rma.uni: Meta-Analysis via Linear (Mixed-Effects) Models In metafor: Meta-Analysis Package for R

Description

Function to fit the meta-analytic fixed- and random/mixed-effects models with or without moderators via linear (mixed-effects) models. See the documentation of the metafor-package for more details on these models. \loadmathjax

Usage

  1 2 3 4 5 6 7 8 9 10 11 12 rma.uni(yi, vi, sei, weights, ai, bi, ci, di, n1i, n2i, x1i, x2i, t1i, t2i, m1i, m2i, sd1i, sd2i, xi, mi, ri, ti, sdi, r2i, ni, mods, scale, measure="GEN", intercept=TRUE, data, slab, subset, add=1/2, to="only0", drop00=FALSE, vtype="LS", method="REML", weighted=TRUE, test="z", level=95, digits, btt, att, tau2, verbose=FALSE, control, ...) rma(yi, vi, sei, weights, ai, bi, ci, di, n1i, n2i, x1i, x2i, t1i, t2i, m1i, m2i, sd1i, sd2i, xi, mi, ri, ti, sdi, r2i, ni, mods, scale, measure="GEN", intercept=TRUE, data, slab, subset, add=1/2, to="only0", drop00=FALSE, vtype="LS", method="REML", weighted=TRUE, test="z", level=95, digits, btt, att, tau2, verbose=FALSE, control, ...) 

Arguments

 yi vector of length \mjseqnk with the observed effect sizes or outcomes. See ‘Details’. vi vector of length \mjseqnk with the corresponding sampling variances. See ‘Details’. sei vector of length \mjseqnk with the corresponding standard errors (only relevant when not using vi). See ‘Details’. weights optional argument to specify a vector of length \mjseqnk with user-defined weights. See ‘Details’. ai see below and the documentation of the escalc function for more details. bi see below and the documentation of the escalc function for more details. ci see below and the documentation of the escalc function for more details. di see below and the documentation of the escalc function for more details. n1i see below and the documentation of the escalc function for more details. n2i see below and the documentation of the escalc function for more details. x1i see below and the documentation of the escalc function for more details. x2i see below and the documentation of the escalc function for more details. t1i see below and the documentation of the escalc function for more details. t2i see below and the documentation of the escalc function for more details. m1i see below and the documentation of the escalc function for more details. m2i see below and the documentation of the escalc function for more details. sd1i see below and the documentation of the escalc function for more details. sd2i see below and the documentation of the escalc function for more details. xi see below and the documentation of the escalc function for more details. mi see below and the documentation of the escalc function for more details. ri see below and the documentation of the escalc function for more details. ti see below and the documentation of the escalc function for more details. sdi see below and the documentation of the escalc function for more details. r2i see below and the documentation of the escalc function for more details. ni see below and the documentation of the escalc function for more details. mods optional argument to include one or more moderators in the model. A single moderator can be given as a vector of length \mjseqnk specifying the values of the moderator. Multiple moderators are specified by giving a matrix with \mjseqnk rows and as many columns as there are moderator variables. Alternatively, a model formula can be used to specify the model. See ‘Details’. scale optional argument to include one or more predictors for the scale part in a location-scale model. See ‘Details’. measure character string to specify the type of data supplied to the function. When measure="GEN" (default), the observed effect sizes or outcomes and corresponding sampling variances (or standard errors) should be supplied to the function via the yi, vi, and sei arguments (only one of the two, vi or sei, needs to be specified). Alternatively, one can set measure to one of the effect size or outcome measures described under the documentation for the escalc function in which case one must specify the required data via the appropriate arguments. intercept logical to specify whether an intercept should be added to the model (the default is TRUE). Ignored when mods is a formula. data optional data frame containing the data supplied to the function. slab optional vector with labels for the \mjseqnk studies. subset optional (logical or numeric) vector to specify the subset of studies that should be used for the analysis. add see the documentation of the escalc function. to see the documentation of the escalc function. drop00 see the documentation of the escalc function. vtype see the documentation of the escalc function. method character string to specify whether a fixed- or a random/mixed-effects model should be fitted. A fixed-effects model (with or without moderators) is fitted when using method="FE". Random/mixed-effects models are fitted by setting method equal to one of the following: "DL", "HE", "SJ", "ML", "REML", "EB", "HS", "HSk", or "GENQ". Default is "REML". See ‘Details’. weighted logical to specify whether weighted (default) or unweighted estimation should be used to fit the model. test character string to specify how test statistics and confidence intervals for the fixed effects should be computed. By default (test="z"), Wald-type tests and CIs are obtained, which are based on a standard normal distribution. When test="t", a t-distribution is used instead. When test="knha", the method by Knapp and Hartung (2003) is used. See ‘Details’. level numeric value between 0 and 100 to specify the confidence interval level (the default is 95). digits integer to specify the number of decimal places to which the printed results should be rounded. If unspecified, the default is 4. btt optional vector of indices to specify which coefficients to include in the omnibus test of moderators. Can also be a string to grep for. See ‘Details’. att optional vector of indices to specify which scale coefficients to include in the omnibus test. Only relevant for location-scale models. See ‘Details’. tau2 optional numeric value to specify the amount of (residual) heterogeneity in a random- or mixed-effects model (instead of estimating it). Useful for sensitivity analyses (e.g., for plotting results as a function of \mjseqn\tau^2). When unspecified, the value of \mjseqn\tau^2 is estimated from the data. verbose logical to specify whether output should be generated on the progress of the model fitting (the default is FALSE). Can also be an integer. Values > 1 generate more verbose output. See ‘Note’. control optional list of control values for the iterative estimation algorithms. If unspecified, default values are defined inside the function. See ‘Note’. ... additional arguments.

Details

Specifying the Data

The function can be used in conjunction with any of the usual effect size or outcome measures used in meta-analyses (e.g., log risk ratios, log odds ratios, risk differences, mean differences, standardized mean differences, log transformed ratios of means, raw correlation coefficients, correlation coefficients transformed with Fisher's r-to-z transformation, and so on). Simply specify the observed effect sizes or outcomes via the yi argument and the corresponding sampling variances via the vi argument. Instead of specifying vi, one can specify the standard errors (the square root of the sampling variances) via the sei argument. The escalc function can be used to compute a wide variety of effect size or outcome measures (and the corresponding sampling variances) based on summary statistics.

Alternatively, the function can automatically calculate the values of a chosen effect size or outcome measure (and the corresponding sampling variances) when supplied with the necessary data. The escalc function describes which effect size or outcome measures are currently implemented and what data/arguments should then be specified/used. The measure argument should then be set to the desired effect size or outcome measure.

Specifying the Model

The function can be used to fit fixed- and random/mixed-effects models, as well as meta-regression models including moderators (the difference between the various models is described in detail on the introductory metafor-package help page).

Assuming the observed effect sizes or outcomes and corresponding sampling variances are supplied via yi and vi, a fixed-effects model can be fitted with rma(yi, vi, method="FE"). Weighted estimation (with inverse-variance weights) is used by default. User-defined weights can be supplied via the weights argument. Unweighted estimation can be used by setting weighted=FALSE (which is the same as setting the weights equal to a constant).

A random-effects model can be fitted with the same code but setting the method argument to one of the various estimators for the amount of heterogeneity:

• method="DL" = DerSimonian-Laird estimator,

• method="HE" = Hedges estimator,

• method="HS" = Hunter-Schmidt estimator,

• method="HSk" = Hunter-Schmidt estimator with a small sample-size correction,

• method="SJ" = Sidik-Jonkman estimator,

• method="ML" = maximum-likelihood estimator,

• method="REML" = restricted maximum-likelihood estimator,

• method="EB" = empirical Bayes estimator,

• method="PM" = Paule-Mandel estimator,

• method="GENQ" = generalized Q-statistic estimator.

For a description of the various estimators, see Brannick et al. (2019), DerSimonian and Kacker (2007), Raudenbush (2009), Viechtbauer (2005), and Viechtbauer et al. (2015). Note that the Hedges estimator is also called the ‘variance component estimator’ or ‘Cochran estimator’, the Sidik-Jonkman estimator is also called the ‘model error variance estimator’, and the empirical Bayes estimator is actually identical to the Paule-Mandel estimator (Paule & Mandel, 1982). Finally, the generalized Q-statistic estimator is a general method-of-moments estimator (DerSimonian & Kacker, 2007) requiring the specification of weights (the HE and DL estimators are just special cases with equal and inverse variance weights, respectively).

One or more moderators can be included in these models via the mods argument. A single moderator can be given as a (row or column) vector of length \mjseqnk specifying the values of the moderator. Multiple moderators are specified by giving an appropriate model matrix (i.e., \mjseqnX) with \mjseqnk rows and as many columns as there are moderator variables (e.g., mods = cbind(mod1, mod2, mod3), where mod1, mod2, and mod3 correspond to the names of the variables for three moderator variables). The intercept is added to the model matrix by default unless intercept=FALSE.

Alternatively, one can use standard formula syntax to specify the model. In this case, the mods argument should be set equal to a one-sided formula of the form mods = ~ model (e.g., mods = ~ mod1 + mod2 + mod3). Interactions, polynomial terms, and factors can be easily added to the model in this manner. When specifying a model formula via the mods argument, the intercept argument is ignored. Instead, the inclusion/exclusion of the intercept is controlled by the specified formula (e.g., mods = ~ mod1 + mod2 + mod3 - 1 would lead to the removal of the intercept).

When the observed effect sizes or outcomes and corresponding sampling variances are supplied via the yi and vi (or sei) arguments, one can also specify moderators via the yi argument (e.g., rma(yi ~ mod1 + mod2 + mod3, vi)). In that case, the mods argument is ignored and the inclusion/exclusion of the intercept again is controlled by the specified formula.

Omnibus Test of Moderators

For models including moderators, an omnibus test of all model coefficients is conducted that excludes the intercept (the first coefficient) if it is included in the model. If no intercept is included in the model, then the omnibus test includes all of the coefficients in the model including the first. Alternatively, one can manually specify the indices of the coefficients to test via the btt argument. For example, with btt=c(3,4), only the third and fourth coefficient from the model would be included in the test (if an intercept is included in the model, then it corresponds to the first coefficient in the model). Instead of specifying the coefficient numbers, one can specify a string for btt. In that case, grep will be used to search for all coefficient names that match the string. The omnibus test is called the \mjseqnQ_M-test and follows, under the assumptions of the model, a chi-square distribution with \mjseqnm degrees of freedom (with \mjseqnm denoting the number of coefficients tested) under the null hypothesis (that the true value of all coefficients tested is equal to 0).

Categorical Moderators

Categorical moderator variables can be included in the model via the mods argument in the same way that appropriately (dummy) coded categorical independent variables can be included in linear models. One can either do the dummy coding manually or use a model formula together with the factor function to let R handle the coding automatically (note that string/character variables in a model formula are automatically converted to factors). An example to illustrate these different approaches is provided below.

Tests and Confidence Intervals

By default, tests of individual coefficients in the model (and the corresponding confidence intervals) are based on a standard normal distribution, while the omnibus test is based on a chi-square distribution (see above). As an alternative, one can set test="t", in which case tests of individual coefficients and confidence intervals are based on a t-distribution with \mjseqnk-p degrees of freedom, while the omnibus test statistic then uses an F-distribution with \mjseqnm and \mjseqnk-p degrees of freedom (with \mjseqnk denoting the total number of estimates included in the analysis and \mjseqnp the total number of model coefficients including the intercept if it is present). Finally, when test="knha", the Knapp and Hartung (2003) method is used, which applies an adjustment to the standard errors of the estimated coefficients (to account for the uncertainty in the estimate of the amount of (residual) heterogeneity) and uses t- and F-distributions as described above.

Test for (Residual) Heterogeneity

A test for (residual) heterogeneity is automatically carried out by the function. Without moderators in the model, this is simply Cochran's \mjseqnQ-test (Cochran, 1954), which tests whether the variability in the observed effect sizes or outcomes is larger than would be expected based on sampling variability alone. A significant test suggests that the true effects/outcomes are heterogeneous. When moderators are included in the model, this is the \mjseqnQ_E-test for residual heterogeneity, which tests whether the variability in the observed effect sizes or outcomes not accounted for by the moderators included in the model is larger than would be expected based on sampling variability alone.

Location-Scale Models

The function can also be used to fit so-called ‘location-scale models’. In such models, one can specify not only predictors for the size of the average true outcome (i.e., for their ‘location’), but also predictors for the amount of heterogeneity in the outcomes (i.e., their ‘scale’). The model is then given by \mjdeqny_i = \beta_0 + \beta_1 x_i1 + \beta_2 x_i2 + ... + \beta_p' x_ip' + u_i + \epsilon_i,y_i = \beta_0 + \beta_1 x_i1 + \beta_2 x_i2 + ... + \beta_p' x_ip' + u_i + \epsilon_i, \mjdeqnu_i \sim N(0, \tau_i^2), \; \epsilon_i \sim N(0, v_i),u_i ~ N(0, tau_i^2), \epsilon_i \sim N(0, v_i), \mjdeqn\ln(\tau_i^2) = \alpha_0 + \alpha_1 z_i1 + \alpha_2 z_i2 + ... + \alpha_q' z_iq',log(tau^2) = \alpha_0 + \alpha z_i1 + \alpha z_i2 + ... + \alpha_q' z_iq', where \mjeqnx_i1, ..., x_ip'x_i1, ..., x_ip' are the values of the \mjseqnp' predictor variables that may be related to the size of the average true outcome (letting \mjseqnp = p' + 1 denote the total number of location coefficients in the model including the model intercept \mjseqn\beta_0) and \mjeqnz_i1, ..., z_iq'z_i1, ..., z_iq' are the values of the \mjseqnq' scale variables that may be related to the amount of heterogeneity in the outcomes (letting \mjseqnq = q' + 1 denote the total number of scale coefficients in the model including the model intercept \mjseqn\alpha_0). Location variables can be specified via the mods argument as described above (e.g., mods = ~ mod1 + mod2 + mod3). Scale variables can be specified via the scale argument (e.g., scale = ~ var1 + var2 + var3). A log link is used for specifying the relationship between the scale variables and the amount of heterogeneity so that \mjseqn\tau_i^2 is guaranteed to be non-negative. Estimates of the location and scale coefficients can be obtained with either maximum likelihood (method="ML") or restricted maximum likelihood (method="REML") estimation. An omnibus test of the scale coefficients is conducted as described above (where the att argument can be used to specify which coefficients to include in the test).

Value

An object of class c("rma.uni","rma"). The object is a list containing the following components:

 beta estimated coefficients of the model. se standard errors of the coefficients. zval test statistics of the coefficients. pval corresponding p-values. ci.lb lower bound of the confidence intervals for the coefficients. ci.ub upper bound of the confidence intervals for the coefficients. vb variance-covariance matrix of the estimated coefficients. tau2 estimated amount of (residual) heterogeneity. Always 0 when method="FE". se.tau2 standard error of the estimated amount of (residual) heterogeneity. k number of studies included in the analysis. p number of coefficients in the model (including the intercept). m number of coefficients included in the omnibus test of moderators. QE test statistic of the test for (residual) heterogeneity. QEp corresponding p-value. QM test statistic of the omnibus test of moderators. QMp corresponding p-value. I2 value of \mjseqnI^2. See print.rma.uni for more details. H2 value of \mjseqnH^2. See print.rma.uni for more details. R2 value of \mjseqnR^2. See print.rma.uni for more details. int.only logical that indicates whether the model is an intercept-only model. yi, vi, X the vector of outcomes, the corresponding sampling variances, and the model matrix. fit.stats a list with the log-likelihood, deviance, AIC, BIC, and AICc values under the unrestricted and restricted likelihood. ... some additional elements/values.

For location-scale models, the object is of class c("rma.ls","rma.uni","rma") and includes the following components in addition to the ones listed above:

 alpha estimated scale coefficients of the model. se.alpha standard errors of the coefficients. zval.alpha test statistics of the coefficients. pval.alpha corresponding p-values. ci.lb.alpha lower bound of the confidence intervals for the coefficients. ci.ub.alpha upper bound of the confidence intervals for the coefficients. va variance-covariance matrix of the estimated coefficients. tau2 as above, but now a vector of values. q number of scale coefficients in the model (including the intercept). QS test statistic of the omnibus test of the scale coefficients. QSp corresponding p-value. ... some additional elements/values.

Methods

The results of the fitted model are formatted and printed with the print.rma.uni function. If fit statistics should also be given, use summary.rma (or use the fitstats.rma function to extract them). Full versus reduced model comparisons in terms of fit statistics and likelihoods can be obtained with anova.rma. Wald-type tests for sets of model coefficients or linear combinations thereof can be obtained with the same function. Permutation tests for the model coefficient(s) can be obtained with permutest.rma.uni. Tests and confidence intervals based on (cluster) robust methods can be obtained with robust.rma.uni.

Predicted/fitted values can be obtained with predict.rma and fitted.rma. For best linear unbiased predictions, see blup.rma.uni and ranef.rma.uni.

The residuals.rma, rstandard.rma.uni, and rstudent.rma.uni functions extract raw and standardized residuals. Additional case diagnostics (e.g., to determine influential studies) can be obtained with the influence.rma.uni function. For models without moderators, leave-one-out diagnostics can also be obtained with leave1out.rma.uni. For models with moderators, variance inflation factors can be obtained with vif.rma.

A confidence interval for the amount of (residual) heterogeneity in the random/mixed-effects model can be obtained with confint.rma.uni.

Forest, funnel, radial, L'Abbé, and Baujat plots can be obtained with forest.rma, funnel.rma, radial.rma, labbe.rma, and baujat.rma (radial and L'Abbé plots only for models without moderators). The qqnorm.rma.uni function provides normal QQ plots of the standardized residuals. One can also just call plot.rma.uni on the fitted model object to obtain various plots at once. For random/mixed-effects models, the profile.rma.uni function can be used to obtain a plot of the (restricted) log-likelihood as a function of \mjseqn\tau^2. For models with moderators, regplot.rma draws scatter plots / bubble plots, showing the (marginal) relationship between the observed outcomes and a selected moderator from the model.

Tests for funnel plot asymmetry (which may be indicative of publication bias) can be obtained with ranktest.rma and regtest.rma. For models without moderators, the trimfill.rma.uni method can be used to carry out a trim and fill analysis and hc.rma.uni provides a random-effects model analysis that is more robust to publication bias (based on the method by Henmi & Copas, 2010). The test of ‘excess significance’ can be carried out with the tes.rma function. Selection models can be fitted with the selmodel function.

For models without moderators, a cumulative meta-analysis (i.e., adding one observation at a time) can be obtained with cumul.rma.uni.

Other extractor functions include coef.rma, vcov.rma, logLik.rma, deviance.rma, AIC.rma, BIC.rma, hatvalues.rma.uni, and weights.rma.uni.

Note

While the HS, HSk, HE, DL, SJ, and GENQ estimators of \mjseqn\tau^2 are based on closed-form solutions, the ML, REML, and EB estimators must be obtained iteratively. For this, the function makes use of the Fisher scoring algorithm, which is robust to poor starting values and usually converges quickly (Harville, 1977; Jennrich & Sampson, 1976). By default, the starting value is set equal to the value of the Hedges (HE) estimator and the algorithm terminates when the change in the estimated value of \mjseqn\tau^2 is smaller than \mjeqn10^-510^(-5) from one iteration to the next. The maximum number of iterations is 100 by default (which should be sufficient in most cases). Information on the progress of the algorithm can be obtained by setting verbose=TRUE. One can also set verbose to an integer (verbose=2 yields even more information and verbose=3 also sets option(warn=1) temporarily).

A different starting value, threshold, and maximum number of iterations can be specified via the control argument by setting control=list(tau2.init=value, threshold=value, maxiter=value). The step length of the Fisher scoring algorithm can also be adjusted by a desired factor with control=list(stepadj=value) (values below 1 will reduce the step length). If using verbose=TRUE shows the estimate jumping around erratically (or cycling through a few values), decreasing the step length (and increasing the maximum number of iterations) can often help with convergence (e.g., control=list(stepadj=0.5, maxiter=1000)).

The PM estimator also involves an iterative algorithm, which makes use of the uniroot function. By default, the desired accuracy (tol) is set equal to .Machine$double.eps^0.25 and the maximum number of iterations (maxiter) to 100 (as above). The upper bound of the interval searched (tau2.max) is set to 100 (which should be large enough for most cases). These values can be adjusted with control=list(tol=value, maxiter=value, tau2.max=value). All of the heterogeneity estimators except SJ can in principle yield negative estimates for the amount of (residual) heterogeneity. However, negative estimates of \mjseqn\tau^2 are outside of the parameter space. For the HS, HSk, HE, DL, and GENQ estimators, negative estimates are therefore truncated to zero. For the ML, REML, and EB estimators, the Fisher scoring algorithm makes use of step halving (Jennrich & Sampson, 1976) to guarantee a non-negative estimate. Finally, for the PM estimator, the lower bound of the interval searched is set to zero by default. For those brave enough to step into risky territory, there is the option to set the lower bound for all these estimators to some other value besides zero (even a negative one) with control=list(tau2.min=value), but the lowest value permitted is -min(vi) (to ensure that the marginal variances are always non-negative). The Hunter-Schmidt estimator for the amount of heterogeneity is defined in Hunter and Schmidt (1990) only in the context of the random-effects model when analyzing correlation coefficients. A general version of this estimator for random- and mixed-effects models not specific to any particular outcome measure is described in Viechtbauer (2005) and Viechtbauer et al. (2015) and is implemented here. The Sidik-Jonkman estimator starts with a crude estimate of \mjseqn\tau^2, which is then updated as described in Sidik and Jonkman (2005b, 2007). If, instead of the crude estimate, one wants to use a better a priori estimate, one can do so by passing this value via control=list(tau2.init=value). Outcomes with non-positive sampling variances are problematic. If a sampling variance is equal to zero, then its weight will be \mjseqn1/0 for fixed-effects models when using weighted estimation. Switching to unweighted estimation is a possible solution then. For random/mixed-effects model, some estimators of \mjseqn\tau^2 are undefined when there is at least one sampling variance equal to zero. Other estimators may work, but it may still be necessary to switch to unweighted model fitting, especially when the estimate of \mjseqn\tau^2 converges to zero. When including moderators in the model, it is possible that the model matrix is not of full rank (i.e., there is a linear relationship between the moderator variables included in the model). The function automatically tries to reduce the model matrix to full rank by removing redundant predictors, but if this fails the model cannot be fitted and an error will be issued. Deleting (redundant) moderator variables from the model as needed should solve this problem. Finally, some general words of caution about the assumptions underlying the models: • The sampling variances (i.e., the vi values) are treated as if they are known constants. This (usually) implies that the distributions of the test statistics and corresponding confidence intervals are only exact and have nominal coverage when the within-study sample sizes are large (i.e., when the error in the sampling variance estimates is small). Certain outcome measures (e.g., the arcsine square root transformed risk difference and Fisher's r-to-z transformed correlation coefficient) are based on variance stabilizing transformations that also help to make the assumption of known sampling variances much more reasonable. • When fitting a mixed/random-effects model, \mjseqn\tau^2 is estimated and then treated as a known constant thereafter. This ignores the uncertainty in the estimate of \mjseqn\tau^2. As a consequence, the standard errors of the parameter estimates tend to be too small, yielding test statistics that are too large and confidence intervals that are not wide enough. The Knapp and Hartung (2003) adjustment can be used to counter this problem, yielding test statistics and confidence intervals whose properties are closer to nominal. • Most effect size or outcome measures are not exactly normally distributed as assumed under the various models. However, the normal approximation usually becomes more accurate for most effect size or outcome measures as the within-study sample sizes increase. Therefore, sufficiently large within-study sample sizes are (usually) needed to be certain that the tests and confidence intervals have nominal levels/coverage. Again, certain outcome measures (e.g., Fisher's r-to-z transformed correlation coefficient) may be preferable from this perspective as well. Author(s) Wolfgang Viechtbauer wvb@metafor-project.org https://www.metafor-project.org References Berkey, C. S., Hoaglin, D. C., Mosteller, F., & Colditz, G. A. (1995). A random-effects regression model for meta-analysis. Statistics in Medicine, 14(4), 395–411. https://doi.org/10.1002/sim.4780140406 Brannick, M. T., Potter, S. M., Benitez, B., & Morris, S. B. (2019). Bias and precision of alternate estimators in meta-analysis: Benefits of blending Schmidt–Hunter and Hedges approaches. Organizational Research Methods, 22(2), 490–514. https://doi.org/10.1177/1094428117741966 Cochran, W. G. (1954). The combination of estimates from different experiments. Biometrics, 10(1), 101–129. https://doi.org/10.2307/3001666 DerSimonian, R., & Laird, N. (1986). Meta-analysis in clinical trials. Controlled Clinical Trials, 7(3), 177–188. https://doi.org/10.1016/0197-2456(86)90046-2 DerSimonian, R., & Kacker, R. (2007). Random-effects model for meta-analysis of clinical trials: An update. Contemporary Clinical Trials, 28(2), 105–114. https://doi.org/10.1016/j.cct.2006.04.004 Harville, D. A. (1977). Maximum likelihood approaches to variance component estimation and to related problems. Journal of the American Statistical Association, 72(358), 320–338. https://doi.org/10.2307/2286796 Hedges, L. V. (1983). A random effects model for effect sizes. Psychological Bulletin, 93(2), 388–395. https://doi.org/10.1037/0033-2909.93.2.388 Hedges, L. V., & Olkin, I. (1985). Statistical methods for meta-analysis. San Diego, CA: Academic Press. Henmi, M., & Copas, J. B. (2010). Confidence intervals for random effects meta-analysis and robustness to publication bias. Statistics in Medicine, 29(29), 2969–2983. https://doi.org/10.1002/sim.4029 Hunter, J. E., & Schmidt, F. L. (2004). Methods of meta-analysis: Correcting error and bias in research findings (2nd ed.). Thousand Oaks, CA: Sage. Jennrich, R. I., & Sampson, P. F. (1976). Newton-Raphson and related algorithms for maximum likelihood variance component estimation. Technometrics, 18(1), 11–17. https://doi.org/10.2307/1267911 Knapp, G., & Hartung, J. (2003). Improved tests for a random effects meta-regression with a single covariate. Statistics in Medicine, 22(17), 2693–2710. https://doi.org/10.1002/sim.1482 Morris, C. N. (1983). Parametric empirical Bayes inference: Theory and applications. Journal of the American Statistical Association, 78(381), 47–55. https://doi.org/10.2307/2287098 Paule, R. C., & Mandel, J. (1982). Consensus values and weighting factors. Journal of Research of the National Bureau of Standards, 87(5), 377–385. https://doi.org/10.6028/jres.087.022 Raudenbush, S. W. (2009). Analyzing effect sizes: Random effects models. In H. Cooper, L. V. Hedges, & J. C. Valentine (Eds.), The handbook of research synthesis and meta-analysis (2nd ed., pp. 295–315). New York: Russell Sage Foundation. Sidik, K., & Jonkman, J. N. (2005a). A note on variance estimation in random effects meta-regression. Journal of Biopharmaceutical Statistics, 15(5), 823–838. https://doi.org/10.1081/BIP-200067915 Sidik, K., & Jonkman, J. N. (2005b). Simple heterogeneity variance estimation for meta-analysis. Journal of the Royal Statistical Society, Series C, 54(2), 367–384. https://doi.org/10.1111/j.1467-9876.2005.00489.x Sidik, K., & Jonkman, J. N. (2007). A comparison of heterogeneity variance estimators in combining results of studies. Statistics in Medicine, 26(9), 1964–1981. https://doi.org/10.1002/sim.2688 Viechtbauer, W. (2005). Bias and efficiency of meta-analytic variance estimators in the random-effects model. Journal of Educational and Behavioral Statistics, 30(3), 261–293. https://doi.org/10.3102/10769986030003261 Viechtbauer, W. (2010). Conducting meta-analyses in R with the metafor package. Journal of Statistical Software, 36(3), 1–48. https://doi.org/10.18637/jss.v036.i03 Viechtbauer, W., López-López, J. A., Sánchez-Meca, J., & Marín-Martínez, F. (2015). A comparison of procedures to test for moderators in mixed-effects meta-regression models. Psychological Methods, 20(3), 360–374. https://doi.org/10.1037/met0000023 See Also rma.mh, rma.peto, rma.glmm, and rma.mv for other model fitting functions. Examples   1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 ### calculate log risk ratios and corresponding sampling variances dat <- escalc(measure="RR", ai=tpos, bi=tneg, ci=cpos, di=cneg, data=dat.bcg) ### fit a random-effects model using the log risk ratios and variances as input ### note: method="REML" is the default, so one could leave this out rma(yi, vi, data=dat, method="REML") ### fit a random-effects model using the log risk ratios and standard errors as input ### note: the second argument of rma() is for the *variances*, so we use the ### named argument 'sei' to supply the standard errors to the function dat$sei <- sqrt(dat$vi) rma(yi, sei=sei, data=dat) ### fit a random-effects model supplying the 2x2 table cell frequencies to the function rma(measure="RR", ai=tpos, bi=tneg, ci=cpos, di=cneg, data=dat) ### fit a mixed-effects model with two moderators (absolute latitude and publication year) rma(yi, vi, mods=cbind(ablat, year), data=dat) ### using a model formula to specify the same model rma(yi, vi, mods = ~ ablat + year, data=dat) ### using a model formula as part of the yi argument rma(yi ~ ablat + year, vi, data=dat) ### manual dummy coding of the allocation factor alloc.random <- ifelse(dat$alloc == "random", 1, 0) alloc.alternate <- ifelse(dat$alloc == "alternate", 1, 0) alloc.systematic <- ifelse(dat$alloc == "systematic", 1, 0) ### test the allocation factor (in the presence of the other moderators) ### note: 'alternate' is the reference level of the allocation factor, ### since this is the dummy/level we leave out of the model ### note: the intercept is the first coefficient, so with btt=2:3 we test ### coefficients 2 and 3, corresponding to the coefficients for the ### allocation factor rma(yi, vi, mods = ~ alloc.random + alloc.systematic + year + ablat, data=dat, btt=2:3) ### using a model formula to specify the same model rma(yi, vi, mods = ~ factor(alloc) + year + ablat, data=dat, btt=2:3) ### factor() is not needed as character variables are automatically converted to factors rma(yi, vi, mods = ~ alloc + year + ablat, data=dat, btt=2:3) ### test all pairwise differences with Holm's method (using the 'multcomp' package if installed) res <- rma(yi, vi, mods = ~ factor(alloc) - 1, data=dat) res if (require(multcomp)) summary(glht(res, linfct=contrMat(c("alternate"=1,"random"=1,"systematic"=1), type="Tukey")), test=adjusted("holm")) ### subgrouping versus using a single model with a factor (subgrouping provides ### an estimate of tau^2 within each subgroup, but the number of studies in each ### subgroup is quite small; the model with the allocation factor provides a ### single estimate of tau^2 based on a larger number of studies, but assumes ### that tau^2 is the same within each subgroup) res.a <- rma(yi, vi, data=dat, subset=(alloc=="alternate")) res.r <- rma(yi, vi, data=dat, subset=(alloc=="random")) res.s <- rma(yi, vi, data=dat, subset=(alloc=="systematic")) res.a res.r res.s res <- rma(yi, vi, mods = ~ factor(alloc) - 1, data=dat) res ############################################################################ ### demonstrating that Q_E + Q_M = Q_Total for fixed-effects models ### note: this does not work for random/mixed-effects models, since Q_E and ### Q_Total are calculated under the assumption that tau^2 = 0, while the ### calculation of Q_M incorporates the estimate of tau^2 res <- rma(yi, vi, data=dat, method="FE") res ### this gives Q_Total res <- rma(yi, vi, mods = ~ ablat + year, data=dat, method="FE") res ### this gives Q_E and Q_M res$QE + res$QM ### decomposition of Q_E into subgroup Q-values res <- rma(yi, vi, mods = ~ factor(alloc), data=dat) res res.a <- rma(yi, vi, data=dat, subset=(alloc=="alternate")) res.r <- rma(yi, vi, data=dat, subset=(alloc=="random")) res.s <- rma(yi, vi, data=dat, subset=(alloc=="systematic")) res.a$QE ### Q-value within subgroup "alternate" res.r$QE ### Q-value within subgroup "random" res.s$QE ### Q-value within subgroup "systematic" res$QE res.a$QE + res.r$QE + res.s$QE ############################################################################ ### an example of a location-scale model dat <- dat.bangertdrowns2004 ### fit a standard random-effects model res <- rma(yi, vi, data=dat) res ### fit the same model as a location-scale model res <- rma(yi, vi, scale = ~ 1, data=dat) res ### check that we obtain the same estimate for tau^2 predict(res, newscale=1, transf=exp) ### add the total sample size (per 100) as a location and scale predictor dat$ni100 <- dat\$ni/100 res <- rma(yi, vi, mods = ~ ni100, scale = ~ ni100, data=dat) res ### variables in the location and scale parts can differ res <- rma(yi, vi, mods = ~ ni100 + meta, scale = ~ ni100 + imag, data=dat) res 

metafor documentation built on June 9, 2021, 5:08 p.m.