similar to: multinom crashes (when I do something stupid) (PR#8358)

Hello all, I came across a behavior of R with environments that I'm not sure what is causing it. It involves changing variables that are found through using model.frame on a formula inside a function. I wonder if it's a "bug" or a "feature". And in either case, how it might be managed. Here is a simple example: # let's say we have an x and y variables: temp_x

Hello all, I wrote a small function to add labels for outliers in a boxplot. This function will only work on a simple boxplot/formula command (e.g: something like boxplot(y~x)). Code + example follows in this e-mail. I'd be happy for any suggestions on how to improve this code, for example: - Handle boxplot.matrix (which shouldn't be too hard to do) - Handle cases of complex

Hello all, A colleague at work set me the challenge to convert some MATLAB code into R, to see which is faster. We'd seen that benchmark comparing MATLAB 6.5 to R1.90 (and others), and so I thought that I should be able to get roughly comparable speeds. The code has lots of multiplications of matrixes, transposes, and MATLAB's "repmat". I did the code conversion, and R was about

Hi all, Easy question for you (which I failed to find the answer to in the FAQ etc). I've recently been forced to switch from linux to windows (currently windows NT), and my usual habit of putting lots of "cat" statements in slow functions to get an idea of the progress rate is no longer useful. Why -- because R waits until the function is completely finished before printing the cat

Hello, Probably a stupidly easy question, but I have done the following in order to make predictions from a fitted glm with new data: my.glm <- glm(lt96~so296[,1:17],family=binomial(link=logit)) p96 <- predict.glm(my.glm,newdata=so293[,1:17],type="response") but I always get the fitted linear predictors from the original model, ie there doesn't seem to be acknowledgement of

Hello and thanks in advance for any advice. I am not clear how, in practice, the multinom() function in nnet and the multinomial() function in VGAM differ in terms of interpretation. I understand that they are fit differently. Are there certain scenarios where one is more appropriate than the other? In my case I have a dependent variable with 4 categories and 1 binary and 4 continuous

Hi, is there a possibility to calculate the different "types" of residuals directly using the multinom function from MASS as it is possible for the functions gam, glm using type="deviance" or "working" or "pearson" or "response"? I tried it but got always the "response" type, I guess. thanx Matthias

Is anyone aware of a way to check whether multinom has converged by checking a component of the output ? I'm not familar with nnet but, since multinom calls nnet , maybe there is an extra argument once can send to multinom to capture this information. Thanks.

Dear R users, I'm using a multinomial LOGIT model to analyse choice behaviour of consumers (as part of my masters thesis research). Using the R documentation and search on the R website I have a working script now. Parameters are estimated and I can access them via coefficients(multinom.out). In order to see if the parameters are significant I like to access the standard errors in the

Dear all users, I"d like to ask you how to make decision about colinearity among categorical independent variables when the model is multinomial logistic regression. Any help is appreciated, Niklas [[alternative HTML version deleted]]

I see that "case weights" can be optioned in multinom(). I wanted to make sure I understand what weights= is expecting. My weights (not really mine but I'm stuck with them) are noninteger, are not scaled to sum to the sample size, and larger weights are intended to increase influence. The description of various types of weights is a perennial confusion for me; sorry. STS

Hi, I am searching for methods to compare regression models with an ordered categorical response variable (polr versus multinom). The pattern of predictions of both methods (using the same predictor variables) is quite different and the AIC is smaller for the multinom approach. I guess polr has more strict premises for the structure of the response variable, which methods can be used to test for

I'd like to fit a multinomial log-linear model for 4 categories of the form log[(P(D=i | X)/P(D=0 | X)] = alpha_i + X beta_i ; i=1,2,3 but with beta_1 constrained to zero. Is there a way to impose such a constraint in the multinom function? Brad ------------------------------------------------------------------------- Brad McNeney email: mcneney at stat.sfu.ca

Dear R help list, i'm using multinom (nnet), the results given to me is the coefficients and std errors. Is there a way to obtain directly (or in an export to latex) the odd-ratio (exp(B)) and it's confident intervals. thanks very much for your help Pierre-Henry Miquel Universit?? de Lille 2 (M??decine) France Vous manquez d??espace pour stocker vos mails ?

Good Day: When I used: multinom(formula = Y ~ X1 + X2 + X3 + X1:X2 + X1:X3 + X3:X2 + X1^2 + X2^2 + X3^2, data = DATASET), I get estimates and AIC for the model containing main effects and interactions only (no squared terms)...and FYI, all predictors are continuous. Is this "normal" behavior? If I run this in S-Plus I get estimates and AIC for the model containing all terms(including

Dear R-helpers, I need to compute probabilties of multinomial observations, eg by doing the following: y=sample(1:3,15,1) prob=matrix(runif(45),15) prob=prob/rowSums(prob) diag(prob[,y]) However, my question is whether this is the most efficient way to do this. In the call prob[,y] a whole matrix is computed which seems a bit of a waste. Is there maybe a vectorized version of dmultinom which

Hello, This is the fitted model: > fit Call: multinom(formula = resp ~ pred$cls + pred$smoke) Coefficients: (Intercept) pred$cls2 pred$cls3 pred$cls4 pred$cls5 pred$smoke2 pred$smoke3 Proteinuria -1.140520 0.1616644 0.05554898 -0.01584927 0.02574805 -0.4057245 -0.2898425 Hypertension -2.691215 -0.3699690 -0.22582107 0.01615898 0.26318005 0.1239051 0.2413282

Dear all, I'm fitting a multinom function to my dataset (multinom(outcome~age+K+D)) and I need to present my results on a poster. Does someone know a nice way of doing that? I think I saw in an archive that you cannot plot a multinom.object, is it true? Thank you by advance for your help, Cheers Camille

Dear Colleagues, I am doing two class classification using logistic regression. I realized that I can either use "glm" function or "multinom" function. I know "multinom" is used for multiclass classification. But if I was it for binary classification, I was wondering if there is an difference in the results compared to "glm" results. Thanks in advance.

>On Sat, 11 Jun 2005, John Fox wrote: > >>Dear Marc, >> >>I get the same results -- same coefficients, standard errors, and fitted >>probabilities -- from multinom() and glm(). It's true that the deviances >>differ, but they, I believe, are defined only up to an additive constant: > >Yes. There are many variations on the definition >of (residual)