interval.logitsurv.discrete {mets} | R Documentation |
logit(P(T >t | x)) = log(G(t)) + x β
P(T >t | x) = \frac{1}{1 + G(t) exp( x β) }
interval.logitsurv.discrete( formula, data, beta = NULL, no.opt = FALSE, method = "NR", stderr = TRUE, weights = NULL, offsets = NULL, exp.link = 1, increment = 1, ... )
formula |
formula |
data |
data |
beta |
starting values |
no.opt |
optimization TRUE/FALSE |
method |
NR, nlm |
stderr |
to return only estimate |
weights |
weights following id for GLM |
offsets |
following id for GLM |
exp.link |
parametrize increments exp(alpha) > 0 |
increment |
using increments dG(t)=exp(alpha) as parameters |
... |
Additional arguments to lower level funtions lava::NR optimizer or nlm |
This is thus also the cumulative odds model, since
P(T ≤q t | x) = \frac{G(t) \exp(x β) }{1 + G(t) exp( x β) }
The baseline G(t) is written as cumsum(exp(α)) and this is not the standard parametrization that takes log of G(t) as the parameters.
Input are intervals given by ]t_l,t_r] where t_r can be infinity for right-censored intervals When truly discrete ]0,1] will be an observation at 1, and ]j,j+1] will be an observation at j+1
Likelihood is maximized:
∏ P(T_i >t_{il} | x) - P(T_i> t_{ir}| x)
Thomas Scheike
data(ttpd) dtable(ttpd,~entry+time2) out <- interval.logitsurv.discrete(Interval(entry,time2)~X1+X2+X3+X4,ttpd) summary(out) pred <- predictlogitSurvd(out,se=FALSE) plotSurvd(pred)