Testing the marginal homogeneity of a repeated measurements factor in a
complete block design.
Usage
## S3 method for class 'formula'
mh_test(formula, data, subset = NULL, ...)
## S3 method for class 'table'
mh_test(object, ...)
## S3 method for class 'SymmetryProblem'
mh_test(object, ...)
Arguments
formula
a formula of the form y ~ x | block where y and x are
factors and block is an optional factor (which is generated
automatically if omitted).
data
an optional data frame containing the variables in the model formula.
subset
an optional vector specifying a subset of observations to be used. Defaults
to NULL.
object
an object inheriting from classes "table" (with identical
dimnames components) or "SymmetryProblem".
...
further arguments to be passed to symmetry_test.
Details
mh_test provides the McNemar test, the Cochran Q test, the
Stuart(-Maxwell) test and the Madansky test of interchangeability. A general
description of these methods is given by Agresti (2002).
The null hypothesis of marginal homogeneity is tested. The response variable
and the measurement conditions are given by y and x,
respectively, and block is a factor where each level corresponds to
exactly one subject with repeated measurements.
This procedure is known as the McNemar test (McNemar, 1947) when both y
and x are binary factors, as the Cochran Q test (Cochran, 1950)
when y is a binary factor and x is a factor with an arbitrary
number of levels, as the Stuart(-Maxwell) test (Stuart, 1955; Maxwell, 1970)
when y is a factor with an arbitrary number of levels and x is a
binary factor, and as the Madansky test of interchangeability (Madansky, 1963),
which implies marginal homogeneity, when both y and x are
factors with an arbitrary number of levels.
If y and/or x are ordered factors, the default scores,
1:nlevels(y) and 1:nlevels(x) respectively, can be altered using
the scores argument (see symmetry_test); this argument
can also be used to coerce nominal factors to class "ordered". If both
y and x are ordered factors, a linear-by-linear association test
is computed and the direction of the alternative hypothesis can be specified
using the alternative argument. This extension was given by Birch
(1965) who also discussed the situation when either the response or the
measurement condition is an ordered factor; see also White, Landis and Cooper
(1982).
The conditional null distribution of the test statistic is used to obtain
p-values and an asymptotic approximation of the exact distribution is
used by default (distribution = "asymptotic"). Alternatively, the
distribution can be approximated via Monte Carlo resampling or computed
exactly for univariate two-sample problems by setting distribution to
"approximate" or "exact" respectively. See
asymptotic, approximate and exact
for details.
Value
An object inheriting from class "IndependenceTest".
Note
This function is currently computationally inefficient for data with a large
number of pairs or sets.
References
Agresti, A. (2002). Categorical Data Analysis, Second Edition.
Hoboken, New Jersey: John Wiley & Sons.
Birch, M. W. (1965). The detection of partial association, II: The general
case. Journal of the Royal Statistical Society B 27(1),
111–124.
Cochran, W. G. (1950). The comparison of percentages in matched
samples. Biometrika37(3/4), 256–266.
Madansky, A. (1963). Tests of homogeneity for correlated samples.
Journal of the American Statistical Association58(301),
97–119.
Maxwell, A. E. (1970). Comparing the classification of subjects by two
independent judges. British Journal of Psychiatry116(535),
651–655.
McNemar, Q. (1947). Note on the sampling error of the difference between
correlated proportions or percentages. Psychometrika12(2),
153–157.
Stuart, A. (1955). A test for homogeneity of the marginal distributions in a
two-way classification. Biometrika42(3/4), 412–416.
White, A. A., Landis, J. R. and Cooper, M. M. (1982). A note on the
equivalence of several marginal homogeneity test criteria for categorical
data. International Statistical Review50(1), 27–34.
Examples
## Performance of prime minister
## Agresti (2002, p. 409)
performance <- matrix(
c(794, 150,
86, 570),
nrow = 2, byrow = TRUE,
dimnames = list(
"First" = c("Approve", "Disprove"),
"Second" = c("Approve", "Disprove")
)
)
## Asymptotic McNemar Test
diag(performance) <- 0 # speed-up: only off-diagonal elements contribute
mh_test(as.table(performance))
## Exact McNemar Test
mh_test(as.table(performance), distribution = "exact")
## Effectiveness of different media for the growth of diphtheria
## Cochran (1950, Tab. 2)
cases <- c(4, 2, 3, 1, 59)
n <- sum(cases)
cochran <- data.frame(
diphtheria = factor(
unlist(rep(list(c(1, 1, 1, 1),
c(1, 1, 0, 1),
c(0, 1, 1, 1),
c(0, 1, 0, 1),
c(0, 0, 0, 0)),
cases))
),
media = factor(rep(LETTERS[1:4], n)),
case = factor(rep(seq_len(n), each = 4))
)
## Asymptotic Cochran Q test (Cochran, 1950, p. 260)
mh_test(diphtheria ~ media | case, data = cochran) # Q = 8.05
## Approximative Cochran Q test
mt <- mh_test(diphtheria ~ media | case, data = cochran,
distribution = approximate(B = 10000))
pvalue(mt) # standard p-value
midpvalue(mt) # mid-p-value
pvalue_interval(mt) # p-value interval
## Opinions on Pre- and Extramarital Sex
## Agresti (2002, p. 421)
opinions <- c("Always wrong", "Almost always wrong",
"Wrong only sometimes", "Not wrong at all")
PreExSex <- matrix(
c(144, 33, 84, 126,
2, 4, 14, 29,
0, 2, 6, 25,
0, 0, 1, 5),
nrow = 4,
dimnames = list(
"PreMaritalSex" = opinions,
"ExtraMaritalSex" = opinions
)
)
## Asymptotic Stuart test
mh_test(as.table(PreExSex))
## Asymptotic Stuart-Birch test
## Note: response as ordinal
mh_test(as.table(PreExSex), scores = list(response = 1:length(opinions)))
## Vote intention
## Madansky (1963, pp. 107-108)
vote <- array(
c(120, 1, 8, 2, 2, 1, 2, 1, 7,
6, 2, 1, 1, 103, 5, 1, 4, 8,
20, 3, 31, 1, 6, 30, 2, 1, 81),
dim = c(3, 3, 3),
dimnames = list(
"July" = c("Republican", "Democratic", "Uncertain"),
"August" = c("Republican", "Democratic", "Uncertain"),
"June" = c("Republican", "Democratic", "Uncertain")
)
)
## Asymptotic Madansky test (Q = 70.77)
mh_test(as.table(vote))
## Cross-over study
## http://www.nesug.org/proceedings/nesug00/st/st9005.pdf
dysmenorrhea <- array(
c(6, 2, 1, 4, 3, 0, 5, 2, 2,
3, 1, 0, 13, 3, 0, 10, 1, 0,
1, 2, 1, 8, 1, 1, 14, 2, 0),
dim = c(3, 3, 3),
dimnames = list(
"Placebo" = c("None", "Moderate", "Complete"),
"High dose" = c("None", "Moderate", "Complete"),
"Low dose" = c("None", "Moderate", "Complete")
)
)
## Asymptotic Madansky-Birch test (Q = 53.76)
## Note: response as ordinal
mh_test(as.table(dysmenorrhea), scores = list(response = 1:3))
## Asymptotic Madansky-Birch test (Q = 47.29)
## Note: response and measurement conditions as ordinal
mh_test(as.table(dysmenorrhea), scores = list(response = 1:3,
conditions = 3:1))