1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
|
\name{predict.lrm}
\alias{predict.lrm}
\alias{predict.orm}
\alias{Mean.lrm}
\alias{Mean.orm}
\title{
Predicted Values for Binary and Ordinal Logistic Models
}
\description{
Computes a variety of types of predicted values for fits from
\code{lrm} and \code{orm}, either from the original dataset or for new
observations. The \code{Mean.lrm} and \code{Mean.orm} functions produce
an R function to compute the predicted mean of a numeric ordered
response variable given the linear predictor, which is assumed to use
the first intercept in its computation.
}
\usage{
\method{predict}{lrm}(object, \dots, type=c("lp", "fitted",
"fitted.ind", "mean", "x", "data.frame",
"terms", "cterms", "ccterms", "adjto","adjto.data.frame",
"model.frame"), se.fit=FALSE, codes=FALSE)
\method{predict}{orm}(object, \dots, type=c("lp", "fitted",
"fitted.ind", "mean", "x", "data.frame",
"terms", "cterms", "ccterms", "adjto","adjto.data.frame",
"model.frame"), se.fit=FALSE, codes=FALSE)
\method{Mean}{lrm}(object, codes=FALSE, \dots)
\method{Mean}{orm}(object, codes=FALSE, \dots)
}
\arguments{
\item{object}{a object created by \code{lrm} or \code{orm}}
\item{\dots}{
arguments passed to \code{predictrms}, such as \code{kint} and \code{newdata}
(which is used if you are predicting \code{out of data}). See
\code{predictrms} to see how NAs are handled. Ignored for other functions.
}
\item{type}{
See \code{predict.rms} for \code{"x", "data.frame", "terms", "cterms",
"ccterms", "adjto", "adjto.data.frame"} and
\code{"model.frame"}. \code{type="lp"} is used to get
linear predictors (using the first intercept by default; specify
\code{kint} to use others). \code{type="fitted"}
is used to get all the probabilities \eqn{Y\geq j}.
\code{type="fitted.ind"} gets all the individual probabilities
\eqn{Y=j} (not recommended for \code{orm} fits). For an ordinal response
variable, \code{type="mean"} computes
the estimated mean \eqn{Y} by summing values of \eqn{Y}
multiplied by the estimated \eqn{Prob(Y=j)}. If \eqn{Y} was a character or
\code{factor} object, the levels are the character values or factor levels,
so these must be translatable to numeric, unless \code{codes=TRUE}.
See the Hannah and Quigley reference below for the method of estimating
(and presenting) the mean score. If you specify
\code{type="fitted","fitted.ind","mean"} you may not specify \code{kint}.
}
\item{se.fit}{
applies only to \code{type="lp"}, to get standard errors.
}
\item{codes}{
if \code{TRUE}, \code{type="mean"}, \code{Mean.lrm}, and \code{Mean.orm}
use the integer codes \eqn{1,2,\ldots,k} for the \eqn{k}-level response
in computing the predicted mean response.
}
}
\value{
a vector (\code{type="lp"} with \code{se.fit=FALSE}, or
\code{type="mean"} or only one
observation being predicted), a list (with elements \code{linear.predictors}
and \code{se.fit} if \code{se.fit=TRUE}), a matrix (\code{type="fitted"}
or \code{type="fitted.ind"}), a data frame, or a design matrix. For
\code{Mean.lrm} and \code{Mean.orm}, the result is an R function.
}
\author{
Frank Harrell\cr
Department of Biostatistics\cr
Vanderbilt University\cr
f.harrell@vanderbilt.edu
}
\references{
Hannah M, Quigley P: Presentation of ordinal regression analysis on the
original scale. Biometrics 52:771--5; 1996.
}
\seealso{
\code{\link{lrm}}, \code{\link{orm}}, \code{\link{predict.rms}},
\code{\link{naresid}}, \code{\link{contrast.rms}}
}
\examples{
# See help for predict.rms for several binary logistic
# regression examples
# Examples of predictions from ordinal models
set.seed(1)
y <- factor(sample(1:3, 400, TRUE), 1:3, c('good','better','best'))
x1 <- runif(400)
x2 <- runif(400)
f <- lrm(y ~ rcs(x1,4)*x2, x=TRUE) #x=TRUE needed for se.fit
# Get 0.95 confidence limits for Prob[better or best]
L <- predict(f, se.fit=TRUE) #omitted kint= so use 1st intercept
plogis(with(L, linear.predictors + 1.96*cbind(-se.fit,se.fit)))
predict(f, type="fitted.ind")[1:10,] #gets Prob(better) and all others
d <- data.frame(x1=c(.1,.5),x2=c(.5,.15))
predict(f, d, type="fitted") # Prob(Y>=j) for new observation
predict(f, d, type="fitted.ind") # Prob(Y=j)
predict(f, d, type='mean', codes=TRUE) # predicts mean(y) using codes 1,2,3
m <- Mean(f, codes=TRUE)
lp <- predict(f, d)
m(lp)
# Can use function m as an argument to Predict or nomogram to
# get predicted means instead of log odds or probabilities
dd <- datadist(x1,x2); options(datadist='dd')
m
plot(Predict(f, x1, fun=m), ylab='Predicted Mean')
# Note: Run f through bootcov with coef.reps=TRUE to get proper confidence
# limits for predicted means from the prop. odds model
options(datadist=NULL)
}
\keyword{models}
\keyword{regression}
\concept{logistic regression model}
|
