Article

A comparison of statistical methods for meta-analysis

Department of Mathematics and Statistics, Richard Berry Building, The University of Melbourne, Victoria 3010, Australia.
Statistics in Medicine (Impact Factor: 2.04). 03/2001; 20(6):825-40. DOI: 10.1002/sim.650
Source: PubMed

ABSTRACT Meta-analysis may be used to estimate an overall effect across a number of similar studies. A number of statistical techniques are currently used to combine individual study results. The simplest of these is based on a fixed effects model, which assumes the true effect is the same for all studies. A random effects model, however, allows the true effect to vary across studies, with the mean true effect the parameter of interest. We consider three methods currently used for estimation within the framework of a random effects model, and illustrate them by applying each method to a collection of six studies on the effect of aspirin after myocardial infarction. These methods are compared using estimated coverage probabilities of confidence intervals for the overall effect. The techniques considered all generally have coverages below the nominal level, and in particular it is shown that the commonly used DerSimonian and Laird method does not adequately reflect the error associated with parameter estimation, especially when the number of studies is small.

Download full-text

Full-text

Available from: Ian Robert Gordon, Sep 19, 2014
2 Followers
 · 
289 Views
  • Source
    • "Examining the heterogeneity of treatment effect estimates across the studies is an important part of any meta-analysis conducted. We consider heterogeneity an estimation issue for those cases where a random effects meta-analysis is most relevant [4] [5] [6], i.e. if we aim to include potential additional variation between studies as uncertainty in our treatment evaluation. Unbiased estimation of between-study variance plays a key role in the treatment effect evaluation. "
    [Show abstract] [Hide abstract]
    ABSTRACT: Estimators for the variance between treatment effects from randomized clinical trials (RCTs) in a meta-analysis may yield divergent or even contradictory results. In a sequential meta-analysis (SMA), their properties are even more important, as they influence the point in time at which definite conclusions are drawn. In this study, we evaluated the properties of estimators of heterogeneity to be used in an SMA. We conducted an extensive simulation study with dichotomous and continuous outcome data and applied the estimators in real life examples. Bias and variance of the estimators were used as primary evaluation criteria, as well as the number of RCTs and patients from the accumulating trials needed to get stable estimates. The simulation studies showed that the well-known DerSimonian-Laird (DL) estimator largely underestimates the true value for dichotomous outcomes. The two-step DL (DL2) significantly improves this behaviour. In general, the DL2 and Paule-Mandel (PM) estimators are recommended for both dichotomous and continuous outcome data for use in an SMA.
    Contemporary clinical trials 12/2013; 37(1). DOI:10.1016/j.cct.2013.11.012 · 1.99 Impact Factor
  • Source
    • "Using this meta-analytic approach, 177 separate time dummy models are estimated (one for each merger announcement) for time-varying betas, ACAR and R 2 thereby saving the estimated coefficients and standard errors. Consistent with Brockwell and Gordon (2001), we then fit a maximum-likelihood random effects model which allows the true effect to vary by announcement. This model solves the log-likelihood function iteratively to produce an estimate of the between-announcement variance. "
    [Show abstract] [Hide abstract]
    ABSTRACT: This article demonstrates that the risk profile of acquiring banks’ common stock changes in the aftermath of a merger announcement when examining 177 large merger deals in the United States spanning from 1998 to 2010 and inclusive of the fifth and sixth merger waves. There is a tendency for beta risk to rise markedly immediately following such announcements and remains relatively high even two years afterwards. This corroborates the view that the newly consolidated big banks resulting from mergers entail higher systematic risk and, instead of providing risk diversification to shareholders, exhibit greater comovement with the market. The broad asset pricing implication here is that the ‘too big to fail’ mentality that arises from large bank mergers actually translates into more risk for shareholders and susceptibility to adverse movements in the aggregate market.
    Journal of International Financial Markets Institutions and Money 12/2013; 27:47–58. DOI:10.1016/j.intfin.2013.07.007 · 0.89 Impact Factor
  • Source
    • "). To evaluate this level of heterogeneity we applied Cochran's Q statistic, in which Q follows a v k-1 2 distribution with k = number of studies (Brockwell and Gordon, 2001). A significant v 2 value indicates a significant level of between-study variation, rather than a single underlying effect. "
    [Show abstract] [Hide abstract]
    ABSTRACT: A conspicuous feature of honey bee social bio-logy is the division of labour between reproductive queens and functionally sterile workers. However, the sterility of workers is conditional and sensitive to genetic and envi-ronmental context. Despite this understanding, we do not yet know how effective differences in genotype versus differ-ences in colony environment are for generating variation in levels of ovary activation in a population of workers. We therefore performed a field study and meta-analysis to esti-mate the standardized effect size g of broad 'environmental' and 'genetic' manipulations on worker ovary scores. Despite considerable differences in methodology and treatment among published studies, we report that both genetic and environmental manipulations were effective at generating differences in ovary phenotype between groups of worker bees. Our analysis found that environmental treatments, such as differences in pheromones and diet, had a larger mean effect on worker ovary activation scores than have genetic factors such as patriline or strain (g = 0.54 vs. 0.39). We conclude by discussing the biological significance of envi-ronmentally sensitive sterility in honey bee societies.
    Insectes Sociaux 08/2012; 59:381-388. DOI:10.1007/s00040-012-0230-1 · 1.31 Impact Factor
Show more