R Graphical Manual

Browse All

Last data update: 2014.03.03

R: Univariate meta-analysis.

uvmeta

R Documentation

Univariate meta-analysis.

Description

This function performs a univariate meta-analysis by assuming fixed or random effects. Whereas the fixed effects model assumes that all studies in the analysis share a common effect size, the random-effects model allows different study-specific effect sizes. Concretely, if we move from fixed-effect weights to random-effects weights, large studies lose influence and small studies gain influence (Borenstein 2010).

Usage

uvmeta(r, vars, model="random", method="MOM", labels, na.action, pars, 
       verbose=FALSE, ...)

Arguments

`r`	vector of numerics containing the effect sizes
`vars`	vector of numerics containing the error variance of the effect sizes
`model`	Assume "random" or "fixed" effects.
`method`	Estimation method: use "MOM" to implement the non-parametric method-of-moment estimator from DerSimonian and Laird, "ml" to implement the maximum-likelihood estimator, "pl" to use the profile-likelihood estimator or "bayes" to implement a Bayesian meta-analysis assuming normality of the random effects (Higgins 2009).
`labels`	vector of characters containing the labels for the studies
`na.action`	a function which indicates what should happen when the data contain NAs. Defaults to "na.fail", other options are "na.omit", "na.exclude" or "na.pass".
`pars`	A list with additional arguments. Use "level" to specify the level of confidence or credibility intervals. The following parameters configure the MCMC sampling procedure and are ignored if `method="MOM"`: `hp.mu.mean` (Hyperparameter: mean of the prior distribution of the fixed/random effects model, defaults to zero), `hp.mu.var` (Hyperparameter: variance of the prior distribution of the fixed/random effects model, defaults to 1000), `n.chains` (specifies the number of parallel chains), `n.adapt` (specifies the number of iterations for adaptation), `n.init` (number of iterations to run for initializing the Markov chain) and `n.iter` (number of iterations to monitor).
`verbose`	if TRUE then messages generated during the fitting process will be displayed.
`...`	arguments to be passed on to other functions

Details

The Bayesian approach uses an uninformative Normal prior for the mean and an uninformative uniform prior for the variance of the pooled effect size (Higgins 2009). For random effects models, a prediction interval for the pooled effect size is displayed. This interval predicts in what range future effect sizes will fall given what has already been observed (Higgins 2009, Riley 2011).

Value

An object of the class uvmeta for which many standard methods are available.

Author(s)

Thomas Debray <thomas.debray@gmail.com>

References

Biggerstaff BJ, Tweedie RL. Incorporating variability in estimates of heterogeneity in the random effects model in meta-analysis. Statistics in Medicine 1997; 16: 753–768.

Borenstein M, Hedges LV, Higgins JPT, Rothstein HR. A basic introduction to fixed-effect and random-effects models for meta-analysis. Research Synthesis Methods 2010; 1: 97–111.

DerSimonian R, Laird N. Meta-analysis in clinical trials. Controlled Clinical Trials 1986; 7: 177–188.

Gasparrini A, Armstrong B, Kenward, MG. Multivariate meta-analysis for non-linear and other multi-parameter associations. Statistics in Medicine 2012; 31: 3821–3839.

Graham PL, Moran JL. Robust meta-analytic conclusions mandate the provision of prediction intervals in meta-analysis summaries. Journal of Clinical Epidemiology 2012; 65: 503–510.

Higgins JPT, Thompson SG. Quantifying heterogeneity in a meta-analysis. Statistics in Medicine 2002; 21: 1539–1558.

Higgins JPT, Thompson SG, Spiegelhalter DJ. A re-evaluation of random-effects meta-analysis. Journal of the Royal Statistical Society. Series A (Statistics in Society) 2009, 172: 137–159.

Riley RD, Higgins JPT, Deeks JJ. Interpretation of random effects meta-analyses. British Medical Journal 2011; 342: d549.

Rucker G, Schwarzer G, Carpenter JR, Schumacher M . Undue reliance on I2 in assessing heterogeneity may mislead. BMC Medical Research Methodology 2008; 8: 79.

Examples

data(Roberts)

#Extract effect size and error variance
r <- Roberts$SDM
vars <- Roberts$SE**2

#Frequentist random-effects meta-analysis
fit1 <- uvmeta(r,vars,labels=rownames(Roberts))
plot(fit1, main="Forest plot") #show a forest plot
fit1


#Bayesian random-effects meta-analysis
#fit2 <- uvmeta(r,vars,method="bayes")
#fit2