similar to: Hauck-Donner

There is both a technical and a theoretical element to my question... Should I be able to use the outputs which arise from the fastbw function as inputs to nomogram(). I seem to be failing at this, -- I obtain a subscript out of range error. That I can't do this may speak to technical failings, but I suspect it is because Prof Harrell thinks/knows it injudicious. However, I can't

I want to employ a parsimonious model to draw nomograms, as the full model is too complex to draw nomograms readily (several interactions of continuous variables). However, one interesting variable stays or leaves based on whether I choose "p value" or "AIC" options to fastbw(). My question boils down to this: Is there a theoretical reason to prefer one over another?

Dear All, For logistic regression models: is it possible to use validate (rms package) to compute bias-corrected AUC, but have variable selection with AIC use step (or stepAIC, from MASS), instead of fastbw? More details: I've been using the validate function (in the rms package, by Frank Harrell) to obtain, among other things, bootstrap bias-corrected estimates of the AUC, when variable

Dear R users, For the purpose of validating a prediction model using validate() from the rms package, I am running into some trouble with using the fastbw() function breaking up natural groups of variables. Is there any way I can specify to keep certain variable together? In particular, if interactions are included I would also like to keep the main effects in the model. Another example is a

Hello I am running R 2.0.1 on Windows, I am attempting to use Frank Harrell's 'fastbw' function (from the Design library), but I get an error that the fit was not created with a Design library fitting function; yet when I go to the help for fastbw (and also look in Frank's book Regression Modeling Strategies) it appears that fastbw should work with a model created with lm.....

Dear prof. Harrell, I'm not able to use the force option with fastbw, here an example of the error I've got (dataset stagec rpart package): > fitstc <- cph(Surv(stagec$pgtime,stagec$pgstat) ~ age + eet + g2 + grade + gleason + ploidy, data=stagec) > fbwstc <- fastbw(fitstc,rule="aic",type="individual") > fbwstc Deleted Chi-Sq d.f. P Residual d.f.

Hello I'm using logistic regression from the Design library (lrm), then fastbw to undertake a backward selection and create a reduced model, before trying to make predictions against an independent set of data using predict.lrm with the reduced model. I wouldn't normally use this method, but I'm contrasting the results with an AIC/MMI approach. The script contains: # Determine full

Hi, it seems that the fastbw() in the Design package only works with variable of class "factor" according to the help page if I understand correctly. Is there any R function/package that do stepwise variable selection for a Cox model with continuous independent variables? Thank you John ____________________________________________________________________________________ Looking

Hello! I am using relatively simple linear model. By applying fastbw() on ols() results from rms package I would like to get subtable "Approximate Estimates after Deleting Factors". However, it seems this is not possible. Am I right? I can only get coefficients for variables kept in the model (for example: x$coefficients), but not S.E., Wald's Z and P? Is there any easy way to

I am using the validate option in the Design package with the Cox survival model. I am using the bw=T option which, like the fastbw function, performs a backward elimination variable selection The output includes a series of columns (below) giving information on eliminated variables. My question is that I am unsure of the difference between the 2 p-values given (the one after Chi-Sq and the one

A new version of rms is now available on CRAN for Linux and Windows (Mac will probably be available very soon). Largest changes include latex methods for validate.* and adding the capability to force a subset of variables to be included in all backwards stepdown models (single model or validation by resampling). Recent updates: * In survplot.rms, fixed bug (curves were undefined if

A new version of rms is now available on CRAN for Linux and Windows (Mac will probably be available very soon). Largest changes include latex methods for validate.* and adding the capability to force a subset of variables to be included in all backwards stepdown models (single model or validation by resampling). Recent updates: * In survplot.rms, fixed bug (curves were undefined if

Dear all, I think I have a rather strange question, but I'd like to give it a try: I want to perform a simulation numerous times, thats why I can't do it by hand. I sample a small dataset from a very large one, and use backward selection to select significant predictors for some arbitrary outcome variable Y. These predictors are to be placed in a model, and regression coefficients

I just ran example(glm) and happened to notice that models based on the Gamma distribution gives a t test, while the Poisson models give a z test. Why? Both are b/s.e., aren't they? I can't find documentation supporting the claim that the distribution is more like t in one case than another, except in the Gaussian case (where it really is t). Aren't all of the others approximations

Dear all, Is there any function for calculating confidence limits for coefficients in an lm() object? I know of the confint() function in the MASS library working very well on my binomial GLMs and I have tried it (using glm () , family=gaussian) but it gives NAs according to below. Does the confint() function not accept gaussian GLMs? Could there be convergence problems in the GLM? Note the

I have a very simple problem. When using glm to fit binary logistic regression model, sometimes I receive the following warning: Warning messages: 1: fitted probabilities numerically 0 or 1 occurred in: glm.fit(x = X, y = Y, weights = weights, start = start, etastart = etastart, 2: fitted probabilities numerically 0 or 1 occurred in: glm.fit(x = X, y = Y, weights = weights, start = start,

As suggested in my earlier message, I have a large population of independent variables and a binary dependent outcome. It is expected that only a few of the independent variables actually contribute to the outcome, and I'd like to find those. If it wasn't already obvious, I am *not* a statistician. Not even close. :-) Statistician colleagues have suggested that I use logistic

Hi Mark, I don't think anyone will dispute that it's easier to hack up a shadow stack (or plug into a conservative collector) to get up and running with GC. That is absolutely the route to go if portability trumps performance. If you review the mailing list history, I think you'll also find that developers who do care about performance have been disappointed with the impact

I am cleaning up the rms package to not export functions not to be called directly by users. rms uses generic functions defined in other packages. For example there is a latex method in the Hmisc package, and rms has a latex method for objects of class "anova.rms" so there are anova.rms and latex.anova.rms functions in rms. I use:

Gordon Henriksen wrote: > Hi Mark, > > I don't think anyone will dispute that it's easier to hack up a shadow > stack (or plug into a conservative collector) to get up and running > with GC. That is absolutely the route to go if portability trumps > performance. Why? LLVM is all about portability AND performance. > > If you review the mailing list history,