Hostname: page-component-cb9f654ff-hn9fh Total loading time: 0 Render date: 2025-08-12T10:07:16.294Z Has data issue: false hasContentIssue false

Meta-analyzing correlation matrices in the presence of hierarchical effect size multiplicity

Published online by Cambridge University Press:  07 August 2025

Ronny Scherer*
Affiliation:
Centre for Educational Measurement (CEMO), Faculty of Educational Sciences, https://ror.org/01xtthb56University of Oslo, Oslo, Norway Centre for Research on Equality in Education (CREATE), Faculty of Educational Sciences, https://ror.org/01xtthb56University of Oslo, Oslo, Norway
Diego G. Campos
Affiliation:
Centre for Educational Measurement (CEMO), Faculty of Educational Sciences, https://ror.org/01xtthb56University of Oslo, Oslo, Norway Centre for Research on Equality in Education (CREATE), Faculty of Educational Sciences, https://ror.org/01xtthb56University of Oslo, Oslo, Norway
*
Corresponding author: Ronny Scherer; Email: ronny.scherer@cemo.uio.no
Rights & Permissions [Opens in a new window]

Abstract

To synthesize evidence on the relations among multiple constructs, measures, or concepts, meta-analyzing correlation matrices across primary studies has become a crucial analytic approach. Common meta-analytic approaches employ univariate or multivariate models to estimate a pooled correlation matrix, which is subjected to further analyses, such as structural equation modeling. In practice, meta-analysts often extract multiple correlation matrices per study from various samples, study sites, labs, or countries, thus introducing hierarchical effect size multiplicity into the meta-analytic data. However, this feature has largely been ignored when pooling correlation matrices for meta-analysis. To contribute to the methodological development in this area, we describe a multilevel, multivariate, and random-effects modeling approach, which pools correlation matrices meta-analytically and, at the same time, addresses hierarchical effect size multiplicity. Specifically, it allows meta-analysts to test various assumptions on the dependencies among random effects, aiding the selection of a meta-analytic baseline model. We describe this approach, present four working models within it, and illustrate them with an example and the corresponding R code.

Information

Type
Tutorial
Creative Commons
Creative Common License - CCCreative Common License - BY
This is an Open Access article, distributed under the terms of the Creative Commons Attribution licence (https://creativecommons.org/licenses/by/4.0), which permits unrestricted re-use, distribution and reproduction, provided the original article is properly cited.
Copyright
© The Author(s), 2025. Published by Cambridge University Press on behalf of The Society for Research Synthesis Methodology

Highlights

What is already known?

  • Meta-analyzing correlation matrices allows meta-analysts to test theories and hypotheses on the relations between multiple variables.

  • Pooling correlation matrices meta-analytically across studies often relies on univariate random-effects models or multivariate random-effects models (MV-REMs).

  • Hierarchical effect size multiplicity can arise when meta-analysts extract multiple correlation matrices from a study, often due to the inclusion of multiple samples, study sites, or countries.

  • Existing pooling approaches either largely ignore hierarchical effect size multiplicity or only partially account for it.

What is new?

  • Multilevel extensions of MV-REMs provide a framework for addressing hierarchical effect size multiplicity in the meta-analytic pooling of correlation matrices.

  • Dependencies among correlations within correlation matrices can be specified in the asymptotic sampling covariance matrices.

  • Multilevel, multivariate, and random-effects models (MLMV-REMs) enable meta-analysts to model the heterogeneity in correlation matrices within and between studies.

  • These models can be estimated using the multilevel and multivariate features in the R package “metafor.”

Potential impact for RSM readers

  • Meta-analysts who wish to synthesize correlation matrices in the presence of hierarchical effect size multiplicity may take an MLMV-REM approach.

  • We aim to encourage the development and evaluation of models addressing effect size multiplicity in the meta-analytic pooling of correlation matrices.

  • We offer an illustrative example along with the corresponding analytical R code that meta-analysts can adapt to their research.

1 Introduction

Understanding how multiple constructs, concepts, or variables are related and testing theories that describe these relations requires the synthesis of multiple correlation coefficients.Reference Becker, Aloe, Cheung, Schmid, Stijnen and White 1 Multiple correlation coefficients are often stored in correlation matrices, and synthesizing them across primary studies means to pool the correlation coefficients via meta-analytic methods into an overall, weighted average correlation matrix.Reference Cheung and Cheung 2 Notably, correlation coefficients within a correlation matrix are not independent, as they are usually derived from the same sample. Given this correlational dependence (“correlational effect size multiplicity”), the pooling of correlation matrices often involves multivariate models that provide more accurate meta-analytic estimates and standard errors than multiple, separate univariate models.Reference Cheung 3 , Reference Jak 4

With the growing availability of published studies and data, multiple independent samples, study sites, or countries may be included per primary study. This scenario is referred to as “hierarchical effect size multiplicity.”Reference Pustejovsky and Tipton 5 For instance, in their study of the relations among technology acceptance variables, Scherer et al.Reference Scherer, Siddiq and Tondeur 6 extracted multiple correlation matrices from primary studies, because some studies included more than one teacher sample (e.g., in-service and pre-service teachers). The resultant meta-analytic data comprised multiple correlation matrices that were hierarchically nested in studies.Reference Campos, Cheung and Scherer 7

While dealing with hierarchical effect size multiplicity has been well described for univariate outcomes (e.g., via multilevel random-effects modelsReference Fernández-Castilla, Jamshidi, Declercq, Beretvas, Onghena and Van den Noortgate 8 or cluster-robust variance estimationReference Hedges, Tipton and Johnson 9 of one correlation coefficient), it has received less attention for multivariate outcomes, especially when correlation matrices are pooled. Similar to a univariate scenario, ignoring hierarchical effect size multiplicity when meta-analyzing multivariate data may result in biased standard errors and heterogeneity estimates.Reference Stolwijk, Jak, Eichelsheim and Hoeve 10 , Reference Ke, Zhang and Tong 11 In our view, existing approaches to pool correlation matrices meta-analytically largely ignored hierarchical effect size multiplicity or addressed it only to a limited extent.Reference Stolwijk, Jak, Eichelsheim and Hoeve 10 For instance, pooling correlation matrices across studies—as a key stage in meta-analytic structural equation modeling (SEM)—is often based on multivariate random-effects models (MV-REMs) assuming that correlation matrices are fully independent. Acknowledging this limitation, Wilson et al.Reference Wilson, Polanin and Lipsey 12 developed a multilevel random-effects modeling approach—referred to as the “WPL approach”—that quantifies within- and between-study heterogeneity of the correlation coefficients within correlation matrices. However, this approach assumes that within- and between-heterogeneity are the same across correlations—an assumption that may or may not hold in practical applications.

To address these limitations, we present a multilevel, multivariate, and random-effects modeling (MLMV-REM) approach for pooling correlation matrices in the presence of hierarchical effect size multiplicity. Extending the existing approaches, the MLMV-REM approach allows meta-analysts to flexibly specify random-effects structures within and between studies, obtain information about correlation-specific heterogeneity, and examine moderator effects. In this tutorial paper, we describe the approach, highlight the key analytical decisions within, present four working models, and illustrate its application with an example, including the respective R code. Despite its flexibility, we hope to stimulate future evaluations of the MLMV-REM approach and to advance the development of approaches that deal with effect size multiplicity when pooling correlation matrices meta-analytically.

2 Existing approaches to pooling correlation matrices meta-analytically

In this section, we explain the different types of dependencies among correlation coefficients that may occur when meta-analyzing correlation matrices and review univariate and multivariate pooling approaches. While various methods have been developed to pool correlation matrices meta-analytically, each presents specific limitations. These shortcomings highlight the need for a more flexible and comprehensive method that can address both the correlational and hierarchical dependencies that may occur in meta-analytic data, ensuring more accurate and reliable meta-analytic conclusions.

2.1 Dependencies among effect sizes

While pooling correlation matrices across studies has gained popularity in meta-analysis, it still faces significant challenges. One of these challenges refers to effect size multiplicity that occurs due to the dependencies among effect sizes both within and across studies.Reference Stolwijk, Jak, Eichelsheim and Hoeve 10 , Reference Wilson, Polanin and Lipsey 12 , Reference Sheng, Kong, Cortina and Hou 13 CheungReference Cheung 3 summarized effect size dependencies as (a) dependencies among the sampling errors that occur due to the administration of multiple measures to the same sample in each study, (b) dependencies among true effect sizes at the population level across studies, and (c) dependencies due to hierarchies in the data, such as multiple correlation coefficients that are nested in studies.

In the context of pooling correlation matrices meta-analytically, such dependencies can occur not only because multiple correlation coefficients are stored in correlation matrices, but also because multiple correlation matrices are extracted from one study. For example, in a meta-analysis of psychological traits—such as anxiety, depression, and stress—studies might report multiple correlation matrices for samples defined by different clinical statuses (e.g., clinical vs. non-clinical samples). In another meta-analysis, authors may extract correlation matrices describing the relations among student achievement, motivation, and engagement in reading, mathematics, and science, using data from international large-scale assessments in education, such as the Programme for International Student Achievement or the Trends in Mathematics and Science Study.Reference Campos, Cheung and Scherer 7 , Reference Scherer, Siddiq and Nilsen 14 Given the inclusion of several countries in these assessments, multiple correlation matrices are available per assessment cycle.

These examples illustrate that dependencies among multiple correlation coefficients may not only exist within a correlation matrix, because multiple correlations were derived from the same sample, but also across multiple constructs or measures (i.e., correlational effect size multiplicity).Reference Pustejovsky and Tipton 5 The inclusion of multiple correlation matrices per study that are derived from independent samples also creates hierarchical effect size multiplicity.Reference López-López, Page, Lipsey and Higgins 15

Failing to account for such dependencies can bias meta-analytic results, lead to inaccurate pooled estimates, and compromise statistical inferences.Reference Pustejovsky and Tipton 5 , Reference López-López, Page, Lipsey and Higgins 15 , Reference Cheung 16 Several statistical approaches have been proposed to address effect size multiplicity when pooling correlation matrices meta-analytically, which we describe briefly in the next sections. Notably, we exclude approaches that ignore effect size multiplicity—be it the fact that multiple correlation coefficients are stored in correlation matrices or multiple correlation matrices are available per study. For an overview of these approaches, we kindly refer readers to Stolwijk et al.Reference Stolwijk, Jak, Eichelsheim and Hoeve 10

2.2 Univariate approaches

Univariate approaches pool each correlation coefficient within a correlation matrix separately across studies via a series of standard, univariate, fixed- or random-effects models.Reference Becker, Aloe, Cheung, Schmid, Stijnen and White 1 , Reference Cheung 3 , Reference Borenstein, Hedges, Higgins and Rothstein 17 For instance, taking a univariate approach, the correlation between anxiety and depression would be meta-analyzed independently of the correlations between anxiety and stress and between stress and depression. In situations where multiple correlations between the same constructs or measures are available due to the inclusion of multiple samples, univariate approaches can account for this hierarchical effect size multiplicity via multilevel random-effects models. Such models estimate the variation of correlation coefficients within and between studies and allow meta-analysts to explore possible causes of this heterogeneity.Reference Van den Noortgate, López-López, Marín-Martínez and Sánchez-Meca 18 , Reference Harrer, Cuijpers, Furukawa and Ebert 19 However, this approach neglects the inherent multivariate structure of correlation matrices, where multiple correlations share common constructs and are derived from the same sample (i.e., correlational effect size multiplicity). Ignoring this dependence among correlations within matrices leads to an underestimation of standard errors and, consequently, inflated Type I error rates.Reference Moeyaert, Ugille, Beretvas, Ferron, Bunuan and Van Den Noortgate 20 , Reference Riley 21 Hence, univariate approaches are not recommended for synthesizing correlation matrices with or without hierarchical effect size multiplicity.Reference Jak 4

2.3 Multivariate approaches

Multivariate approaches can account for dependencies among correlation coefficients within studies, which arise from measuring multiple outcomes on the same sample.Reference Becker, Aloe, Cooper, Hedges and Valentine 22 , Reference Becker 23 These models incorporate not only the sampling variances but also some information about the covariances of correlation coefficients, allowing for the estimation of a pooled correlation matrix that reflects the interrelationships between multiple outcomes.Reference Jak 4 By simultaneously leveraging all available correlations, multivariate approaches “borrow strength,” for instance, from the within-study correlations, leading to more precise estimates of mean effect sizes and more reliable assessments of between-study heterogeneity.Reference Riley 21 , Reference Jackson, White, Price, Copas and Riley 24 , Reference Jackson, White, Riley, Schmid, Stijnen and White 25 Moreover, multivariate approaches often result in joint confidence intervals and distributions and allow meta-analysts to predict an effect controlling for another, correlated effect.Reference Jackson, White, Riley, Schmid, Stijnen and White 25 However, the application of multivariate approaches in practice is often constrained by the limited availability of information about sampling covariances or correlations in primary studies, and estimation problems may occur due to the increased number of parameters and their joint distributions as compared to separate univariate meta-analyses.Reference Jackson, White, Riley, Schmid, Stijnen and White 25 , Reference Ahn, Ames and Myers 26 In the following section, we briefly describe some multivariate approaches to pooling correlation matrices meta-analytically.

2.3.1 Generalized least squares approach

BeckerReference Becker 23 , Reference Card 27 developed a multi-step generalized least squares (GLS) approach, in which multiple correlation coefficients can be synthesized while accounting for their dependencies. Specifically, in this approach, the correlation coefficients and sampling variances are extracted from the primary studies, and sampling covariances are estimated for each pair of correlation coefficients. Using the sampling covariance matrix and between-study heterogeneity estimates for each correlation coefficient to obtain weights under a random-effects assumption, a weighted average, that is, a pooled correlation matrix, is then obtained.Reference Card 27 While the GLS approach can handle the dependencies among multiple correlations within correlation matrices, it does not account for hierarchical effect size multiplicity when multiple correlations are available for the same constructs across different samples within studies. It would likely need a multilevel extension of the step in which heterogeneity variances are estimated and incorporated in the weights.

2.3.2 Multivariate, random-effects modeling approach

Given that correlation matrices within studies contain multiple correlation coefficients from the same sample, dependencies among these effect sizes occur (i.e., correlational effect size multiplicity). A multivariate, random-effects modeling (MV-REM) approach can handle these dependencies and overcome issues associated with estimating multiple, separate, univariate random-effects models, such as the risk of bias and reduced precision.Reference Cheung 3 , Reference Riley 21 Specifically, MV-REMs decompose multiple observed correlation coefficients into true effect sizes, random effects, and sampling error. Given the multivariate nature of the data, the sampling covariance matrix of a study does not only contain the sampling variances in the diagonal but also nonzero sampling covariances in the off-diagonal.Reference Cheung 3 These covariance matrices are study-specific, so that two primary studies may show different degrees of dependencies. At the population level, the dependencies among effect sizes are modeled by the structure of the random effects, that is, the structure of the heterogeneity covariance matrix. In this matrix, the between-study heterogeneity variances are stored in the diagonal, and the off-diagonal represents the assumptions on the dependencies at the population level. Several structures have been proposed (e.g., unstructured, diagonal, or compound symmetric [CS] matrices),Reference Cheung 3 , Reference Viechtbauer 28 and we will describe them in greater detail in Section 3. Overall, while the MV-REM approach handles correlational effect size multiplicity within correlation matrices, it does not consider hierarchical effect size multiplicity. Hence, taking this approach, meta-analysts are limited to meta-analytic data in which each study contributes one correlation matrix.

2.3.3 Structural equation modeling approach

To synthesize correlation matrices across studies, CheungReference Cheung 29 translated multivariate meta-analysis into the SEM framework. In this framework, the vector of correlation coefficients and their variance components are modeled as mean and covariance structures in SEM. This approach follows two stepsReference Cheung 3 , Reference Cheung and Chan 30 : In the first step, researchers obtain the sampling variance–covariance matrix for each study via confirmatory factor analyses. In the second step, the vector of correlation coefficients is pooled via an MV-REM that also incorporates the sampling covariance matrices. In addition to a pooled correlation matrix, a between-study heterogeneity covariance matrix is estimated. The SEM approach offers several advantages, including the ability to handle missing correlation coefficients in some primary studies. Specifically, given that the meta-analytic models used to synthesize correlation matrices are translated into an SEM framework, the Full-Information Maximum-Likelihood procedure is available to handle missing correlations in the primary studies’ correlation matrices.Reference Cheung 3 This way, all available information from the correlation matrices is used to estimate a pooled correlation matrix and the respective sampling covariance matrix. In the pooling stage, the data are often vectorized, and missing entries in these vectors are kept. This procedure is more efficient than listwise deletion, in which correlation matrices with missing values are deleted.Reference Jak 4 , Reference Jak and Cheung 31

Despite its utility in accounting for the dependencies among correlation coefficients within a correlation matrix (correlational effect size multiplicity), the SEM approach is less applicable when multiple correlation matrices are reported for independent samples within a study (hierarchical effect size multiplicity), because it allows each study to contribute only one correlation to each cell of the synthesized correlation matrix.Reference Wilson, Polanin and Lipsey 12 Hence, this approach is constrained by its assumption of independence between correlation matrices across studies.Reference Stolwijk, Jak, Eichelsheim and Hoeve 10

2.3.4 Wilson–Polanin–Lipsey approach

The three-level random-effects modeling approach, proposed by Wilson et al.,Reference Wilson, Polanin and Lipsey 12 addresses both correlational and hierarchical effect size multiplicity when pooling correlation matrices. This approach includes information from all available (dependent) effect sizes per study and explicitly models these dependencies in the pooled correlation matrix. Furthermore, it estimates both within- and between-study variances, allowing for the examination of heterogeneity at different levels (CheungReference Cheung 16 ). In essence, the WPL approach is based on a multivariate, multilevel, and random-effects model, which yields a pooled correlation matrix, a within-study variance estimate, and a between-study variance estimate.Reference Stolwijk, Jak, Eichelsheim and Hoeve 10 , Reference Wilson, Polanin and Lipsey 12 However, the WPL approach assumes that these heterogeneity variance estimates are the same for all correlation coefficients. This assumption may not hold in practice, especially when the correlation coefficients represent associations among diverse constructs that may vary substantially across studies. In such a situation, meta-analysts may expect different amounts of heterogeneity across correlation coefficients. Overall, the three-level random-effects approach by Wilson et al. addresses both correlational and hierarchical dependencies, but its simplifying assumption about within- and between-study heterogeneity limits its utility.

3 A multilevel, multivariate, and random-effects modeling approach

When multiple, independent samples provide correlation matrices, both correlational and hierarchical effect size multiplicity exist. One approach to account for these multiplicities is to model the dependencies among multiple correlations within a correlation matrix via a multivariate component and to model the hierarchical nesting of multiple, independent correlation matrices in primary studies via a multilevel component. This approach can be based on MLMV-REMsReference McShane and Böckenholt 32 that systematically extend the existing multivariate approaches (e.g., the WPL and MV-REM approaches).

3.1 Specification of the MLMV-REM

Suppose we have a meta-analytic dataset that comprises correlation coefficients as effect sizes and their respective sampling variances.Footnote i Let ${\widehat{r}}_{ijk}$ denote the estimate of the ith correlation coefficient in the jth correlation matrix (or, respectively, the jth sample) extracted from the kth primary study. Given this terminology, the jth correlation matrix within the kth primary study that contains I correlation coefficients can then be written as ${\widehat{\mathbf{r}}}_{jk}={\left[{\widehat{r}}_{ijk}\right]}_{i=1}^I$ .

A generic version of an MLMV-REM can be specified by distinguishing between three levels of analysis. At Level 1, the effect size estimates ${\widehat{\mathbf{r}}}_{jk}$ are decomposed into true effect size ${\boldsymbol{\unicode{x3bb}}}_{jk}$ and sampling errorReference Harrer, Cuijpers, Furukawa and Ebert 19 :

$$\begin{align*}\mathrm{Level}\;1\left(\mathrm{sampling}\kern0.17em \mathrm{error}\right)\kern-1.5pt{:}\ {\widehat{\mathbf{R}}}_{jk}={\boldsymbol{\unicode{x3bb}}}_{jk}+{\mathbf{e}}_{jk}\;\mathrm{with}\;{\mathbf{e}}_{jk}\sim MVN\left(\mathbf{0},{\mathbf{V}}_{jk}\right)\!.\end{align*}$$

Given the multivariate nature of the data (i.e., multiple correlations within a correlation matrix due to multiple outcomes or constructs of interest), the sampling variances ${v}_{ijk}$ are stored in the diagonal of the sampling covariance matrices ${\mathbf{V}}_{jk}$ , and possible covariances among the sampling correlation coefficients are stored off the diagonal. ${\mathbf{V}}_{jk}$ take the following symmetric form:

$$\begin{align*}\mathrm{Sampling}\kern0.17em \mathrm{covariance}\kern0.17em \mathrm{matrix}{:}\ {\mathbf{V}}_{jk}=\left[\begin{array}{cccc}{v}_{1 jk}& & & \\ {}{\mathit{\operatorname{cov}}}_{\left(2,1\right) jk}& {v}_{2 jk}& & \\ {}\vdots & \vdots & \ddots & \\ {}{\mathit{\operatorname{cov}}}_{\left(I,1\right) jk}& {\mathit{\operatorname{cov}}}_{\left(I,2\right) jk}& \cdots & {v}_{Ijk}\end{array}\right]\end{align*}$$

To access the sampling covariances ${\mathit{\operatorname{cov}}}_{\left(i,{i}^{\prime}\right) jk}$ for the $i$ th and ${i}^{\prime }$ th variables, several solutions have been proposed.Footnote ii Wei and HigginsReference Wei and Higgins 33 described a strategy in which the available information about the correlations among effect sizes is meta-analyzed, and the resultant, pooled correlation coefficient is then used to impute the correlations for studies or matrices that do not provide this information. Another solution is to specify values of the correlations among effect sizes, either informed by existing research or estimated correlations among constructs from the primary study data and then derive the resultant sampling covariances. However, oftentimes, this approach requires restrictive assumptions on correlations between effect sizes, such as the assumption of a constant sampling correlation.Reference Pustejovsky and Tipton 5 CheungReference Cheung 3 suggested to translate the meta-analytic models that pool the correlation matrices into the framework of SEM, which makes available an asymptotic sampling covariance matrix. Olkin and FinnReference Olkin and Finn 34 reviewed the explicit expressions for constructing the covariances between two correlation coefficients that were proposed by Olkin and SiotaniReference Olkin and Siotani 35 , who derived them based on large-sample theory. Specifically, skipping the abovementioned indices $j$ and $k$ for simplicity, for variables $A$ , $B$ , $C$ , $D$ , and $N$ representing the sample size, the asymptotic covariance between the correlation coefficients ${r}_{AB}$ and ${r}_{CD}$ isReference Olkin and Finn 34

$$\begin{align*}\mathit{\operatorname{cov}}\left({r}_{AB},{r}_{CD}\right)&=\frac{1}{N-1}\bigg[\frac{1}{2}{\rho}_{AB}{\rho}_{CD}\left({\rho}_{AC}^2+{\rho}_{AD}^2+{\rho}_{BC}^2+{\rho}_{BD}^2\right)+\left({\rho}_{AC}{\rho}_{BD}+{\rho}_{AD}{\rho}_{BC}\right)\\&\quad-\left({\rho}_{AB}{\rho}_{AC}{\rho}_{AD}+{\rho}_{AB}{\rho}_{BC}{\rho}_{BD}+{\rho}_{AC}{\rho}_{BC}{\rho}_{CD}+{\rho}_{AD}{\rho}_{BD}{\rho}_{CD}\right)\bigg].\end{align*}$$

Conversely, the asymptotic covariance between ${r}_{AB}$ and ${r}_{AC}$ —two correlation coefficients sharing variable $A$ —isReference Olkin and Finn 34

$$\begin{align*}\mathit{\operatorname{cov}}\left({r}_{AB},{r}_{AC}\right)=\frac{1}{N-1}\left[\left({\rho}_{BC}-\frac{1}{2}{\rho}_{AB}{\rho}_{AC}\right)\left(1-{\rho}_{AB}^2-{\rho}_{AC}^2\right)+\frac{1}{2}{\rho}_{AB}{\rho}_{AC}{\rho}_{BC}^2\right].\end{align*}$$

Finally, the asymptotic sampling variance of a correlation coefficient ${r}_{AB}$ reduces toReference Borenstein, Hedges, Higgins and Rothstein 17

$$\begin{align*}{v}_{AB}={\left(1-{\rho}_{AB}^2\right)}^2/\left(N-1\right).\end{align*}$$

Similar equations have also been derived for Fisher- ${Z}_r$ transformed correlations coefficients.Reference Card 27 , Reference Pustejovsky 36 Irrespective of the choice for an approach to approximate or construct the sampling covariance matrix, the Level 1 specification of the MLMV-REM is the first instance in which meta-analysts must decide on how to represent the effect size multiplicity within correlation matrices.

To further address the nesting of multiple correlation matrices (or samples) in primary studies, two additional levels of analysis are specified. Level 2 contains information about the within-study heterogeneity of correlation coefficients, and Level 3 contains information about the between-study heterogeneity of correlation coefficients.Reference Fernández-Castilla, Jamshidi, Declercq, Beretvas, Onghena and Van den Noortgate 8 , Reference Van den Noortgate, López-López, Marín-Martínez and Sánchez-Meca 18 , Reference Harrer, Cuijpers, Furukawa and Ebert 19 The decomposition of effect sizes and the respective assumptions can be specified as follows:

$$\begin{align*}\mathrm{Level}\;2\;\left(\mathrm{within}\kern0.17em \mathrm{studies}\right)\kern-1.5pt{:}\ {\boldsymbol{\unicode{x3bb}}}_{jk}={\boldsymbol{\unicode{x3b7}}}_k+{\mathbf{u}}_{jk}\mathrm{with}\;{\mathbf{u}}_{jk}\sim MVN\left(\boldsymbol{0},\boldsymbol{\Sigma} \right)\!,\end{align*}$$
$$\begin{align*}\mathrm{Level}\;3\;\left(\mathrm{between}\kern0.17em \mathrm{studies}\right)\kern-1.5pt{:}\ {\boldsymbol{\unicode{x3b7}}}_k=\boldsymbol{\unicode{x3bc}} +{\mathbf{g}}_k\;\mathrm{with}\;{\mathbf{g}}_k\sim MVN\left(\boldsymbol{0},\boldsymbol{\Omega} \right)\!.\end{align*}$$

In this specification, the vector of correlation coefficients ${\boldsymbol{\unicode{x3bb}}}_{jk}$ is decomposed into an average study-level vector of correlations ${\boldsymbol{\unicode{x3b7}}}_k$ and its matrix- or sample-specific deviations from it, ${\mathbf{u}}_{jk}$ . At Level 2, these deviations may exhibit within-study heterogeneity, which is stored in the respective heterogeneity matrix $\boldsymbol{\Sigma}$ . Accordingly, ${\boldsymbol{\unicode{x3b7}}}_k$ is then further decomposed into an overall effect size estimate $\boldsymbol{\unicode{x3bc}}$ and the study-specific deviations from it, ${\mathbf{g}}_k$ . At Level 3, these deviations may exhibit between-study heterogeneity, which is stored in the respective heterogeneity matrix $\boldsymbol{\Omega}$ . Combining all levels of analysis results in the following overall model specification:

$$\begin{align*}{\widehat{\mathbf{R}}}_{jk}=\boldsymbol{\unicode{x3bc}} +{\mathbf{g}}_k+{\mathbf{u}}_{jk}+{\mathbf{e}}_{jk}\;\mathrm{with}\kern0.24em {\mathbf{e}}_{jk}\sim MVN\left(\mathbf{0},{\mathbf{V}}_{jk}\right),{\mathbf{u}}_{jk}\sim MVN\left(\mathbf{0},\boldsymbol{\Sigma} \right),{\mathbf{g}}_k\sim MVN\left(\mathbf{0},\boldsymbol{\Omega} \right)\!.\end{align*}$$

This MLMV-REM allows meta-analysts to quantify correlation-specific heterogeneity within and between primary studies (see Model 1 in Table 1). In this context, “correlation-specific” means that the parameters are estimated for each of the correlations within a correlation matrix. For instance, if meta-analysts wish to synthesize the correlations among four variables, then each correlation matrix contains six correlations off the diagonal, which represent the associations among the six pairs of variables. In our specification of the MLMV-REM, each of these six types of correlations has their own estimates of the within- and between-study heterogeneity variances. In the next section, we describe a range of structures that the heterogeneity matrices $\boldsymbol{\Sigma}$ and $\boldsymbol{\Omega}$ may have.

Table 1 Specification of meta-analytic working models with the MLMV-REM approach

Note. The variable “Cell” is defined as a factor indicating the different types of correlations in the correlation matrix (e.g., “I2–I4” and “I11–I12”). “Correlation-specific” means that the parameters are estimated for each of the 10 correlations. “Constrained” means that the parameter is constrained to be equal across the types of correlations. “Correlation” contains the extracted Pearson correlation coefficients. “V” represents the sampling covariance matrix. “Cell” represents a nominal variable of the names of the different types of correlations. The random-effects correlations $\rho$ and $\phi$ refer to the correlations among random effects within and, respectively, between studies. These correlations were fixed to zero here. “HCS” specifies a heteroscedastic compound symmetric (CS) structure of the random effects, while “CS” specifies a CS structure.Reference Viechtbauer 28

The MLMV-REM approach can be further extended to a mixed-effects meta-regression model by adding explanatory variables, such as sample, study, publication, or measurement characteristics. In such an extension, ${\mathbf{u}}_{jk}$ and ${\mathbf{g}}_k$ may become residuals in a regression sense, depending on the level at which the explanatory variables are defined.Reference Snijders and Bosker 37 Moreover, heterogeneity could be explained specifically for each correlation coefficient at two levels of analysis (i.e., within and between studies). This entails that multiple measures of variance explanations are accessible for each correlation coefficient, once explanatory variables are added to the MLMV-REM (i.e., ${R}_{(2)i}^2$ and ${R}_{(3)i}^2$ for each of the $i=1,\dots, I$ ) correlations).Reference Cheung 3 , Reference Rights and Sterba 38

Finally, the MLMV-REM approach has the ability to pool correlation matrices with partially missing correlations using likelihood estimation. In this approach, missing correlation coefficients are typically omitted from the likelihood calculation, while all available correlations for each study are retained.Reference McShane and Böckenholt 39 , Reference Lu 40 Thus, rather than imputing unreported values or excluding entire studies, each unobserved correlation simply contributes no data to the joint likelihood, and only the observed correlations are modeled. This effectively results in pairwise deletion at the level of individual correlation coefficients. However, unlike naive pairwise methods that estimate each correlation separately, a joint multivariate model “borrows strength” across the various outcome measures and shared study-level effects, thereby leveraging the observed correlations from each study to improve the estimation of the weighted average correlations.Reference Jackson, White, Price, Copas and Riley 24 , Reference Riley, Thompson and Abrams 41

3.2 Specification of the Level 2 and Level 3 random effects

As noted earlier, heterogeneity can be modeled with the MLMV-REM both within and between studies for each type of correlation. Specifically, the symmetric heterogeneity matrices $\boldsymbol{\Sigma}$ and $\boldsymbol{\Omega}$ store the correlation-specific heterogeneity variances in the diagonal and the heterogeneity covariances in the off-diagonal. For I correlation coefficients, these matrices can take the following, often referred to as “unstructured” forms:

$$\begin{align*}\mathrm{Level}-2\;\mathrm{heterogeneity}\kern0.17em \mathrm{covariance}\kern0.17em \mathrm{matrix}{:}\ \boldsymbol{\Sigma} =\left[\begin{array}{cccc}{\tau}_1^2& & & \\ {}{\rho}_{\left(2,1\right)}{\tau}_2{\tau}_1& {\tau}_2^2& & \\ {}\vdots & \vdots & \ddots & \\ {}{\rho}_{\left(I,1\right)}{\tau}_I{\tau}_1& {\rho}_{\left(I,2\right)}{\tau}_I{\tau}_2& \cdots & {\tau}_I^2\end{array}\right],\end{align*}$$
$$\begin{align*}\mathrm{Level}-3\;\mathrm{heterogeneity}\kern0.17em \mathrm{covariance}\kern0.17em \mathrm{matrix}{:}\ \boldsymbol{\Omega} =\left[\begin{array}{cccc}{\gamma}_1^2& & & \\ {}{\phi}_{\left(2,1\right)}{\gamma}_2{\gamma}_1& {\gamma}_2^2& & \\ {}\vdots & \vdots & \ddots & \\ {}{\phi}_{\left(I,1\right)}{\gamma}_I{\gamma}_1& {\phi}_{\left(I,2\right)}{\gamma}_I{\gamma}_2& \cdots & {\gamma}_I^2\end{array}\right].\end{align*}$$

These random-effects covariance matrices contain the within-study variances ${\tau}_i^2$ , the between-study variances ${\gamma}_i^2$ for each correlation, and the covariances that are composed of the within- or between-study standard deviations, ${\tau}_i$ and ${\gamma}_i$ , and the correlations among the within- or between-level random effects of the correlation coefficients, ${\rho}_{\left(i,{i}^{\prime}\right)}$ and ${\phi}_{\left(i,{i}^{\prime}\right)}$ , for $i,{i}^{\prime}=1,\dots, I$ .

This form of $\boldsymbol{\Sigma}$ and $\boldsymbol{\Omega}$ is “unstructured” and allows meta-analysts to include the covariances among random effects that are specific to each pair $\left(i,{i}^{\prime}\right)$ of correlation coefficients. However, although the specification of this form has a large degree of flexibility, meta-analysts may not be able to estimate all parameters within the heterogeneity covariance matrices. For instance, the likely small number of available correlations and primary studies could lead to unstable variances and small statistical power of the respective estimates. Moreover, information about the correlations among the random effects within and between studies may be lacking. Hence, meta-analysts often must simplify the structures of $\boldsymbol{\Sigma}$ and $\boldsymbol{\Omega}$ . We would like to highlight two approaches to simplifying the heterogeneity matrices. The first approach concerns the correlations among the random effects, ${\rho}_{\left(i,{i}^{\prime}\right)}$ and ${\phi}_{\left(i,{i}^{\prime}\right)}$ . Like the specification of the sampling covariance matrices ${\mathbf{V}}_{jk}$ , meta-analysts could fix ${\rho}_{\left(i,{i}^{\prime}\right)}$ and ${\phi}_{\left(i,{i}^{\prime}\right)}$ to some nonzero constant values, $\rho$ and $\phi$ , that are the same for each random effect or approximate them in some way. The structure with constant random-effects correlations is referred to as the “heteroscedastic compound symmetry (HCS)” structure. If $\rho$ or $\phi$ are fixed to zero, then the matrices $\boldsymbol{\Sigma}$ and $\boldsymbol{\Omega}$ have a diagonal structure—a special case of the HCS structure in which meta-analysts can model only the within- and between-study heterogeneity variances and assume that random effects at Levels 2 and 3 are independent. This diagonal structure is often used in the first stage of MASEM, in which multiple correlation matrices are pooled via an MV-REM.Reference Cheung 3 The heterogeneity matrices with an HCS or diagonal structure are as follows:

$$\begin{align*}\mathrm{HCS}\;\mathrm{structures}{:}\ \boldsymbol{\Sigma} =\left[\begin{array}{cccc}{\tau}_1^2& & & \\ {}{\rho \tau}_2{\tau}_1& {\tau}_2^2& & \\ {}\vdots & \vdots & \ddots & \\ {}{\rho \tau}_I{\tau}_1& {\rho \tau}_I{\tau}_2& \cdots & {\tau}_I^2\end{array}\right],\boldsymbol{\Omega} =\left[\begin{array}{cccc}{\gamma}_1^2& & & \\ {}\phi {\gamma}_2{\gamma}_1& {\gamma}_2^2& & \\ {}\vdots & \vdots & \ddots & \\ {}\phi {\gamma}_I{\gamma}_1& \phi {\gamma}_I{\gamma}_2& \cdots & {\gamma}_I^2\end{array}\right],\end{align*}$$
$$\begin{align*}\mathrm{Diagonal}\kern0.17em \mathrm{structures}{:}\ \boldsymbol{\Sigma} =\left[\begin{array}{cccc}{\tau}_1^2& & & \\ {}0& {\tau}_2^2& & \\ {}\vdots & \vdots & \ddots & \\ {}0& 0& \cdots & {\tau}_I^2\end{array}\right],\boldsymbol{\Omega} =\left[\begin{array}{cccc}{\gamma}_1^2& & & \\ {}0& {\gamma}_2^2& & \\ {}\vdots & \vdots & \ddots & \\ {}0& 0& \cdots & {\gamma}_I^2\end{array}\right]\;.\end{align*}$$

Notably, these structures contain heterogeneity variances that are specific to the random effects at Levels 2 and 3 and thus to the correlation coefficients. In other words, for each correlation coefficient (i.e., effect size), within- and between-study variance components are estimated.

A second approach to simplifying $\boldsymbol{\Sigma}$ and $\boldsymbol{\Omega}$ is to constrain these variance components to be equal across random effects. Imposing this constraint to an HCS structure results in a so-called “CS” structure. In the special case where $\rho$ or $\phi$ are fixed to zero, the structure is that of a “scaled identity” structure.Reference Viechtbauer 28 Notably, the term “scaled identity structure” is based on the fact that the two heterogeneity matrices have the forms $\boldsymbol{\Sigma} ={\tau}^2{\boldsymbol{I}}_I$ and $\boldsymbol{\Omega} ={\gamma}^2{\boldsymbol{I}}_I$ , where ${\boldsymbol{I}}_I$ represents an $I\times I$ identity matrix:

$$\begin{align*}\mathrm{CS}\;\mathrm{structures}{:}\ \boldsymbol{\Sigma} =\left[\begin{array}{cccc}{\tau}^2& & & \\ {}{\rho \tau}^2& {\tau}^2& & \\ {}\vdots & \vdots & \ddots & \\ {}{\rho \tau}^2& {\rho \tau}^2& \cdots & {\tau}^2\end{array}\right],\boldsymbol{\Omega} =\left[\begin{array}{cccc}{\gamma}^2& & & \\ {}\phi {\gamma}^2& {\gamma}^2& & \\ {}\vdots & \vdots & \ddots & \\ {}\phi {\gamma}^2& \phi {\gamma}^2& \cdots & {\gamma}^2\end{array}\right],\end{align*}$$
$$\begin{align*}\mathrm{Scaled}\kern0.17em \mathrm{identity}\kern0.17em \mathrm{structures}{:}\ \boldsymbol{\Sigma} =\left[\begin{array}{cccc}{\tau}^2& & & \\ {}0& {\tau}^2& & \\ {}\vdots & \vdots & \ddots & \\ {}0& 0& \cdots & {\tau}^2\end{array}\right],\boldsymbol{\Omega} =\left[\begin{array}{cccc}{\gamma}^2& & & \\ {}0& {\gamma}^2& & \\ {}\vdots & \vdots & \ddots & \\ {}0& 0& \cdots & {\gamma}^2\end{array}\right].\end{align*}$$

If meta-analysts choose one of the structures that assume fixed correlations among random effects, $\rho$ or $\phi$ , in our view, they should examine the extent to which the meta-analytic findings and inferences are sensitive to the choice of these fixed values.Reference Hedges, Tipton and Johnson 9 , Reference Harrer, Cuijpers, Furukawa and Ebert 19

3.3 Decision scheme and working models

3.3.1 Analytical decisions

To address hierarchical effect size multiplicity when pooling correlation matrices, meta-analysts must make several analytic decisions. Figure 1 provides an overview of these decisions and their possible outcomes. First, when meta-analysts have extracted the correlation matrices and the respective correlation coefficients from the primary study data, they must decide on the structure of the sampling covariance matrices ${\mathbf{V}}_{jk}$ . Specifically, assuming that correlations within correlation matrices/samples are independent results in diagonal sampling covariance matrices that essentially only contain the sampling variances of the correlation coefficients. Alternatively, meta-analysts could assume some dependence among the sampling errors and then specify ${\mathbf{V}}_{jk}$ with sampling covariances off the diagonals. In scenarios where multiple measures of the same sample have been taken over time or multiple, correlated constructs were assessed for each sample, a dependency structure with some correlation among sampling errors is conceptually reasonable and preferable over an independence structure.Reference Cheung 3 , Reference Pustejovsky and Tipton 5 , Reference Harrer, Cuijpers, Furukawa and Ebert 19 The extracted correlation coefficients ${\widehat{\mathbf{R}}}_{jk}$ and sampling covariance matrices can then be submitted to meta-analytic modeling.

Figure 1 Decision scheme for selecting a working model.

Second, if hierarchical effect size multiplicity is present in the meta-analytic data, meta-analysts must decide if they wish to account for it. In case they wish to ignore effect size multiplicity—for instance, due to close-to-zero within-study heterogeneity in the correlation coefficients—they may choose an MV-REM to pool the correlation matrices.Reference Becker, Aloe, Cheung, Schmid, Stijnen and White 1 , Reference Stolwijk, Jak, Eichelsheim and Hoeve 10 , Reference Sheng, Kong, Cortina and Hou 13 To model hierarchical effect size multiplicity, meta-analysts may choose a working model within the MLMV-REM approach (see the following section).

Third, meta-analysts must decide on the structures of the within- and between-study heterogeneity matrices. Specifically, do they allow for correlation-specific within- and between-study variances in correlation coefficients, and which assumption do they make on the random-effects correlations within and between studies (i.e., independence assumption with $\rho =\phi =0$ or dependence assumptions with some nonzero values of $\rho$ and $\phi$ )? While the former decision determines which of the four working models is selected, the latter decision must be made for any model that has been selected.

3.3.2 Selection of a working model

Table 1 provides an overview of four working models within the MLMV-REM approach, which estimate weighted average effect sizes and the within- and between-study variance components and assume constant correlations among the Level 2 and Level 3 random effects. Model 1 allows the heterogeneity estimates at Levels 2 and 3 to be specific to each correlation coefficient. Hence, $\boldsymbol{\Sigma}$ and $\boldsymbol{\Omega}$ may simplify to an HCS or a diagonal structure. Model 2 estimates correlation-specific within-study heterogeneity variances but constrains the between-study heterogeneity variances to be equal across correlation coefficients. Hence, $\boldsymbol{\Sigma}$ follows an HCS or a diagonal structure, while $\boldsymbol{\Omega}$ follows a CS or scaled identity structure. This model assumes that the correlation coefficients have the same amount of heterogeneity between (yet not within) studies. Conversely, Model 3 estimates correlation-specific between-study heterogeneity and constrains the within-study heterogeneity to be equal across correlation coefficients. Finally, Model 4 imposes the equality constraints on both the within- and between-study heterogeneity variances. In this model, $\boldsymbol{\Sigma}$ and $\boldsymbol{\Omega}$ both have a CS or scaled identity structure. Notably, this model is similar to the meta-analytic model underlying the WPL approach.Reference Wilson, Polanin and Lipsey 12

This selection of working models shows that meta-analysts have several options to specify and simplify similar or different Level 2 and Level 3 variance structures (see Table 1). We chose Models 1–4 as working models because, in our view, they are relevant for meta-analytic modeling practice for several reasons. First, these models allow meta-analysts to estimate correlation-specific heterogeneity variances. The information about the extent to which specific correlations within correlation matrices vary within and between studies identifies the levels at which heterogeneity is located and could be explained by potential moderators.

Second, as noted earlier, the oftentimes lacking or limited information about the correlations among the Level 2 and Level 3 random effects in a meta-analysis may not allow meta-analysts to estimate these correlations for each correlation coefficient separately. Hence, fixing these random-effects correlations to some constant value represents a pragmatic solution to still accounting for random-effects dependencies despite the lack of information about their extent.Reference Pustejovsky and Tipton 5

Third, the constant random-effects correlations $\rho$ and $\phi$ can also be fixed to zero, assuming the independence of random effects at Levels 2 and 3. This assumption is sometimes made when meta-analysts are only interested in estimating the heterogeneity variances.Reference Cheung 3 , Reference Jak 4 However, in this situation, the dependencies among the Level-2 and Level-3 random effects are not accounted for and may introduce some bias to the meta-analytic estimates and their standard errors.

Fourth, in Working Models 2–4, the within- and/or between-study heterogeneity variances are constrained to be equal across correlation coefficients. While these models make a strong and oftentimes unrealistic assumption about the heterogeneity in the meta-analytic data, they are still useful for meta-analytic practices. Specifically, comparing a model with such equality constraints to Working Model 1 allows meta-analysts to test hypotheses on the similarity or differences in heterogeneity variance estimates across the correlation coefficients.Reference Cheung 3 The constraints in Models 2–4 offer meta-analysts ways to simplify the meta-analytic models and thus strive for model parsimony.

Fifth, Models 1–4 represent several substantive assumptions that may represent meta-analysts’ theories and hypotheses on the heterogeneity in the meta-analytic data. Model 1 represents the most flexible model among the working models and is useful in situations where the correlation-specific effect sizes and heterogeneity variances are in meta-analysts’ key interest. Moreover, Model 4 may also be useful to model meta-analytic data that comprised highly heterogeneous samples within primary studies of different characteristics. Model 2 assumes that correlation coefficients can vary within studies to a different extent and between studies to the same extent. This model could be useful for meta-analytic data that comprised highly heterogeneous samples within primary studies that follow the same study protocol (e.g., multilab studies). Conversely, Model 3 might be useful for modeling meta-analytic data with highly homogeneous samples within studies but with substantial differences in study characteristics (e.g., different sampling protocols). Model 4 may be useful in situations where (a) meta-analysts wish to account for heterogeneity both within and between studies yet do not have a large-enough meta-analytic database to model correlation-specific heterogeneity variances; (b) within- and between-study heterogeneity variances are homogeneous (e.g., consistently close to zero across all correlation coefficients); or (c) meta-analysts wish to test hypotheses on the equality of the heterogeneity variances.

In addition to these substantive decisions, meta-analysts may also rely on fit information and the results of model comparisons. For instance, Working Models 2 and 3 are both nested in Model 1, so that, in addition to inspecting information criteria, likelihood-ratio testing could provide information about which model may be preferred. In this sense, the selection of a working model is both a substantive and statistical decision.

4 Illustrative example

To support meta-analysts with translating their decisions on the dependencies among sampling errors, the modeling of within- and between-study heterogeneity variances, and the choices for the Level 2 and Level 3 correlations among random effects, we illustrate the implementation of Models 1–4 in the R software with an example dataset. Specifically, we showcase the model estimation using the R package “metafor”Reference Viechtbauer 28 and the generation of cluster-robust standard errors using the R package “clubSandwich.” 42 Please find the full data in Supplementary Material S1 and the detailed R code with the respective output in Supplementary Materials S2 and S3.

In our illustrative example, we chose a subset of an openly available dataset that was published by Schroeders et al.Reference Schroeders, Kubera and Gnambs 43 This meta-analytic dataset contains the correlations among the items of the Toronto Alexithymia Scale (TAS-20) and/or the statistical parameters to derive them (e.g., standardized factor loadings and factor correlations). The goal of the original meta-analysis was to examine the factor structure underlying the scale, generalize reliability evidence, and ultimately craft a validity argument. While this goal may be specific to the study of assessments, our example can also be generalized to other situations and goals. For instance, the selected variables may not necessarily represent item responses, but other quantitative variables such as scores derived from other scales. Note that our illustrative example focuses on the meta-analytic pooling of the correlation matrices via the MLMV-REM approach, yet not the potential follow-up steps using the pooled correlation matrix. These steps may include conducting factor analyses, examining the psychometric network structure underlying the scale, or testing specific hypotheses on the relations among the variables.

4.1 Description of the meta-analytic data

For illustrative purposes, we chose the data focusing on participants’ self-reported difficulties with describing feelings. This dimension was measured by the following five items of the TAS-20 (see Schroeders et al.Reference Schroeders, Kubera and Gnambs 43 ):

  • Item 2 (I2): It is difficult for me to find the right words for my feelings.

  • Item 4 (I4): I am able to describe my feelings easily. (reverse-coded)

  • Item 11 (I11): I find it hard to describe how I feel about people.

  • Item 12 (I12): People tell me to describe my feelings more.

  • Item 17 (I17): It is difficult for me to reveal my innermost feeling, even to close friends.

Given the focus on these five items, the correlation matrices we intended to pool meta-analytically were 5  $\times$  5 matrices with 10 correlation coefficients each. The data were accessed from a csv file, stored in an R object named tas20, and then supplemented by an effect size identifier ESID.

Output:

To facilitate the meta-analytic modeling and to indicate the different analytical levels, we created identifiers of the effect sizes (ESID), the samples (SampleID), the primary studies (StudyID), and the type of correlation (Cell; e.g., “I2–I4” indicates the correlation between Items 2 and 4). The correlation coefficients were stored in the variable Correlation, and the names of the corresponding variables are indicated by Var1 and Var2. The dataset contains 880 correlation coefficients that were extracted from 88 samples in 62 primary studies, totaling 69,722 participants. On average, the samples comprised 792 participants (M = 792.3, SD = 1,508.8, Mdn = 327, Min = 99, Max = 1,2706), and each study included 1.4 correlation matrices (SD = 0.6, Mdn = 1, Min = 1, Max = 4). Overall, 22 primary studies contained more than two correlation matrices (i.e., samples).

4.2 Constructing sampling covariance matrices

If meta-analysts assume independent sampling errors, then they may directly use the sampling variances of the correlation coefficients as input for the meta-analytic working model. If meta-analysts decide to account for the dependencies among sampling errors within correlation matrices for a specific sample, they may construct sampling covariance matrices using the explicit expressions proposed by Olkin and Siotani.Reference Olkin and Siotani 35 There are several ways to generate the sampling matrices. For instance, the R package “metafor” contains the rcalc() function, which creates a list of sampling covariance matrices (V) for each sample-specific correlation matrix.

This function takes the variable containing the correlation coefficients (Correlation), the names of the respective variables (Var1, Var2), the clustering variable (SampleID), the sample sizes (ni = N), and the name of the dataset (data = tas20) as arguments. The first part specifies that multiple correlations are nested in sample-specific correlation matrices, because each independent sample provided one correlation matrix, and the last part specifies that sampling variances and covariances are generated using the raw correlation coefficients. The resultant object V contains the full sampling covariance matrix with a block design to indicate the correlation matrices.

## Sampling covariance matrix of the first sample

blsplit(V, tas20$SampleID)$`1`

Output:

For the R code using Fisher’s- ${Z}_r$ transformation (via the additional option rtoz = TRUE), please see Supplementary Material S3.

Another option to construct sampling covariance matrices is to use the asyCov() function in the R package “metaSEM.”Reference Cheung 44 This function has similar utilities to the above-described function. Nonetheless, it contains some additional features, such as the possibility to also construct sampling correlation matrices (via the cor.analysis option).

V <- asyCov(cmatnew, dat$n, cor.analysis = FALSE)

Several coding steps are leading up to using this function, such as generating a list of correlation matrices from the study samples (cmatnew), along with their sample sizes (n). Please find the details of these steps in Supplementary Material S2. Notice that, from version 1.2.6 of the “metaSEM” package, the asyCov() function no longer implements Cheung and Chan’sReference Cheung and Chan 45 SEM approach that rests on a multivariate normality assumption but the equations by Olkin and Finn.Reference Olkin and Finn 34 In the following analyses, we used the V matrix generated from the rcalc() function.

4.3 Approaches to pooling correlation matrices

Once the correlation matrices and sampling (co-)variance matrices have been extracted or generated, meta-analysts can employ the working models we proposed to estimate a pooled correlation matrix and heterogeneity (co-)variances. In this section, we describe in detail the implementation of Working Model 1 and explain briefly the implementation of Models 2–4 in R.

4.3.1 Working Model 1

Given the multivariate nature of the data (i.e., multiple correlations per sample that are based on multiple, measured variables in the primary studies) and the hierarchical effect size multiplicity (i.e., multiple correlation matrices nested in primary studies), we used the rma.mv() function in “metafor” to estimate Model 1. Figure 2 shows an example code with explanations of its elements. Specifically, this function contains the specification of the correlation coefficients (yi), the sampling covariance matrices (V), and the dataset in which these elements are stored (data). Next, the random effects within (|ESID) and between studies (|StudyID) are specified as a list. Given that Model 1 assumes correlation-specific estimates of these random effects, the option ~factor(Cell) is added to this list. Note that the sampling covariances were generated in the previous step using SampleID as a clustering identifier to account for the nesting of multiple correlation coefficients in the samples’ correlation matrices. In the pooling step, the sample specificity of the correlation coefficients is represented by ESID (i.e., each sample contributes only one correlation coefficient within a correlation matrix), and the hierarchical nesting of multiple correlation matrices in primary studies is represented by StudyID.

Figure 2 Elements of the analytic code to specify Working Model 1 in the R package “metafor”.

Suppose that meta-analysts do have some evidence that the Level 2 and Level 3 random effects can be considered independent. Specifying an HCS structure within and between studies (struc = c(“HCS,”"HCS”)), they may assume a correlation of zero within and zero between studies (rho = 0 and, respectively, phi = 0). Next, the estimation procedure is set; in our example, we chose restricted maximum likelihood estimation (method = “REML”). Finally, we specified the variable indicating the type of correlation as a moderator and turned off the intercept of this meta-regression part (mods = ~factor(Cell)-1). This way, we get estimates of each pooled correlation coefficient directly. Labelling the rma.mv() object “tas20.mlmvrem1” and using the summary() function, we obtained the following output of this model estimation:

First, the output contains information about the log-likelihood, deviance, and information criteria, which can be used for further model comparisons. Second, the output displays the within-study variance estimates (tau^2.1-tau^2.10) and the between-study variance estimates (gamma^2.1-gamma^2.10) for each of the 10 correlation coefficients. These estimates exhibited a considerable range, with some variances being close to zero and others as high as 0.0088 (tau^2.7). Third, the Q statistics are shown, indicating, for instance, that statistically significant (residual) heterogeneity existed in the data ( ${Q}_E$ [870] = 6,986.58, p < .01). Fourth, the estimates of the weighted average correlation coefficients, their standard errors, z-statistics, p-values, and the respective confidence intervals are shown (section Model results). Notably, all pooled coefficients were positive and were statistically significantly different from zero.

Table 2 Weighted average effect sizes and variance components in Working Models 1–4 with independent random effects (with $\rho = \phi=0$ )

Note: $\overline{r}$  = Weighted average Pearson correlation coefficient with cluster-robust standard error and 95% confidence interval. “I” refers to the item in the TAS-20. Correlations are labeled using the item numbers (e.g., “I11–I12” refers to the correlation between Items 11 and 12). AIC = Akaike’s Information Criterion; BIC = Bayesian Information Criterion; AICc = Corrected AIC.Reference Viechtbauer 28 All models assume independent random effects (i.e., $\rho =\phi =0$ ) and block-diagonal sampling covariance matrices.

If meta-analysts wish to further obtain cluster-robust standard errors of the pooled correlation coefficient estimates (with primary studies as clusters, indicated by StudyID), they may use the convenience function robust() in the R package “metafor” that calls the “clubSandwich” package:

Selected output:

Table 2 summarizes the key estimates of Model 1. However, the previous output does not contain any information about the preference of Model 1 over alternative models. Hence, in the next section, we describe the estimation of Models 2–4 and perform model comparisons to select an appropriate working model for the meta-analytic pooling of the correlation matrices.

4.3.2 Working Models 2–4

The specification of Working Models 2–4 is like that of Model 1. While Model 1 allowed for correlation-specific estimates of the heterogeneity variances, Models 2–4 restricted these estimates to be the same across correlations. This is reflected in the specification of the structures of Level 2 and Level 3 random effects. Table 1 shows sample codes to estimate Models 2–4 in “metafor.” For instance, in Model 2, the random-effects structures are specified as struc = c(“HCS,”"CS”), and the respective output contains the following variance estimates:

Notably, the within-study variance estimates are correlation-specific (tau^2.1-tau^2.10), and only one heterogeneity variance is estimated at the between level (gamma^2). Similarly, Model 3 can be specified using the option struc = c(“CS,”"HCS”) to implement the assumption of correlation-specific heterogeneity variances between studies but one uniform variance estimate within studies. Finally, the structure option in Model 4 specifies two compound structures struc = c(“CS,”"CS”). Please find the output of the estimation of Models 3 and 4 in Supplementary Material S2. Table 2 shows the parameters of Models 2–4.

4.3.3 Model selection

To select a working model, we compared the information criteria and performed likelihood-ratio tests across Models 1–4. Table 2 contains the respective values of the AIC, BIC, and a corrected AIC. Consistently across these information criteria, Model 2 showed the lowest values and thus the best fit. Moreover, a likelihood-ratio test between Models 1 and 2 suggested that these models did not differ statistically significantly in their fit, ${\unicode{x3c7}}^2$ (9) = 7.60, p = .57. We performed this test using the anova() function:

Output:

Further likelihood-ratio tests suggested the preference of Model 2 over Models 3 and 4. This preference of Model 2 indicated that the between-level variance estimates did not differ significantly and can thus be considered statistically equal across correlation coefficients. Hence, a meta-analyst may select Model 2 as her working model to pool the correlation matrices meta-analytically and conduct subsequent moderator analyses. Notice that our specifications of Models 1–4 assumed that $\rho =\phi =0$ .

4.4 Moderator analyses

As mentioned earlier, Models 1–4 can be extended to mixed-effects meta-regression models to explore moderator effects and examine the extent to which heterogeneity can be explained by the characteristics of the primary studies, samples, or measures. In this section, we illustrate this possible extension by a categorical (i.e., type of the sample, coded as 0 = non-clinical and 1 = clinical; ClinicalBinary) and a continuous moderator (i.e., the proportion of female participants in the primary study; PropFemale).

4.4.1 Categorical moderator

There are several ways of exploring the possible moderating effect of a categorical variable using the MLMVREM approach. One way is to extend the model specification of Model 2 in “metafor” by specifying a crossed effect between the moderator and the variable that indicates the type of correlation (i.e., factor(Cell)). This specification still contains the direct effect of factor(Cell), which serves as the intercepts, that is, the expected weighted average correlations for each type of correlation when the sample is nonclinical. Hence, the model specification no longer contains an overall intercept (−1).

Selected output of the model summary:

The estimates of the model parameters indicate statistically significant and negative moderator effects of the type of the sample on three correlation coefficients: I2–I4 (B = −0.084, SE = 0.032, p = .010), I4–I11 (B = −0.075, SE = 0.024, p = .002), and I4–I17 (B = −0.064, SE = 0.027, p = .017). Hence, these correlations tended to be smaller for clinical samples than for nonclinical samples. The cluster-robust standard errors supported this result. Next, we computed the variance explanations for each level of analysis and some selected correlations.

About 6.4% of the between-study variation in effect sizes was explained by the type of sample, and about 5.6% (6.7%, 0.5%) of the within-study variation in the correlations I2–I4 (I4–11, I4–I17) was explained.

Another way of exploring the effect sizes across the moderator’s categories is to estimate the subgroup-specific effect sizes directly. To achieve this, the moderator option in the above-described specification is changed into

mods =~ factor(Cell):factor(ClinicalBinary)-1.

Consequently, meta-analysts obtain the effect size estimates of each correlation for studies with nonclinical (code 0) and clinical samples (code 1). However, this way, the similarities and differences in effects are not tested statistically.

Selected output:

4.4.2 Continuous moderator

Introducing continuous moderators works in the same way as adding categorical moderators. However, given that continuous moderators do not have discrete categories, subgroup-specific estimates of the effect sizes are not accessible. In this example, we chose the proportion of female study participants as a possible moderator without any variance-stabilizing transformation for better interpretability and to avoid issues related to back-transformations.Reference Lin and Xu 46 , Reference Schwarzer, Chemaitelly, Abu-Raddad and Rücker 47 The model specification code is similar to that of the categorical example and contains a modified mods option (model tas20.gender in Supplementary Material S2):

mods =~ factor(Cell)1 + factor(Cell):PropFemale.

The results indicated that the correlation I11–I17 was moderated by the proportion of female study participants, B = −0.001, SE = 0.001, p = .008. This effect was supported by the cluster-robust standard errors. Please find the detailed output in Supplementary Material S2.

4.5 Sensitivity analyses

As noted earlier, some decisions on model parameters in the MLMV-REM may be informed by prior research or meta-analysts’ substantive expertise. To examine the sensitivity of the meta-analytic findings, we first explored how different a-priori fixed values of $\rho$ (0.0, 0.5, 1.0) and $\phi$ (0.0, 0.5, 1.0) affected (a) model selection and (b) key parameter estimates of Model 1, such as the weighted average correlation coefficients or heterogeneity variances. Second, we conducted the analyses using Fisher’s- ${Z}_r$ transformed scores.

For the raw correlations, Models 1 and 2 were suitable meta-analytic working models to pool the correlation matrices across the conditions. When $\rho$ and $\phi$ were small, Model 2 was likely to be preferred over Model 1. The relation between $\rho$ (or, respectively, $\phi$ ) and the within-sample variance estimates was mostly V-shaped, and the largest values of the respective variance estimates occurred for $\rho =\phi =1$ . The between-sample variance estimates showed the same pattern. Finally, pooled correlation coefficients increased with higher values of $\rho$ and $\phi$ . For Fisher’s- ${Z}_r$ transformed scores and given values of $\rho$ and $\phi$ , pooled correlation coefficients, within- and between-sample variance estimates were slightly larger than those obtained from the raw correlations.

The detailed results of these sensitivity analyses are shown in Supplementary Materials S2 and S3. Overall, the pooled correlations and heterogeneity estimates were sensitive to the choices of $\rho$ and $\phi$ , and simulation studies are needed to examine this sensitivity systematically.

5 Discussion and recommendations

In this tutorial paper, we have described an MLMV-REM approach to addressing hierarchical effect size multiplicity when pooling correlation matrices across primary studies. This approach can be specified via several working models under various assumptions on the correlation-specific within- and between-study variances. We illustrated the implementation of four working models using the R package “metafor” and the respective cluster-robust standard errors using the R package “clubSandwich.” This illustration contained a description of the working models’ assumptions, the R code to specify and estimate them, and the resultant R output.

As noted earlier, the MLMV-REM approach allows meta-analysts to model within- and between-study heterogeneity, either specific for each correlation coefficient or under the assumption of equal heterogeneity across correlations at the respective levels of analysis. To the best of our knowledge, this approach is the first to make explicit these heterogeneity variances—existing approaches either provide one heterogeneity variance estimate for all correlation coefficients at the within and between levelReference Wilson, Polanin and Lipsey 12 or correlation-specific heterogeneity variance estimates at a single level of analysis.Reference Cheung 3 We therefore consider the MLMV-REM approach an extension and integration of the existing approaches. This approach not only provides meta-analysts with information about associations among multiple constructs, concepts, or variables and their heterogeneity but also has the potential to advance meta-analytic SEM, in which meta-analysts can test their theories and hypotheses about these associations.Reference Stolwijk, Jak, Eichelsheim and Hoeve 10

The working models within the MLMV-REM approach represented several assumptions meta-analysts may have on the degree of sampling error dependence and the structure of the within- and between-level random effects. This not only allows meta-analysts to test hypotheses on these structures (e.g., equal vs. correlation-specific heterogeneity variances) but also enables them to select a model for the pooling of correlation matrices that represents the structure and nature of their meta-analytic data. We recommend selecting a meta-analytic working model by estimating and comparing models with several assumptions, evaluating model fit, and considering substantive aspects, such as theories or existing evidence from other studies. Moreover, we encourage meta-analysts to also consider the feasibility of working models with many correlation-specific parameter estimates. Sometimes, too few studies and correlation matrices are available to reliably estimate such models with sufficient statistical power.Reference Cheung 3 It remains for future research to examine systematically the sample size requirements of the working models and identify situations in which the respective fixed and random effects can be estimated reliably.

The MLMV-REM approach has several limitations and leaves some open questions that point to future research directions. First, while this approach offers a flexible specification of model assumptions on the structure of random effects, model selection becomes not only an opportunity but a necessity. Oftentimes, meta-analytic data contain too few studies to estimate all possible variance components of correlations within a correlation matrix.Reference Cheung 3 In this situation, meta-analysts would need to strive for model parsimony and select a model that may not capture all correlation-specific heterogeneity variances within and between studies, but some. Modern methods such as Bayesian meta-analysis may address the challenges of few available studies.

Nonetheless, we currently see the lack of guidance concerning the required number of studies and effect sizes when pooling correlation matrices meta-analytically. Some prior research has pointed to the need for sufficiently large numbers of studies when specifying complex random effects. For instance, Vembye et al.Reference Vembye, Pustejovsky and Pigott 48 showcased the associations between the required number of studies and effect sizes, how many model parameters are estimated, how large their estimates are, the magnitude of the within- and between-study heterogeneity, and the strengths of the correlational dependencies. These authors have developed approximations and tools to estimate statistical power and explore sample size requirements for random-effects models with correlational and hierarchical effect size multiplicity.Reference Vembye, Pustejovsky and Pigott 49 CheungReference Cheung 3 pointed out too few studies may not result in trustworthy, pooled correlation matrices, especially if many variables and thus correlations are involved. If, for instance, the number of estimated parameters in the pooling stage exceeds the number of studies, non-convergence or non-admissible solutions may occur. In our illustrative example, we focused on five variables and thus 10 correlation coefficients per correlation matrix. In Model 1, 10 pooled correlations are estimated, 10 within-study variances, and another 10 between-study variances, totaling 30 model parameters. Although the data obtained from the 88 samples may provide enough statistical power to estimate the pooled correlations, they may not provide enough statistical power to detect potentially small heterogeneity variances for all correlation coefficients. Despite these speculations, we call for simulation studies that shed more light on the meta-analytic sample size requirements and the factors associated with them.

Second, the current implementation of the MLMV-REM approach synthesizes correlation matrices meta-analytically and estimates the respective heterogeneity (co-)variances. Hence, meta-analysts who are interested in testing hypotheses and theories about the relations among variables would then submit these elements to SEM or psychometric network modeling.Reference Cheung and Chan 30 , Reference Cheung and Chan 50 This two-stage approach addresses hierarchical effect size multiplicity at Stage 1. We encourage meta-analysts to develop and explore alternative approaches that may integrate these two analytic stages into one stage. Such an integration could improve estimation efficiency and reduce possible bias in the variance estimates.Reference Jak and Cheung 51

Third, the MLMV-REM approach requires substantive decisions on the model specification, such as the structure of random effects or the sampling covariance matrix $\boldsymbol{V}$ and the choices of the correlations between sampling errors in the matrix $\boldsymbol{V}$ and the correlations between random effects in the matrices $\boldsymbol{\Sigma}$ and $\boldsymbol{\Omega}$ . These decisions also include the choices for an approach to construct sampling covariance matrices and the within- and between-level random effects. Oftentimes, researchers do not have specific hypotheses or prior evidence on these correlations, which leaves them to guess these values. In future model specifications, however, it may be possible to estimate such correlations.Reference Pustejovsky and Tipton 5 , Reference Viechtbauer 28 , Reference McShane and Böckenholt 32 Overall, we recommend backing the analytical decisions meta-analysts have to make by conducting sensitivity analyses to examine the impact of these decisions on meta-analytic estimates.

Fourth, the MLMV-REM approach employs maximum-likelihood estimation to pool correlation matrices with partially missing data. However, it typically assumes that any unreported correlations are missing at random or missing completely at random.Reference Pigott, Cooper and Hedges 52 If, in reality, certain correlations are systematically omitted (i.e., missing not at random [MNAR]), the resulting pooled estimates can become biased. Moreover, when information on the within-sample correlation structure is limited or absent, the “borrowing strength” across studies is less effective, leading to larger standard errors for pooled estimates.Reference Riley 21 Although alternative methods for handling missing correlations through multiple imputation have been proposed,Reference Lu 40 , Reference Pigott, Cooper and Hedges 52 further research is needed to assess their performance when estimating MLMV-REM models in the presence of the MNAR mechanism.

Fifth, effect size multiplicity may be hierarchical, correlational, or both.Reference Pustejovsky and Tipton 5 In our view, the MLMV-REM we have proposed is best suited for meta-analytic datasets with hierarchical multiplicity, in which multiple independent correlation matrices are nested in primary studies. In such a scenario, the variance components have a clear interpretation as within- and between-study heterogeneity variances of the correlation coefficients. In scenarios in which multiple correlation matrices are available for the same sample within a study, our proposed approach may not be best-suited and may only approximate the true weighted average correlation coefficients and their heterogeneity. Specifically, the MLMV-REM approach we proposed assumes zero correlation between correlation matrices within studies. Current methodological developments in the meta-analytic pooling of univariate effect sizes, such as the working models proposed by Pustejovsky and Tipton,Reference Pustejovsky and Tipton 5 and meta-analytic models describing longitudinal dataReference Viechtbauer 28 could serve as starting points to develop an approach that handles meta-analytic data with correlational multiplicity.

In our view, the MLMV-REM approach represents one possible approach to address hierarchical effect size multiplicity when pooling correlation matrices meta-analytically, and alternative approaches may be developed in the future. This approach will have to stand trial against other approaches, and, in our view, investigating its accuracy, efficiency, and statistical power is a natural next step toward evaluating its performance.

Supplementary material

To view supplementary material for this article, please visit http://doi.org/10.1017/rsm.2025.10027.

Acknowledgements

We are grateful for the work by Schroeders, Kubina, and Gnambs,Reference Schroeders, Kubera and Gnambs 43 who made their meta-analytic dataset on the factor structure of the TAS-20 openly available.

Author contributions

Conceptualization: R.S., D.G.C.; Formal analysis: R.S.; Investigation: R.S., D.G.C.; Methodology: R.S.; Software: R.S.; Visualization: R.S.; Writing—original draft: R.S., D.G.C.; Writing—review and editing: R.S., D.G.C.

Competing interest statement

The authors declare that no competing interests exist.

Data availability statement

This study was pre-registered in the Open Science Framework (OSF) at https://doi.org/10.17605/OSF.IO/ZNWR6. The analytical R code, data, and supplementary materials are fully disclosed and contained in the respective OSF project at https://doi.org/10.17605/OSF.IO/XU8ES. The data underlying the illustrative example were extracted from Schroeders, Kubera, and Gnambs,Reference Schroeders, Kubera and Gnambs 43 who have made them openly available at https://osf.io/uxtks/.

Funding statement

This work is part of the Centres of Excellence scheme, funded by the Research Council of Norway (Project No. 331640), and the project “Mapping of Longitudinal Data of Inequalities in Education (MapIE),” funded by the European Commission under the Horizon Europe scheme (Project No. 101132474).

Footnotes

This article was awarded Open Data and Open Materials badges for transparent practices. See the Data availability statement for details.

i. Although we focus on synthesizing correlation coefficients in this study, the formulation of the MLMV-REM can be extended to other kinds of effect sizes.

ii. In this list of approaches, we focus on those that approximate or construct sampling covariance matrices for correlation matrices. If the MLMV-REM is specified for other kinds of effect sizes, other approaches may be accessible, such as fixing a constant sampling correlation and then approximating the sampling covariance matrices.Reference Pustejovsky and Tipton 5

References

Becker, BJ, Aloe, AM, Cheung, MW-L. Meta-analysis of correlations, correlation matrices, and their functions. In: Schmid, CH, Stijnen, T, White, IR, eds. Handbook of Meta-Analysis. 1st ed. CRC Press and Routledge/Taylor & Francis Group, New York; 2021:347369.Google Scholar
Cheung, MW-L, Cheung, SF. Random-effects models for meta-analytic structural equation modeling: review, issues, and illustrations. Res Synth Methods. 2016;7(2):140155. https://doi.org/10.1002/jrsm.1166.CrossRefGoogle ScholarPubMed
Cheung, MW-L. Meta-Analysis: A Structural Equation Modeling Approach. John Wiley & Sons Ltd., Chichester; 2015.10.1002/9781118957813CrossRefGoogle Scholar
Jak, S. Meta-Analytic Structural Equation Modelling. Springer International Publishing, Cham; 2015.10.1007/978-3-319-27174-3CrossRefGoogle Scholar
Pustejovsky, JE, Tipton, E. Meta-analysis with robust variance estimation: expanding the range of working models. Prev Sci. 2022;23:425438. https://doi.org/10.1007/s11121-021-01246-3.CrossRefGoogle ScholarPubMed
Scherer, R, Siddiq, F, Tondeur, J. The technology acceptance model (TAM): a meta-analytic structural equation modeling approach to explaining teachers’ adoption of digital technology in education. Comput Educ. 2019;128:1335. https://doi.org/10.1016/j.compedu.2018.09.009.CrossRefGoogle Scholar
Campos, DG, Cheung, MW-L, Scherer, R. A primer on synthesizing individual participant data obtained from complex sampling surveys: a two-stage IPD meta-analysis approach. Psychol Methods. 2025;30(1):83111. https://doi.org/10.1037/met0000539.CrossRefGoogle ScholarPubMed
Fernández-Castilla, B, Jamshidi, L, Declercq, L, Beretvas, SN, Onghena, P, Van den Noortgate, W. The application of meta-analytic (multi-level) models with multiple random effects: a systematic review. Behav Res Methods. 2020;52(5):20312052. https://doi.org/10.3758/s13428-020-01373-9.CrossRefGoogle ScholarPubMed
Hedges, LV, Tipton, E, Johnson, MC. Robust variance estimation in meta-regression with dependent effect size estimates. Res Synth Methods. 2010;1(1):3965. https://doi.org/10.1002/jrsm.5.CrossRefGoogle ScholarPubMed
Stolwijk, I, Jak, S, Eichelsheim, V, Hoeve, M. Dealing with dependent effect sizes in MASEM. Z Psychol. 2022;230(1):1632. https://doi.org/10.1027/2151-2604/a000485.Google Scholar
Ke, Z, Zhang, Q, Tong, X. Bayesian meta-analytic SEM: a one-stage approach to modeling between-studies heterogeneity in structural parameters. Struct Equ Model Multidiscip J. 2019;26(3):348370. https://doi.org/10.1080/10705511.2018.1530059.CrossRefGoogle Scholar
Wilson, SJ, Polanin, JR, Lipsey, MW. Fitting meta-analytic structural equation models with complex datasets. Res Synth Methods. 2016;7(2):121139. https://doi.org/10.1002/jrsm.1199.CrossRefGoogle ScholarPubMed
Sheng, Z, Kong, W, Cortina, JM, Hou, S. Analyzing matrices of meta-analytic correlations: current practices and recommendations. Res Synth Methods. 2016;7(2):187208. https://doi.org/10.1002/jrsm.1206.CrossRefGoogle ScholarPubMed
Scherer, R, Siddiq, F, Nilsen, T. The potential of international large-scale assessments for meta-analyses in education. Large-Scale Assessments in Education. 2024;12(1):4. https://doi.org/10.1186/s40536-024-00191-1.CrossRefGoogle Scholar
López-López, JA, Page, MJ, Lipsey, MW, Higgins, JPT. Dealing with effect size multiplicity in systematic reviews and meta-analyses. Res Synth Methods. 2018;9(3):336351. https://doi.org/10.1002/jrsm.1310.CrossRefGoogle ScholarPubMed
Cheung, MW-L. A guide to conducting a meta-analysis with non-independent effect sizes. Neuropsychol Rev. 2019;29(4):387396. https://doi.org/10.1007/s11065-019-09415-6.CrossRefGoogle ScholarPubMed
Borenstein, M, Hedges, LV, Higgins, JPT, Rothstein, HR. Introduction to Meta-Analysis. John Wiley & Sons, Ltd., Chichester; 2009.CrossRefGoogle Scholar
Van den Noortgate, W, López-López, JA, Marín-Martínez, F, Sánchez-Meca, J. Three-level meta-analysis of dependent effect sizes. Behav Res Methods. 2013;45(2):576594. https://doi.org/10.3758/s13428-012-0261-6.CrossRefGoogle ScholarPubMed
Harrer, M, Cuijpers, P, Furukawa, TA, Ebert, DD. Doing Meta-Analysis in R: A Hands-On Guide. Chapman & Hall/CRC Press, Boca Raton and London; 2022.Google Scholar
Moeyaert, M, Ugille, M, Beretvas, NS, Ferron, J, Bunuan, R, Van Den Noortgate, W. Methods for dealing with multiple outcomes in meta-analysis: a comparison between averaging effect sizes, robust variance estimation and multilevel meta-analysis. Int J Soc Res Methodol. 2017;20(6):559572. https://doi.org/10.1080/13645579.2016.1252189.CrossRefGoogle Scholar
Riley, RD. Multivariate meta-analysis: the effect of ignoring within-study correlation. J R Stat Soc A Stat Soc. 2009;172(4):789811. https://doi.org/10.1111/j.1467-985X.2008.00593.x.CrossRefGoogle Scholar
Becker, BJ, Aloe, AM. Model-based meta-analysis and related approaches. In: Cooper, H, Hedges, LV, Valentine, JC, eds. The Handbook of Research Synthesis and Meta-Analysis. 3rd ed. Russell Sage Foundation, New York; 2019:339366:chap 16.10.7758/9781610448864.19CrossRefGoogle Scholar
Becker, BJ. Using results from replicated studies to estimate linear models. J Educ Stat. 1992;17(4):341362. https://doi.org/10.3102/10769986017004341.CrossRefGoogle Scholar
Jackson, D, White, IR, Price, M, Copas, J, Riley, RD. Borrowing of strength and study weights in multivariate and network meta-analysis. Stat Methods Med Res. 2017;26(6):28532868. https://doi.org/10.1177/0962280215611702.CrossRefGoogle ScholarPubMed
Jackson, D, White, IR, Riley, RD. Multivariate meta-analysis. In: Schmid, CH, Stijnen, T, White, IR, eds. Handbook of Meta-Analysis. CRC Press, New York; 2021:163186:chap 9.Google Scholar
Ahn, S, Ames, AJ, Myers, ND. A review of meta-analyses in education: methodological strengths and weaknesses. Rev Educ Res. 2012;82(4):436476. https://doi.org/10.3102/0034654312458162.CrossRefGoogle Scholar
Card, NA. Applied Meta-Analysis for Social Science Research. Guilford Press, New York; 2012.Google Scholar
Viechtbauer, W. Conducting meta-analyses in R with the metafor package. J Stat Softw. 2010;36(3):148.10.18637/jss.v036.i03CrossRefGoogle Scholar
Cheung, MW-L. Multivariate meta-analysis as structural equation models. Struct Equ Model Multidiscip J. 2013;20(3):429454. https://doi.org/10.1080/10705511.2013.797827.CrossRefGoogle Scholar
Cheung, MW-L, Chan, W. Meta-analytic structural equation modeling: a two-stage approach. Psychol Methods. 2005;10(1):4064. https://doi.org/10.1037/1082-989X.10.1.40.CrossRefGoogle ScholarPubMed
Jak, S, Cheung, MWL. Accounting for missing correlation coefficients in fixed-effects MASEM. Multivar Behav Res. 2018;53(1):114. https://doi.org/10.1080/00273171.2017.1375886.CrossRefGoogle ScholarPubMed
McShane, BB, Böckenholt, U. Multilevel multivariate meta-analysis made easy: an introduction to MLMVmeta. Behav Res Methods. 2023;55(5):23672386. https://doi.org/10.3758/s13428-022-01892-7.CrossRefGoogle ScholarPubMed
Wei, Y, Higgins, JPT. Estimating within-study covariances in multivariate meta-analysis with multiple outcomes. Stat Med. 2013;32(7):11911205. https://doi.org/10.1002/sim.5679.CrossRefGoogle ScholarPubMed
Olkin, I, Finn, JD. Testing correlated correlations. Psychol Bull. 1990;108(2):330333. https://doi.org/10.1037/0033-2909.108.2.330.CrossRefGoogle Scholar
Olkin, I, Siotani, M. Asymptotic Distribution of Functions of a Correlation Matrix. Technical Report No. 6. Laboratory for Quantitative Research in Education, Stanford University; 1964. 1976. Essays in probability and statistics.Google Scholar
Pustejovsky, JE. Correlations between z-transformed correlation coefficients; 2024. https://jepusto.com/posts/correlated-z-transformed-correlations/. Updated June 6, 2024.Google Scholar
Snijders, TAB, Bosker, RJ. Multilevel Analysis: An Introduction to Basic and Advanced Multilevel Modeling. 2nd ed. Sage Publications, Thousand Oaks; 2012.Google Scholar
Rights, JD, Sterba, SK. Quantifying explained variance in multilevel models: an integrative framework for defining R-squared measures. Psychol Methods. 2019;24(3):309338. https://doi.org/10.1037/met0000184.CrossRefGoogle ScholarPubMed
McShane, BB, Böckenholt, U. Multilevel multivariate meta-analysis with application to choice overload. Psychometrika. 2018;83(1):255271. https://doi.org/10.1007/s11336-017-9571-z.CrossRefGoogle ScholarPubMed
Lu, M. Computing within-study covariances, data visualization, and missing data solutions for multivariate meta-analysis with metavcov.. Front Psychol. 2023;14:1185012. https://doi.org/10.3389/fpsyg.2023.1185012.CrossRefGoogle ScholarPubMed
Riley, RD, Thompson, JR, Abrams, KR. An alternative model for bivariate random-effects meta-analysis when the within-study correlations are unknown. Biostatistics. 2008;9(1):172186. https://doi.org/10.1093/biostatistics/kxm023.CrossRefGoogle ScholarPubMed
clubSandwich: Cluster-Robust (Sandwich) Variance Estimators with Small-Sample Corrections . R package version 0.5.3; 2021. https://CRAN.R-project.org/package=clubSandwich.Google Scholar
Schroeders, U, Kubera, F, Gnambs, T. The structure of the Toronto Alexithymia Scale (TAS-20): a meta-analytic confirmatory factor analysis. Assessment. 2022;29(8):18061823. https://doi.org/10.1177/10731911211033894.Google ScholarPubMed
Cheung, MW-L. metaSEM: an R package for meta-analysis using structural equation modeling. Front Psychol. 2015;5:1521. https://doi.org/10.3389/fpsyg.2014.01521.CrossRefGoogle Scholar
Cheung, MWL, Chan, W. Testing dependent correlation coefficients via structural equation modeling. Organ Res Methods. 2004;7(2):206223. https://doi.org/10.1177/1094428104264024.CrossRefGoogle Scholar
Lin, L, Xu, C. Arcsine-based transformations for meta-analysis of proportions: pros, cons, and alternatives. Health Science Reports. 2020;3(3):e178. https://doi.org/10.1002/hsr2.178.CrossRefGoogle ScholarPubMed
Schwarzer, G, Chemaitelly, H, Abu-Raddad, LJ, Rücker, G. Seriously misleading results using inverse of Freeman–Tukey double arcsine transformation in meta-analysis of single proportions. Res Synth Methods. 2019;10(3):476483. https://doi.org/10.1002/jrsm.1348.CrossRefGoogle ScholarPubMed
Vembye, MH, Pustejovsky, JE, Pigott, TD. Power approximations for overall average effects in meta-analysis with dependent effect sizes. J Educ Behav Stat. 2023;48(1):70102. https://doi.org/10.3102/10769986221127379.CrossRefGoogle Scholar
Vembye, MH, Pustejovsky, JE, Pigott, TD. Conducting power analysis for meta-analysis with dependent effect sizes: common guidelines and an introduction to the POMADE R package. Res Synth Methods. 2024;15(6):12141230. https://doi.org/10.1002/jrsm.1752.CrossRefGoogle Scholar
Cheung, MW-L, Chan, W. A two-stage approach to synthesizing covariance matrices in meta-analytic structural equation modeling. Struct Equ Model Multidiscip J. 2009;16(1):2853. https://doi.org/10.1080/10705510802561295.CrossRefGoogle Scholar
Jak, S, Cheung, MW-L. Meta-analytic structural equation modeling with moderating effects on SEM parameters. Psychol Methods. 2020;25(4):430455. https://doi.org/10.1037/met0000245.CrossRefGoogle ScholarPubMed
Pigott, TD. Methods for handling missing data in research synthesis. In: Cooper, H, Hedges, LV, eds. The Handbook of Research Synthesis. Russell Sage Foundation, New York; 1994:163175.Google Scholar
Figure 0

Table 1 Specification of meta-analytic working models with the MLMV-REM approach

Figure 1

Figure 1 Decision scheme for selecting a working model.

Figure 2

Figure 2 Elements of the analytic code to specify Working Model 1 in the R package “metafor”.

Figure 3

Table 2 Weighted average effect sizes and variance components in Working Models 1–4 with independent random effects (with $\rho = \phi=0$)

Supplementary material: File

Scherer and Campos supplementary material

Scherer and Campos supplementary material
Download Scherer and Campos supplementary material(File)
File 2.4 MB