Also, vgam() and vsmooth.spline() should not be noticeably different from before. But cao() is still working... getting it going soon hopefully. 2009/07/13 cqo() fails... I think it is due to initial values being faulty. Hope to look into it soon. 2009/06/18 For a given VGAM family function, arguments such as parallel, exchangeable etc. will not work if the RHS of the formula is an intercept only. For example, parallel = FALSE ~ 1 and exchangeable = TRUE ~ 1 will fail. Instead, try something like parallel = TRUE ~ x2 + x3 + x4 -1 and exchangeable = FAlSE ~ x2 + x3 + x4 + x5 -1 respectively. 2009/01/01 prediction with vgam( ~ offset(myoffsetmatrix) + ... ) fails inside a function because myoffsetmatrix cannot be found. 2008/08/12 Under Windows, the vgam() example involving the Hunua data seems to fail. It is under investigation. 2008/08/04 VGAM interferes with other packages, e.g., predict() and summary(). This is due to S3 and S4 interference, and currently I haven't sussed out the full details (e.g., NAMESPACES). For now it is best to attach VGAM only when needed and detach it when other packages are to be used. This can be done with library(VGAM) and detach("package:VGAM") 2008/05/16 zipf() did not handle 0 < s < 1. The prerelease version fixes this. 2008/03/12 A call such as mydof = 4 Fit = vgam(y ~ s(x, df=mydof), fam=poissonff) will result in failure when plot(Fit) Instead, one needs Fit = vgam(y ~ s(x, df=4), fam=poissonff) 2008/02/16 The VGAM package interferes with other functions, for example, if VGAM is loaded and lmobject is an "lm" object then fitted(lmobject) predict(lmobject) resid(lmobject) residuals(lmobject) will fail. 2006/05/18 dirmul() is not working yet. 2005/11/16 cao() now works in Windows. The argument xij does not work properly. 2005/8/31 The windows version of cao() seems to hang. It does not hang in Linux. 2005/6/10 cao() works in Linux but seems to hang in Windows. The latter (distributed in a .zip file format) is made from a R Cross Build process which may be a reason for the bug. I'm slowly looking into the bug. 2005/5/6 The VGAM package interferes with other code, including glm(). This may be due to the smart prediction code, or be due to the NAMESPACE facility. In order to use other functions outside the VGAM package you may need to type "detach()". 2003/7/14 vgam(y ~ s(x, df=2), subset= x > 2) will fail in R because the subset argument has the effect that the "df" and "spar" attributes are removed from the data frame containing the smoothing variables. Current fix: create a separate data frame satisfying the subset= condition, and then run vgam() on this smaller data frame. Thanks for Eugene Zwane for finding this bug.