similar to: weird behavior of drop1() for polr models (MASS)

In Venables \& Ripley 3rd edition (p. 231) the proportional odds model is described as: logit(p<=k) = zeta_k + eta but polr apparently thinks there is a minus in front of eta, as is apprent below. Is this a bug og a feature I have overlooked? Here is the naked code for reproduction, below the results. ------------------------------------------------------------------------ --- version

Hi, I am trying to test the proportional-odds model using the "polr" function in the MASS library with the dataset of "housing" contained in the MASS book ("Sat" (factor: low, medium, high) is the dependent variable, "Infl" (low, medium, high), "Type" (tower, apartment, atrium, terrace) and "Cont" (low, high) are the predictor variables

Hi everyone, If I compute a "Ordered Logistic or Probit Regression" with the polr function from MASS package. the summary give me : coefficients, Standard error and Tvalue.. but not directly the p.value. I can compute "manualy" the Pvalue, but Is there a way to directly obtain the pa.value, and I wonder why the p.valeu is not directly calculated, is there a reason? exemple

Hi, I want to do a global likelihood ratio test for the proportional odds logistic regression model and am unsure how to go about it. I am using the polr() function in library(MASS). 1. Is the p-value from the likelihood ratio test obtained by anova(fit1,fit2), where fit1 is the polr model with only the intercept and fit2 is the full polr model (refer to example below)? So in the case of the

I am new to R, ordered logistic regression, and polr. The "Examples" section at the bottom of the help page for polr<http://stat.ethz.ch/R-manual/R-patched/library/MASS/html/polr.html>(that fits a logistic or probit regression model to an ordered factor response) shows options(contrasts = c("contr.treatment", "contr.poly")) house.plr <- polr(Sat ~ Infl +

Dear All, I have been struggling to understand why for the housing data in MASS library R and stata give coef. estimates that are really different. I also tried to come up with many many examples myself (see below, of course I did not have the set.seed command included) and all of my `random' examples seem to give verry similar output. For the housing data, I have changed the data into numeric

Hi all, The nobs method of (MASS:::polr class) takes into account of weight, but nobs method of glm does not. I wonder what is the rationale of such design behind nobs.glm. Thanks in advance. Best Regards. > library(MASS) > house.plr <- polr(Sat ~ Infl + Type + Cont, weights = Freq, data = housing) > house.logit <- glm(I(Sat=='High') ~ Infl + Type + Cont, binomial,weights

Dear list members, The polr() function in the MASS package takes an optional weights argument for case weights. Is there any way to retrieve the case weights from the fitted "polr" object? Examining both the object and the code, I don't see how this can be done, but perhaps I've missed something. Any help would be appreciated. John ------------------------------ John Fox,

Hi all, I have a problem to estimate Std. Error and t-value by ?polr? in library Mass. They result from the summary of a polr object. I can obtain them working in the R environment with the following statements: temp <- polr(formula = formula1, data = data1) coeff <- summary(temp), but when the above statements are enclosed in a function, summary reports the following error:

Hi, I am trying to build an ordinal regression model using polr (from the MASS package). In order to construct an initial model (without an error aborting it) in my setting, I must pass in a "start" parameter. I would then like to use the "step" function to remove unnecessary variables from the model. However, this fails with the error message: > mod1 <-

Dear R Help, I have noticed some inconsistent behaviour of add1 and drop1 with a weighted linear model, which affects the interpretation of the results. I have these data to fit with a linear model, I want to weight them by the relative size of the geographical areas they represent. _________________________________________________________________________________________ > example

Hi all, I'm trying to discover the options available to me for logistic and linear regression. I'm doing some tests on a dataset and want to see how different flavours of the algorithms cope. So far for logistic regression I've tried glm(MASS) and lrm (Design) and found there is a big difference. Is there a list anywhere detailing the options available which details the specific

the data is : >table.8.3<-data.frame(expand.grid( marijuana=factor(c("Yes","No"),levels=c("No","Yes")), cigarette=factor(c("Yes","No"),levels=c("No","Yes")), alcohol=factor(c("Yes","No"),levels=c("No","Yes"))), count=c(911,538,44,456,3,43,2,279))

Dear all, I have been trying to investigate the behaviour of different weights in weighted regression for a dataset with lots of missing data. As a start I simulated some data using the following: library(MASS) N <- 200 sigma <- matrix(c(1, .5, .5, 1), nrow = 2) sim.set <- as.data.frame(mvrnorm(N, c(0, 0), sigma)) colnames(sim.set) <- c('x1', 'x2') # x1 & x2 are

Dear List, recently tried to reproduce the results of some custom model selection function after updating R, which unfortunately failed. However, I ultimately found the issue to be that testing with pchisq() in drop1() seems to have changed. In the below example, earlier versions (e.g. R 2.4.1) produce a missing P-value for the variable x, while newer versions (e.g. R 2.7.1) produce 0 (2.2e-16).

By changing three lines in drop1 from access based on $ to access based on standard accessor methods (terms() and residuals()), it becomes *much* easier to extend drop1 to work with other model types. The use of $ rather than accessors in this context seems to be an oversight rather than a design decision, but maybe someone knows better ... In particular, if one makes these changes (which I am

Hi, I was wondering why anova() and drop1() give different tail probabilities for F tests. I guess overdispersion is calculated differently in the following example, but why? Thanks for any advice, Tom For example: > x<-c(2,3,4,5,6) > y<-c(0,1,0,0,1) > b1<-glm(y~x,binomial) > b2<-glm(y~1,binomial) > drop1(b1,test="F") Single term deletions Model: y ~

Dear R Core Team I've a proposal to improve drop1(). The function should change the contrast from the default ("treatment") to "sum". If you fit a model with an interaction (which ist not signifikant) and you display the main effect with drop1( , scope = .~., test = "F") If you remove the interaction, then everything's okay. There is no way to fit a

Hi, If I used drop1 in a weighted lm fit, it seems to ignore the weights in the AIC calculation of the dropped terms, see the example below. Can this be right? Yan -------------------- library(car) > unweighted.model <- lm(trSex ~ (river+length +depth)^2- length:depth, dno2) > Anova(unweighted.model) Anova Table (Type II tests) Response: trSex Sum Sq Df F value

This is probably simple, but I just can't see it... I want to calculate the R^2s for a series of linear models where each term is dropped in turn. I can get the RSS from drop1(), and the r.squared from summary() for a given model, but don't know how to use the result of drop1() to get the r.squared for each model with one term dropped. Working example: library(vcd) # for