metatest
fits and tests a metaregression model. In addition to
the traditional z test on the estimated coefficients, metatest
also yields more reliable statistics: the t-test, log likelihood ratio
test, Bartlett corrected log likelihood ratio test, and the permutation
test. The Bartlett corrected log likelihood ratio test and the
permutation test are to be recommended since their type 1 errors are
adequate.
metatest(formula, variance, data, threshold = 1e-05, maxiter = 100, npermut = 1000, ...)# S3 method for metatest
summary(object, digits = 4, ...)
# S3 method for metatest
print(x, ...)
formula
specifying the meta regression model;
use y~x
to specify a meta regression of effect sizes y
moderated by x
; the moderators can be either continuous or
categorical variables; an intercept is included by default (use
y~x-1
to drop the intercept); use y~1
for an intercept
only model, i.e. a meta-analysis model.
The variances of the effect sizes to be modelled (a
vector or a variable name interpreted in data
).
A data.frame
to interpet the variables in arguments
formula
and variance
.
The threshold used in estimating the model; the threshold is the change in the value of the random effects variance parameter.
Maximum number of iterations allowed in estimating the model.
Number of permutations performed by the permutation test.
Object of class metatest
.
Determines the number of digits to use in printing the results.
Not currently used.
metatest
returns an object of class metatest
which is a named list
with the following elements:
Convergence info; 0 indicates convergence; -1 signals that the estimator of between study variance was set to zero during estimation (with a warning).
Number of iterations used in optimizing the parameters.
Matrix with between study variance estimate, its associated log likelihood ratio statistic, df and p-value.
Estimated coefficients.
Standard errors of the coefficients.
The t-ratios of the coefficients.
The p-values associated with the z-test.
The df's associated with the t-tests.
The p-values associated with the t-test.
The log likelihood ratio statistics.
The p-values associated with the LLR statistics.
The Bartlett corrected LLR statistics.
The Bartlett scaling factor used to compute the corrected LLR statistics.
The p-values associated with the Bartlett corrected LLR statistics.
The p-values of the permutation tests.
The function call that created the metatest object.
The effect sizes to be analyzed can be of arbitrary type; some
transformations
between different effect size measures
are provided. For many more see the package compute.es.
The print and summary methods are currently identical (this may change in the future), and print the random effects variance, the coefficients, and all the computed statistics and associated p-values.
Hilde M. Huizenga, Ingmar Visser & Conor V. Dolan (2011). Hypothesis testing in random effects meta-regression, British Journal of Mathematical and Statistical Psychology, 64, 1-19.
# NOT RUN {
data(metadata)
res <- metatest(y~x,yvar,data=metadata)
res
# }
Run the code above in your browser using DataCamp Workspace