similar to: Specifying weird models

None of Andy's comments) are inconsistent with the point that rlm() and lqs(), if they disagree with lm(), likely offer better places to start than does lm(), in identifying points that should be examined as in some sense outliers. All such methods are to be used, not as crutches, but as sources of insight. Incidentally, the "rlm" class inherits from "lm", and plot.lm()

To estimate the covariance matrix of e you could use the sample covariance matrix of the residuals. If desired, use its cholesky decomposition to transform to make the error approximately uncorrelated, then refit (and back-transform the coefficient matrix). Stacking the columns of Y and replicating X won't do what you write; it forces each univariate regression to have the same coefficients.

How about: nis.na <- complete.cases --- From: <Ted.Harding at nessie.mcc.ac.uk> Hi Folks, My recent response to Laura Quinn's query about matrix subsetting reminded of a question. I wrote: iDir <- ((Winds[,20]<45)|(Winds[,20]>315))&(!is.na(Winds[,20])) Now, I find "!is.na" a bit awkward to type, so I might prefer to type it as "nis.na".

Hi Folks, Given k RVs with MVN distribution N(mu,S) (S a kxk covariance matrix), let (w.l.o.g.) X1 denote the first r of them, and X2 the last (k-r). Likewise, let mu1 and mu2 denote their respective expectations. Then, of course, the expectation of X2 given X1=x1 is mu2 + S21*inv(S22)*(x1 - mu1) and the covariance matrix of X2 given X1=x2 is S22 - S21*inv(X11)*S12 where Sij is the

Hi folks Does R have anything straightforwardly resembling the commands fprintf, sprintf, printf (derived from C, and present in octave and matlab)? As in printf(format_string, ... ) where "format_string" defines the print format (including any fixed text) and "..." is a list of variables whose values are to be inserted into the line. Example: printf("Case

> -----Original Message----- > From: r-help-bounces at stat.math.ethz.ch > [mailto:r-help-bounces at stat.math.ethz.ch] On Behalf Of > Ted.Harding at nessie.mcc.ac.uk > Sent: Sunday, December 05, 2004 7:14 PM > To: r-help at stat.math.ethz.ch > Subject: Re: [R] What is the most useful way to detect > nonlinearity in lo > > > On 05-Dec-04 Peter Dalgaard wrote:

Hi Folks, No doubt a question with a well-known answer, but I'm unfortunately not managing to find it readily ... ! I have a quantitative variable Y and a 4-level ordered factor A (with very unequal numbers at the different levels, by the way). The command lm(Y ~ A) returns (amongst other stuff) an intercept, and coefficients A.L, A.Q and A.C for the Linear, Quadratic and Cubic effects.

Greetings Folks, When R code (as entered or read from a courced file) is executed, is it interpreted from the input form every time having once been read in, or do subsequent invocations use an "intermediate" (pre-interpreted) form? Or, putting it another way, is the execution of R code faster second time time round (and later) because the pre-interpretation has already been done once

Hi Folks, I am interested in extending the repertoire of link functions in glm(Y~X, family=binomial(link=...)) to include a "tan" link: eta = (4/pi)*tan(mu) i.e. this link bears the same relation to the Cauchy distribution as the probit link bears to the Gaussian. I'm interested in sage advice about this from people who know their way aroung glm. >From the surface, it looks

Hi Folks, I'm currently installing R-1.7.1 off CRAN. As it happens, I have a CD (kindly made for me by Linux Emporium) containing all the libraries which were on CRAN early this year when I installed R-1.6.1. This is highly convenient, since the alternative would be several hours on-line. While a recent library will on installation announce the fact should it need a newer version of R than

Hi Folks, I'm analysing some data which, in its simplest aspect, has 3 factors A, B, C each at 2 levels. If I do lm1 <- lm(y ~ A*B) say, and then summary(lm1, corr=T) I get the correlation matrix of the estimated coeffcients with numerical values for the correlations (3 coeffs in this case). Likewise with 'glm' instead of 'lm'. However, if I do lm2 <- lm(y ~

Something I don't understand ... (!) With the lattice library loaded, I have a loop for( Z in ... ) { ... xyplot(y~x | t, xlab=..., ylab=... ) } and no plot appears on the R graphics device. However, when I run the commands within {...} separately for each instance of Z, I get the plot displayed each time. So it looks as though "xyplot" is not outputting to the graphics

Thanks to Fernando Tusell and especially to Brian Ripley for their work on 'mix', leading to an apparently good package mow available on CRAN. Going through the R code for the function prelim.mix, I am wondering why the following method of calculation is used at one point: umd <- as.integer(round(exp(cumsum(log(d))))) (d is a vector containing, in effect, the numbers of levels of

Documentation for S3 functions is apparently written in troff markup with macro tags like .BG .FN .TL .DN .CS ... Inspection of S3 documentation source files gives a pretty clear idea of what these mean, semantically (and Sd2Rd is a perl script which converts this markup into the Rd format, providing further semantic information along the way). My query is: Can anyone point to troff

Hi folks, On quitting R with q(), is it possible to save the workspace to a directory other than the one R was started from? (I sometimes have a project "master" directory with the major R code and data in that directory, but divisions of the project having their specific stuff in sub-directories. So if I quit while running a sub-project, I'd like to save the workspace back into

Hi Folks, The inspiration for this query is described below, but it prompts a general question: If one wants to use only one or a few functions from a library, is there a way to load only these, without loading the library, short of going into the package source and extracting what is needed (including of course any auxiliary functions and compiled code they may depend on)? What prompted this

Greetings folks, Shafer's S-plus package "norm" for multiple imputation of missing values in multivariate normal data has been most kindly and usefully ported to R by Alvaro A. Novo. Shafer's website http://www.stat.psu.edu/~jls/ lists four S-plus packages in all: NORM - multiple imputation of multivariate continuous data CAT - multiple imputation of multivariate

Hi Folks, Fernando Tussell and I have been working on an R package of Shafer's CAT software for S-plus, for multiple imputation of categorical data. A very first version of this ("0.0-1") now seems to work, in that blatant bugs and segfaults seem to have been worked around. It now needs some testing in the wild, so if anyone would like to have a copy of cat_0.0-1.tar.gz to try out

Hi Folks, I want to do multivariate regression in R, i.e. basically (but with a complication -- see below): given an Nxp matrix Y of p-variate responses, and an Nxk matrix X of covariates, to fit the model Y = X*B + e with estimation of the kxp matrix of coefficients B and estimation of the pxp matrix of covariances between the p variates in Y. I haven't managed to find a

Hi Folks, Thanks to several people's suggestions and clarifications, I think I have implemented a function which computes the conditional mean and covariance matrix of a subset of the dimensions of an MV-normal variable, given the values on the other dimensions (these conditioning value can be presented as a matrix, to deal with several cases at once). The code is below, for anyone who would