mets (version 1.3.4)

mlogit: Multinomial regression based on phreg regression

Description

Fits multinomial regression model $$ P_i = \frac{ \exp( X^\beta_i ) }{ \sum_{j=1}^K \exp( X^\beta_j ) }$$ for $$i=1,..,K$$ where $$\beta_1 = 0$$, such that $$\sum_j P_j = 1$$ using phreg function. Thefore the ratio $$\frac{P_i}{P_1} = \exp( X^\beta_i )$$

Usage

mlogit(formula, data, offset = NULL, weights = NULL, fix.X = FALSE, ...)

Arguments

formula

formula with outcome (see coxph)

data

data frame

offset

offsets for partial likelihood

weights

for score equations

fix.X

to have same coefficients for all categories

...

Additional arguments to lower level funtions

Author

Thomas Scheike

Details

Coefficients give log-Relative-Risk relative to baseline group (first level of factor, so that it can reset by relevel command). Standard errors computed based on sandwhich form $$ DU^-1 \sum U_i^2 DU^-1$$.

Can also get influence functions (possibly robust) via iid() function, response should be a factor.

Can fit cumulative odds model as a special case of interval.logitsurv.discrete

Examples

data(bmt)
dfactor(bmt) <- cause1f~cause
drelevel(bmt,ref=3) <- cause3f~cause
dlevels(bmt)

mreg <- mlogit(cause1f~+1,bmt)
summary(mreg)

mreg <- mlogit(cause1f~tcell+platelet,bmt)
summary(mreg)

mreg3 <- mlogit(cause3f~tcell+platelet,bmt)
summary(mreg3)

## inverse information standard errors 
lava::estimate(coef=mreg3$coef,vcov=mreg3$II)

## predictions based on seen response or not 
newdata <- data.frame(tcell=c(1,1,1),platelet=c(0,1,1),cause1f=c("2","1","0"))
predictmlogit(mreg,newdata,response=FALSE)
predictmlogit(mreg,newdata)