similar to: Coarsening Factors

Bother! This cold has made me accident-prone. I meant to hit Reply-all. Clarification below. -------- Original Message -------- Subject: Re: [R] Parameterization puzzle Date: Fri, 21 Jul 2006 19:10:03 +1200 From: Murray Jorgensen <maj at waikato.ac.nz> To: Prof Brian Ripley <ripley at stats.ox.ac.uk> References: <44C063E5.3020703 at waikato.ac.nz>

I can ask my question using and example from Chapter 1 of Pinheiro & Bates. > # 1.4 An Analysis of Covariance Model > > OrthoFem <- Orthodont[ Orthodont$Sex == "Female", ] > fm1OrthF <- + lme( distance ~ age, data = OrthoFem, random = ~ 1 | Subject ) > summary( fm1OrthF ) Linear mixed-effects model fit by REML Data: OrthoFem AIC BIC

Having just update to R 2.6.2 on my old Windows laptop I notice that the number of packages is growing exponentially and my usual approach of get-em-all may not be viable much longer. Has any thought been given to dividing "contributed" binaries into a recommended set, perhaps a couple of hundred, and the remained. That way one could install the recommended ones routinely and add in

A small example before I begin my query: > educ <- read.table(efile, header=TRUE) > educ Education Age_Group Count 1 IncompleteHS 25-34 5416 2 IncompleteHS 35-44 5030 3 IncompleteHS 45-54 5777 4 IncompleteHS 55-64 7606 5 IncompleteHS >64 13746 6 CompletedHS 25-34 16431 7 CompletedHS 35-44 1855 8 CompletedHS 45-54 9435 9

I am exploring the result of clustering a large multivariate data set into a number of groups, represented, say, by a factor G. I wrote a function to see how categorical variables vary between groups: > ddisp <- function(dvar) { + csqt <- chisq.test(G,dvar) + print(csqt$statistic) + print(csqt$observed) + print(round(csqt$expected)) + round(csqt$residuals) + } > > x

I forgot to mention that I am using Windows XP. -------- Original Message -------- Subject: Pasting data into scan() Date: Tue, 02 May 2006 11:55:03 +1200 From: Murray Jorgensen <maj at stats.waikato.ac.nz> To: r-help at stat.math.ethz.ch The file TENSILE.DAT from the Hand et al "Handbook of Small Data Sets" looks like this: [...] -- Dr Murray Jorgensen

Hi all, I have just found out that the machine learning group in our Faculty has a lot of spare capacity on their FreeNX servers. I do not know a lot about these beasts but I understand that they are a free version of something produced by a firm called "NoMachine". They are designed for executing parallel algorithms and I thought that they might be of use in a project of mine

The birth weight example from ?stepAIC in package MASS runs well as indeed it should. However when I change stepAIC() calls to step() calls I get warning messages that I don't understand, although the output is similar. Warning messages: 1: In model.response(m, "numeric") : using type="numeric" with a factor response will be ignored (and three more the same.) Checked

Consider the following: > A <- 1:10 > A [1] 1 2 3 4 5 6 7 8 9 10 > dim(A) NULL > dim(A) <- c(2,5) > A [,1] [,2] [,3] [,4] [,5] [1,] 1 3 5 7 9 [2,] 2 4 6 8 10 > dim(A) [1] 2 5 > dim(A) <- 10 > A [1] 1 2 3 4 5 6 7 8 9 10 > dim(A) [1] 10 Would it not make sense to have dim(A) = length(A) for all vectors?

I have just installed R 2.11.1 on my XP laptop. I like html help for browsing but text help for on-the-fly look-ups. I was a bit surprised when I was asked to choose between them during the installation. I chose text, thinking I could fix the html help later, which is what I am trying to do now. Now when I ask for html help my browser goes to 'http://-ip-number-/doc/html/index.html'

I found myself wanting to average a vector [vec] within each level of a factor [Fac], returning a vector of the same length as vec. After a while I realised that lm1 <- lm(vec ~ Fac) fitted(lm1) did what I want. But there must be another way to do this, and it would be good to be able to apply other functions than mean() in this way. Cheers, Murray -- Dr Murray Jorgensen

I assigned a class the first problem in Pinheiro & Bates, which uses the data set PBIB from the SASmixed package. I have recently downloaded 2.0.1 and its associated packages. On trying library(SASmixed) data(PBIB) library(nlme) plot(PBIB) I get a warning message Warning message: replacing previous import: coef in: namespaceImportFrom(self, asNamespace(ns)) after library(nlme) and a

I want to extra the part of the formula not including the response variable from an lm object. For example if the lm object ABx.lm was created by the call ABx.lm <- lm( y ~ A + B + x, ...) Then ACx.lm is saved as part of a workspace. I wish to extract "~ A + B + x". Later in my code I will fit another linear model of the form z ~ A + B + x for some other response variable z. I

I'm going to try my hand at converting some Fortran programs to R. Does anyone know of any good articles giving hints at such tasks? I will post a selective summary of my gleanings. Cheers, Murray -- Dr Murray Jorgensen http://www.stats.waikato.ac.nz/Staff/maj.html Department of Statistics, University of Waikato, Hamilton, New Zealand Email: maj at waikato.ac.nz

I am introducing the scan() function to my class. Consider the following file (Scanexamp.Rnw ) \documentclass[12pt]{article} \begin{document} <<>>= height = scan() 64 62 66 65 62 69 72 72 70 part = scan(what = character(0)) "Soprano" "Soprano" "Soprano" "Alto" "Alto" "Tenor" "Tenor" "Bass"

I am wondering if stepAIC in the MASS library may be used for model selection in an overdispersed Poisson situation. What I thought of doing was to get an estimate of the overdispersion parameter phi from fitting a model with all or most of the available predictors (we have a large number of observations so this should not be problematical) and then use stepAIC with scale = phi. Should this

We have a choice when calculating the Huber location estimate: > set.seed(221205) > y <- 7 + 3*rt(30,1) > library(MASS) > huber(y)$mu [1] 5.9117 > coefficients(rlm(y~1)) (Intercept) 5.9204 I was surprised to get two different results. The function huber() works directly with the definition whereas rlm() uses iteratively reweighted least squares. My surprise is

Consider the following output [R2.2.0; Windows XP] > set.seed(160706) > X <- matrix(rnorm(40),nrow=10,ncol=4) > Xpc <- princomp(X,cor=FALSE) > summary(Xpc,loadings=TRUE, cutoff=0) Importance of components: Comp.1 Comp.2 Comp.3 Comp.4 Standard deviation 1.2268300 0.9690865 0.7918504 0.55295970 Proportion of Variance 0.4456907 0.2780929

I am in a discriminant analysis situation with a frame containing several variables and a grouping factor, if you like: set.seed(200906) exampledf <- as.data.frame(matrix(rnorm(50,5,2),nrow=10,ncol=5)) exampledf$Group <- factor(rep(c(1,2,3),c(3,3,4))) exampledf I'm sure there must be a simple way to get the within group pooled covariance matrix but I haven't found it yet. I

[Using R 2.2.0 on Windows XP; OK, OK, I will update soon!] I have noticed some undesirable behaviour when applying ifelse to a data frame. Here is my code: A <- scan() 1.000000 0.000000 0.000000 0 0.00000 0.027702 0.972045 0.000253 0 0.00000 A <- matrix(A,nrow=2,ncol=5,byrow=T) A == 0 ifelse(A==0,0,-A*log(A)) A <- as.data.frame(A) ifelse(A==0,0,-A*log(A)) and this is the output: