Frequentist Meta-Analysis

An approach combining evidence using classical frequentist methods, producing a point estimate and confidence interval without formally incorporating prior beliefs.

Source: DerSimonian R, Laird N. Meta-analysis in clinical trials. Controlled Clinical Trials. 1986;7(3):177-188. doi:10.1016/0197-2456(86)90046-2.

Frequentist meta-analysis combines studies using classical statistical methods, producing a pooled point estimate and a confidence interval without bringing in any prior belief about the effect. It draws its conclusions purely from the data at hand, treating the true effect as a fixed unknown and describing how the estimate would behave over repeated sampling. This keeps it straightforward and free of the need to specify priors, though it cannot formally fold in external information the way a Bayesian analysis can. An estimate from the data alone is what it gives. Borenstein and colleagues (2009) describe this approach.

Source: Borenstein et al. 2009

Frequentist meta-analysis works by extracting the effect estimate and its precision from each study, combining them through inverse-variance weighting under a fixed-effect or random-effects model to produce a pooled estimate, and expressing uncertainty as a confidence interval, with p-values for significance. Heterogeneity is assessed with statistics such as I-squared. The methods rely on the data alone, without prior distributions. So frequentist meta-analysis combines studies by weighted averaging within the frequentist framework, producing a point estimate and confidence interval that summarise the evidence based on the observed study results and the chosen model for heterogeneity.

Source: DerSimonian & Laird 1986

The advantages of frequentist meta-analysis include its familiarity and wide acceptance, its computational simplicity for standard pairwise syntheses, and its avoidance of the need to specify prior distributions, which sidesteps concerns about the influence of priors. Results are expressed in familiar confidence intervals and p-values. For straightforward meta-analyses, it is efficient and transparent. So frequentist meta-analysis is advantageous for its simplicity, familiarity, and lack of prior specification, making it the default approach for standard pairwise meta-analysis, though it is less flexible than Bayesian methods for complex syntheses such as networks.

Source: DerSimonian & Laird 1986

The limitations of frequentist meta-analysis include less natural handling of complex models, such as network meta-analysis and hierarchical structures, than Bayesian methods; difficulty fully representing uncertainty in the between-study variance, especially with few studies; and the inability to formally incorporate external prior information or to make direct probabilistic statements, such as the probability that a treatment is best. These limitations mean that, for complex syntheses or where uncertainty and prior information are important, Bayesian meta-analysis may be preferred, while frequentist meta-analysis remains suitable and standard for straightforward pairwise syntheses where its simplicity and familiarity are advantages.

Source: Dias et al. 2013

Frequentist meta-analysis combines studies using classical methods, producing a point estimate and confidence interval based on the data alone, while Bayesian meta-analysis incorporates prior information and updates it with the data to produce posterior distributions, expressing uncertainty probabilistically. Bayesian methods handle complex models and uncertainty more flexibly and allow probabilistic statements, but require priors and computation, whereas frequentist methods are simpler and more familiar but less flexible. So the two differ in whether prior information is formally used and in how results and uncertainty are expressed, with the choice depending on the complexity of the synthesis and the need for prior information.

Source: Dias et al. 2013