Package generalhoslem

December 16, 2016

Type Package
Title Goodness of Fit Tests for Logistic Regression Models
Version 1.2.4
Date 2016-12-16
Author Matthew Jay [aut, cre]
Maintainer Matthew Jay <[email protected]>
Description Functions to assess the goodness of fit of binary, multinomial and ordinal logistic models.
Included are the Hosmer-Lemeshow tests (binary, multinomial and ordinal) and the Lipsitz and
Pulkstenis-Robinson tests (ordinal).
Depends reshape, MASS, epiR, R (>= 3.3.1)
License GPL-2
Suggests nnet, mlogit, ordinal
NeedsCompilation no
Repository CRAN
Date/Publication 2016-12-16 11:20:32

R topics documented:
generalhoslem-package . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
lipsitz.test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
logitgof . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
pulkrob.chisq . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

Index 9

1
2 generalhoslem-package

generalhoslem-package Goodness of Fit Tests for Logistic Regression Models

Description
Functions to assess the goodness of fit of binary, multinomial and ordinal logistic models. Included
are the Hosmer-Lemeshow tests (binary, multinomial and ordinal) and the Lipsitz and Pulkstenis-
Robinson tests (ordinal).

Details

Package: generalhoslem
Type: Package
Title: Goodness of Fit Tests for Logistic Regression Models
Version: 1.2.4
Date: 2016-12-16
Authors@R: person("Matthew", "Jay", role = c("aut", "cre"), email = "[email protected]")
Author: Matthew Jay [aut, cre]
Maintainer: Matthew Jay <[email protected]>
Description: Functions to assess the goodness of fit of binary, multinomial and ordinal logistic models. Included are the Hos
Depends: reshape, MASS, epiR, R (>= 3.3.1)
License: GPL-2
Suggests: nnet, mlogit, ordinal

Index of help topics:

generalhoslem-package Goodness of Fit Tests for Logistic Regression

Models
lipsitz.test Lipsitz goodness of fit test for ordinal
response models.
logitgof Hosmer-Lemeshow Tests for Logistic Regression
Models
pulkrob.chisq Pulkstenis-Robinson goodness of fit tests for
ordinal response models.

Author(s)
Matthew Jay [aut, cre] Maintainer: Matthew Jay <[email protected]>

References
Fagerland MW, Hosmer DW, Bofin AM. Multinomial goodness-of-fit tests for logistic regres-
sion models. Statistics in Medicine 2008;27(21):4238-53.
Fagerland MW, Hosmer DW. A goodness-of-fit test for the proportional odds regression model.
Statistics in Medicine 2013;32:2235-2249.
lipsitz.test 3

Fagerland MW, Hosmer DW. Tests for goodness of fit in ordinal logistic regression models.
Journal of Statistical Computation and Simulation 2016. DOI: 10.1080/00949655.2016.1156682.
Hosmer DW, Lemeshow S, Sturdivant RX. Applied Logistic Regression, 3rd Edition. 2013.
New York, USA: John Wiley and Sons.
Lipsitz SR, Fitzmaurice GM, Molenberghs G. Goodness-of-Fit Tests for Ordinal Response
Regression Models. Journal of the Royal Statistical Society 1996;45(2):175-190.
Pulkstenis E, Robinson TJ. Goodness-of-fit tests for ordinal response regression models. Statis-
tics in Medicine 2004;23:999-1014.

lipsitz.test Lipsitz goodness of fit test for ordinal response models.

Description
Performs the Lipsitz goodness-of-fit test for ordinal logistic regresison models.

Usage
lipsitz.test(model, g = 10)

Arguments
model an ordinal response model. Must be an object of class polr or clm.
g number of quantiles of risk, defaults to 10.

Details
The Lipsitz test is a goodness-of-fit test for ordinal response logistic regression models. It involves
firstly binning the observed data into percentiles in the same manner as for the multinomial Hosmer-
Lemeshow test. Strictly speaking, the Lipsitz test first requires an ordinal score be assigned to each
subject. This score is the sum of the predicted probabilities for each outcome level within each
subject multiplied by equally spaced integer weights but this has been shown to be idenitical to
the more straightforward binning method (Fagerland and Hosmer, 2013). The user can specify the
number of groups by assigning an integer value to g, which is 10 by default.
Given this partitioning of the data, dummy variables, I, are derived such that, for each group, I = 1
if the subject is in region g and I = 0 if not. The model is then re-fit with these dummy variables. If
the model has good fit, then the logits for all these dummy variables simultaneously = 0. Lipsitz et
al (1996) suggest that likelihood ratio, Wald or score tests can be used but lipsitz.test just uses
the likelihood ratio test with g-1 degrees of freedom.
Note that the outcome variable MUST be converted to a factor before running the model. Us-
ing as.factor() within the model function will cause an error because of the way in which
lipsitz.test uses the update() function to re-fit the mode.
It is recommended (Fagerland and Hosmer, 2016) that the Lipsitz test be run alongside the Hosmer-
Lemeshow test (logitgof) and the Pulkstenis-Robinson tests (pulkrob.chisq and pulkrob.deviance).
4 lipsitz.test

Value

A list of class htest containing:

statistic the value of the likelihood ratio statistic.

parameter degrees of freedom used.
p.value the p-value.
method a character string indicating the name of the test.
data.name a character string indicating the model formula used.
newmoddata a data frame containing the data used in fitting the updated model (essentially
the original data plus the dummy variables).
predictedprobs a data frame of predicted probabilities from the original model.

Author(s)

Matthew Alexander Jay

References

Fagerland MW, Hosmer DW. A goodness-of-fit test for the proportional odds regression model.
Statistics in Medicine 2013;32:2235-2249.
Fagerland MW, Hosmer DW. Tests for goodness of fit in ordinal logistic regression models.
Journal of Statistical Computation and Simulation 2016. DOI: 10.1080/00949655.2016.1156682.
Lipsitz SR, Fitzmaurice GM, Molenberghs G. Goodness-of-Fit Tests for Ordinal Response
Regression Models. Journal of the Royal Statistical Society 1996;45(2):175-190.

See Also

logitgof, pulkrob.chisq.

Examples

data(mtcars)
mtcars$gear <- as.factor(mtcars$gear) # Outcome must be converted to factor
# before fitting model otherwise
# lipsitz.test() will fail.
mod1 <- polr(gear ~ mpg + cyl, data = mtcars)
lipsitz.test(mod1)
logitgof 5

logitgof Hosmer-Lemeshow Tests for Logistic Regression Models

Description
Performs the Hosmer and Lemeshow goodness-of-fit tests for binary, multinomial and ordinal re-
gression models.

Usage
logitgof(obs, exp, g = 10, ord = FALSE)

Arguments
obs a vector of observed values. See details.
exp expected values fitted by the model. See details.
g number of quantiles of risk, 10 by default.
ord logical indicating whether to run the ordinal version, FALSE by default.

Details
The Hosmer-Lemeshow tests The Hosmer-Lemeshow tests are goodness of fit tests for binary,
multinomial and ordinal logistic regression models. logitgof is capable of performing all three.
Essentially, they compare observed with expected frequencies and compute a test statistic which is
roughly distributed according to the chi-squared distribution. The degrees of freedom depend upon
the number of quantiles used and the number of outcome categories. A non-significant p value
indicates that there is no evidence that the observed and expected frequencies differ.
Binary version If obs is a vector of 1s and 0s or a factor vector with 2 levels, then the binary version
of the test is run. exp must be the fitted values obtained from the model, which can be accessed
using the fitted() function.
Multinomial version If obs is a factor with three or more levels and ord = FALSE, the multinomial
version of the test is run. If using the mlogit package to run a model, ensure outcome = FALSE in
the fitted() function. See examples.
Ordinal version If obs is a factor with three or more levels and ord = TRUE, the ordinal version of
the test is run. See examples for how to extract fitted values from models constructed using polr in
package MASS or clm in ordinal.
Note that Fagerland and Hosmer recommend running the Hosmer-Lemeshow test for ordinal mod-
els alongisde the Lipsitz test (lipsitz.test) and Pulkstenis-Robinson tests (pulkrob.chisq and
pulkrob.deviance), as each detects different types of lack of fit.

Value
A list of class htest containing:

statistic the value of the relevant test statistic.

6 logitgof

parameter the number of degrees of freedom used.

p.value the p-value.
method a character string indicating whether the binary or multinomial version of the
test was performed.
data.name a character string containing the names of the data passed to obs and exp.
observed a table of observed frequencies with g rows. Either an xtabs generated table
(used in the binary version) or a cast generated data frame (multinomial ver-
sion).
expected a table of expected frequencies with g rows. Either an xtabs generated table or
a cast generated data frame.
stddiffs a table of the standardised differences. See Hosmer, Lemeshow and Sturdivant
(2013), p 162.

Author(s)
Matthew Alexander Jay, with code adapted from the hoslem.test function written by Peter Soly-
mos for the ResourceSelection package.

References
Fagerland MW, Hosmer DW, Bofin AM. Multinomial goodness-of-fit tests for logistic regres-
sion models. Statistics in Medicine 2008;27(21):4238-53.
Fagerland MW, Hosmer DW. A goodness-of-fit test for the proportional odds regression model.
Statistics in Medicine 2013;32:2235-2249.
Fagerland MW, Hosmer DW. Tests for goodness of fit in ordinal logistic regression models.
Journal of Statistical Computation and Simulation 2016. DOI: 10.1080/00949655.2016.1156682.
Hosmer DW, Lemeshow S, Sturdivant RX. Applied Logistic Regression, 3rd Edition. 2013.
New York, USA: John Wiley and Sons.

See Also
lipsitz.test, pulkrob.chisq.

Examples
## Binary model
# 1/0 coding
data(mtcars)
mod1 <- glm(vs ~ cyl + mpg, data = mtcars, family = binomial)
logitgof(mtcars$vs, fitted(mod1))

# factor name coding

mtcars$engine <- factor(ifelse(mtcars$vs==0, "V", "S"), levels = c("V", "S"))
mod2 <- glm(engine ~ cyl + mpg, data = mtcars, family = binomial)
logitgof(mtcars$engine, fitted(mod2))

## Multinomial model
# with nnet
pulkrob.chisq 7

library(nnet)
mod3 <- multinom(gear ~ mpg + cyl, data = mtcars)
logitgof(mtcars$gear, fitted(mod3))

# with mlogit
library(mlogit)
data("Fishing", package = "mlogit")
Fish <- mlogit.data(Fishing, varying = c(2:9), shape = "wide", choice = "mode")
mod4 <- mlogit(mode ~ 0 | income, data = Fish)
logitgof(Fishing$mode, fitted(mod4, outcome = FALSE))

## Ordinal model
# polr in package MASS
mod5 <- polr(as.factor(gear) ~ mpg + cyl, data = mtcars)
logitgof(mtcars$gear, fitted(mod5), g = 5, ord = TRUE)

# clm in package ordinal

library(ordinal)
mtcars$gear <- as.factor(mtcars$gear)
mod6 <- clm(gear ~ mpg + cyl, data = mtcars)
predprob <- data.frame(mpg = mtcars$mpg, cyl = mtcars$cyl)
fv <- predict(mod6, newdata = predprob, type = "prob")$fit
logitgof(mtcars$gear, fv, g = 5, ord = TRUE)

pulkrob.chisq Pulkstenis-Robinson goodness of fit tests for ordinal response models.

Description
Performs the Pulkstenis-Robinson goodness of fit chi-squared and deviance tests for ordinal logistic
regresison models.

Usage
pulkrob.chisq(model, catvars)
pulkrob.deviance(model, catvars)

Arguments
model an ordinal response model. Must be an object of class polr or clm.
catvars a character vector containing the names of the categorical covariates.

Details
The Pulkstenis-Robinson tests are goodness-of-fit tests for ordinal response models. They are ca-
pable of accommodating models with continuous as well as categorical predictors. The data are
partitioned according to observed covariate patterns using the categorical covariates only. Any
unobserved covariate patterns are discarded and only categorical predictors are used to avoid par-
titioning among an unacceptably high number of covariate patterns. Each subject is assigned an
8 pulkrob.chisq

ordinal response score by summing the predicted probabilities of each subject for each outcome
level multiplied by equally spaced integer weights. The covariate patterns are then split into two at
the median score within each.
Based on this partitioning, observed and expected frequencies are calculated and the test statistic
computed. This statistic is distributed by the chi-squared distribution with (2I 1)(J 1) k 1
degress of freedom, where I is the number of covariate patterns, J is the number of of response
categories and k is the number of categorical variables in the model.
It is recommended (Fagerland and Hosmer, 2016) that the Pulkstenis-Robinson tests be run along-
side the Hosmer-Lemeshow test (logitgof) and the Lipsitz test (lipsitz.test).

Value
A list of class htest containing:
statistic the chi-squared or deviance statistic.
parameter degrees of freedom used.
p.value the p-value.
method a character string indicating the name of the test.
data.name a character string indicating the model formula used.
observed a cast generated data frame of observed frequencies.
expected a cast generated data frame of expected frequencies.
stddiffs a table of the standardised differences. See Hosmer, Lemeshow and Sturdivant
(2013), p 162.

Author(s)
Matthew Alexander Jay

References
Fagerland MW, Hosmer DW. Tests for goodness of fit in ordinal logistic regression models.
Journal of Statistical Computation and Simulation 2016. DOI: 10.1080/00949655.2016.1156682.
Hosmer DW, Lemeshow S, Sturdivant RX. Applied Logistic Regression, 3rd Edition. 2013.
New York, USA: John Wiley and Sons.
Pulkstenis E, Robinson TJ. Goodness-of-fit tests for ordinal response regression models. Statis-
tics in Medicine 2004;23:999-1014.

See Also
logitgof, lipsitz.test.

Examples
data(mtcars)
mod1 <- polr(as.factor(gear) ~ mpg + cyl + vs, data = mtcars)
pulkrob.chisq(mod1, c("vs"))
pulkrob.deviance(mod1, c("vs"))
Index

Topic package
generalhoslem-package, 2
Topic regression
lipsitz.test, 3
logitgof, 5
pulkrob.chisq, 7

generalhoslem (generalhoslem-package), 2
generalhoslem-package, 2

lipsitz.test, 3, 6, 8
logitgof, 4, 5, 8

pulkrob.chisq, 4, 6, 7
pulkrob.deviance (pulkrob.chisq), 7