`rma.glmm.Rd`

Function to fit meta-analytic equal-, fixed-, and random-effects models and (mixed-effects) meta-regression models using a generalized linear (mixed-effects) model framework. See below and the introduction to the metafor-package for more details on these models.

```
rma.glmm(ai, bi, ci, di, n1i, n2i, x1i, x2i, t1i, t2i, xi, mi, ti, ni,
mods, measure, intercept=TRUE, data, slab, subset,
add=1/2, to="only0", drop00=TRUE, vtype="LS",
model="UM.FS", method="ML", coding=1/2, cor=FALSE, test="z",
level=95, digits, btt, nAGQ=7, verbose=FALSE, control, ...)
```

- 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.- 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.- ti
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 \(k\) specifying the values of the moderator. Multiple moderators are specified by giving a matrix with \(k\) rows and as many columns as there are moderator variables. Alternatively, a model

`formula`

can be used to specify the model. See ‘Details’.- measure
character string to specify the outcome measure to use for the meta-analysis. Possible options are the odds ratio (

`"OR"`

), the incidence rate ratio (`"IRR"`

), the logit transformed proportion (`"PLO"`

), or the log transformed incidence rate (`"IRLN"`

).- intercept
logical to specify whether an intercept should be added to the model (the default is

`TRUE`

).- data
optional data frame containing the data supplied to the function.

- slab
optional vector with labels for the \(k\) studies.

- subset
optional (logical or numeric) vector to specify the subset of studies that should be used for the analysis.

- add
non-negative number to specify the amount to add to zero cells, counts, or frequencies when calculating the observed effect sizes or outcomes of the individual studies. See below and the documentation of the

`escalc`

function for more details.- to
character string to specify when the values under

`add`

should be added (either`"only0"`

,`"all"`

,`"if0all"`

, or`"none"`

). See below and the documentation of the`escalc`

function for more details.- drop00
logical to specify whether studies with no cases/events (or only cases) in both groups should be dropped. See the documentation of the

`escalc`

function for more details.- vtype
character string to specify the type of sampling variances to calculate when calculating the observed effect sizes or outcomes. See the documentation of the

`escalc`

function for more details.- model
character string to specify the general model type to use for the analysis (either

`"UM.FS"`

(the default),`"UM.RS"`

,`"CM.EL"`

, or`"CM.AL"`

). See ‘Details’.- method
character string to specify whether an equal- or a random-effects model should be fitted. An equal-effects model is fitted when using

`method="EE"`

. A random-effects model is fitted by setting`method="ML"`

(the default). See ‘Details’.- coding
numeric scalar to indicate how the group variable should be coded in the random effects structure for random/mixed-effects models (the default is

`1/2`

). See ‘Note’.- cor
logical to indicate whether the random study effects should be allowed to be correlated with the random group effects for random/mixed-effects models when

`model="UM.RS"`

(the default is`FALSE`

). See ‘Note’.- 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. See ‘Details’.- level
numeric value between 0 and 100 to specify the confidence interval level (the default is 95).

- digits
optional integer to specify the number of decimal places to which the printed results should be rounded. If unspecified, the default is 4. See also here for further details on how to control the number of digits in the output.

- 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’.- nAGQ
positive integer to specify the number of points per axis for evaluating the adaptive Gauss-Hermite approximation to the log-likelihood. The default is 7. Setting this to 1 corresponds to the Laplacian approximation. See ‘Note’.

- 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 estimation algorithms. If unspecified, default values are defined inside the function. See ‘Note’.

- ...
additional arguments.

The function can be used in combination with the following effect size or outcome measures:

`measure="OR"`

for odds ratios (analyzed in log units)`measure="IRR"`

for incidence rate ratios (analyzed in log units)`measure="PLO"`

for logit transformed proportions (i.e., log odds)`measure="IRLN"`

for log transformed incidence rates.

The `escalc`

function describes the data/arguments that should be specified/used for these measures.

A variety of model types are available when analyzing \(2 \times 2\) table data (i.e., when `measure="OR"`

) or two-group event count data (i.e., when `measure="IRR"`

):

`model="UM.FS"`

for an unconditional generalized linear mixed-effects model with fixed study effects`model="UM.RS"`

for an unconditional generalized linear mixed-effects model with random study effects`model="CM.AL"`

for a conditional generalized linear mixed-effects model (approximate likelihood)`model="CM.EL"`

for a conditional generalized linear mixed-effects model (exact likelihood).

For `measure="OR"`

, models `"UM.FS"`

and `"UM.RS"`

are essentially (mixed-effects) logistic regression models, while for `measure="IRR"`

, these models are (mixed-effects) Poisson regression models. A choice must be made on how to model study level variability (i.e., differences in outcomes across studies irrespective of group membership). One can choose between using fixed study effects (which means that \(k\) dummy variables are added to the model) or random study effects (which means that random effects corresponding to the levels of the study factor are added to the model).

The conditional model (`model="CM.EL"`

) avoids having to model study level variability by conditioning on the total numbers of cases/events in each study. For `measure="OR"`

, this leads to a non-central hypergeometric distribution for the data within each study and the corresponding model is then a (mixed-effects) conditional logistic model. Fitting this model can be difficult and computationally expensive. When the number of cases in each study is small relative to the group sizes, one can approximate the exact likelihood by a binomial distribution, which leads to a regular (mixed-effects) logistic regression model (`model="CM.AL"`

). For `measure="IRR"`

, the conditional model leads directly to a binomial distribution for the data within each study and the resulting model is again a (mixed-effects) logistic regression model (no approximate likelihood model is needed here).

When analyzing proportions (i.e., `measure="PLO"`

) or incidence rates (i.e., `measure="IRLN"`

) of individual groups, the model type is always a (mixed-effects) logistic or Poisson regression model, respectively (i.e., the `model`

argument is not relevant here).

Aside from choosing the general model type, one has to decide whether to fit an equal- or a random-effects model to the data. An *equal-effects model* is fitted by setting `method="EE"`

. A *random-effects model* is fitted by setting `method="ML"`

(the default). Note that random-effects models with dichotomous data are often referred to as ‘binomial-normal’ models in the meta-analytic literature. Analogously, for event count data, such models could be referred to as ‘Poisson-normal’ models.

One or more moderators can be included in a model via the `mods`

argument. A single moderator can be given as a (row or column) vector of length \(k\) specifying the values of the moderator. Multiple moderators are specified by giving an appropriate model matrix (i.e., \(X\)) with \(k\) 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 fitting a particular model, actually up to three different models are fitted within the function:

the equal-effects model (i.e., where \(\tau^2\) is set to 0),

the saturated model (i.e., the model with a deviance of 0), and

the random/mixed-effects model (i.e., where \(\tau^2\) is estimated) (only if

`method="ML"`

).

The saturated model is obtained by adding as many dummy variables to the model as needed so that the model deviance is equal to zero. Even when `method="ML"`

, the equal- and saturated models are also fitted, as they are used to compute the test statistics for the Wald-type and likelihood ratio tests for (residual) heterogeneity (see below).

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 \(Q_M\)-test and follows, under the assumptions of the model, a chi-square distribution with \(m\) degrees of freedom (with \(m\) denoting the number of coefficients tested) under the null hypothesis (that the true value of all coefficients tested is equal to 0).

Categorical moderator variables can be included in the model via the `mods`

argument in the same way that appropriately (dummy) coded categorical 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 autoamte the coding (note that string/character variables in a model formula are automatically converted to factors).

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 \(k-p\) degrees of freedom, while the omnibus test statistic then uses an F-distribution with \(m\) and \(k-p\) degrees of freedom (with \(k\) denoting the total number of estimates included in the analysis and \(p\) the total number of model coefficients including the intercept if it is present). Note that `test="t"`

is not the same as `test="knha"`

in `rma.uni`

, as no adjustment to the standard errors of the estimated coefficients is made.

Two different tests for (residual) heterogeneity are automatically carried out by the function. The first is a Wald-type test, which tests the coefficients corresponding to the dummy variables added in the saturated model for significance. The second is a likelihood ratio test, which tests the same set of coefficients, but does so by computing \(-2\) times the difference in the log-likelihoods of the equal-effects and the saturated models. These two tests are not identical for the types of models fitted by the `rma.glmm`

function and may even lead to conflicting conclusions.

The various models do not require the calculation of the observed effect sizes or outcomes of the individual studies (e.g., the observed odds ratios of the \(k\) studies) and directly make use of the table/event counts. Zero cells/events are not a problem (except in extreme cases, such as when one of the two outcomes never occurs or when there are no events in any of the studies). Therefore, it is unnecessary to add some constant to the cell/event counts when there are zero cells/events.

However, for plotting and various other functions, it is necessary to calculate the observed effect sizes or outcomes for the \(k\) studies. Here, zero cells/events can be problematic, so adding a constant value to the cell/event counts ensures that all \(k\) values can be calculated. The `add`

and `to`

arguments are used to specify what value should be added to the cell/event counts and under what circumstances when calculating the observed effect sizes or outcomes. The documentation of the `escalc`

function explains how the `add`

and `to`

arguments work. Note that `drop00`

is set to `TRUE`

by default, since studies where `ai=ci=0`

or `bi=di=0`

or studies where `x1i=x2i=0`

are uninformative about the size of the effect.

An object of class `c("rma.glmm","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="EE"`

.- sigma2
estimated amount of study level variability (only for

`model="UM.RS"`

).- 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.Wld
Wald-type test statistic of the test for (residual) heterogeneity.

- QEp.Wld
corresponding p-value.

- QE.LRT
likelihood ratio test statistic of the test for (residual) heterogeneity.

- QEp.LRT
corresponding p-value.

- QM
test statistic of the omnibus test of moderators.

- QMp
corresponding p-value.

- I2
value of \(I^2\).

- H2
value of \(H^2\).

- 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.

- ...
some additional elements/values.

The results of the fitted model are formatted and printed with the `print`

function. If fit statistics should also be given, use `summary`

(or use the `fitstats`

function to extract them).

When `measure="OR"`

or `measure="IRR"`

, `model="UM.FS"`

or `model="UM.RS"`

, and `method="ML"`

, one has to choose a coding scheme for the group variable in the random effects structure. When `code=1/2`

(the default), the two groups are coded with `+1/2`

and `-1/2`

(i.e., contrast coding), which is invariant under group label switching.

When `code=1`

, the first group is coded with `1`

and the second group with `0`

. Finally, when `code=0`

, the first group is coded with `0`

and the second group with `1`

. Note that these coding schemes are not invariant under group label switching.

Also, when `model="UM.RS"`

and `method="ML"`

, one has to decide whether the random study effects are allowed to be correlated with the random group effects. By default (i.e., when `cor=FALSE`

), no such correlation is allowed (which is typically an appropriate assumption when `code=1/2`

). When using a different coding scheme for the group variable (i.e., `code=1`

or `code=0`

), allowing the random study and group effects to be correlated (i.e., using `cor=TRUE`

) is usually recommended.

Fitting the various types of models requires several different iterative algorithms:

For

`model="UM.FS"`

and`model="CM.AL"`

, iteratively reweighted least squares (IWLS) as implemented in the`glm`

function is used for fitting the equal-effects and the saturated models. For`method="ML"`

, adaptive Gauss-Hermite quadrature as implemented in the`glmer`

function is used. The same applies when`model="CM.EL"`

is used in combination with`measure="IRR"`

or when`measure="PLO"`

or`measure="IRLN"`

(regardless of the model type).For

`model="UM.RS"`

, adaptive Gauss-Hermite quadrature as implemented in the`glmer`

function is used to fit all of the models.For

`model="CM.EL"`

and`measure="OR"`

, the quasi-Newton method optimizer as implemented in the`nlminb`

function is used by default for fitting the equal-effects and the saturated models. For`method="ML"`

, the same algorithm is used, together with adaptive quadrature as implemented in the`integrate`

function (for the integration over the density of the non-central hypergeometric distribution). Standard errors of the parameter estimates are obtained by inverting the Hessian, which is numerically approximated using the`hessian`

function from the`numDeriv`

package.One can also chose a different optimizer via the

`control`

argument (e.g.,`control=list(optimizer="optim")`

). When using`optim`

, one can set the particular method via the`optmethod`

argument (e.g.,`control=list(optimizer="optim", optmethod="BFGS")`

). Besides`nlminb`

and`optim`

, one can also choose one of the optimizers from the`minqa`

package (i.e.,`uobyqa`

,`newuoa`

, or`bobyqa`

), one of the (derivative-free) algorithms from the`nloptr`

package, the Newton-type algorithm implemented in`nlm`

, the various algorithms implemented in the`dfoptim`

package (`hjk`

for the Hooke-Jeeves,`nmk`

for the Nelder-Mead, and`mads`

for the Mesh Adaptive Direct Searches (MADS) algorithm), the quasi-Newton type optimizers`ucminf`

and`lbfgsb3c`

and the subspace-searching simplex algorithm`subplex`

from the packages of the same name, the Barzilai-Borwein gradient method decent implemented in`BBoptim`

, or the parallelized version of the L-BFGS-B algorithm implemented in`optimParallel`

from the package of the same name.The optimizer name must be given as a character string (i.e., in quotes). Additional control parameters can be specified via the

`optCtrl`

elements of the`control`

argument (e.g.,`control=list(optCtrl=list(iter.max=1000, rel.tol=1e-8))`

). For`nloptr`

, the default is to use the BOBYQA implementation from that package with a relative convergence criterion of`1e-8`

on the function value (i.e., log-likelihood), but this can be changed via the`algorithm`

and`ftop_rel`

arguments (e.g.,`control=list(optimizer="nloptr", optCtrl=list(algorithm="NLOPT_LN_SBPLX", ftol_rel=1e-6))`

). For`optimParallel`

, the control argument`ncpus`

can be used to specify the number of cores to use for the parallelization (e.g.,`control=list(optimizer="optimParallel", ncpus=2)`

). With`parallel::detectCores()`

, one can check on the number of available cores on the local machine.

When `model="CM.EL"`

and `measure="OR"`

, actually `model="CM.AL"`

is used first to obtain starting values for `optim`

, so either 4 (if `method="EE"`

) or 6 (if `method="ML"`

) models need to be fitted in total.

Various additional control parameters can be adjusted via the `control`

argument:

`glmCtrl`

is a list of named arguments to be passed on to the`control`

argument of the`glm`

function,`glmerCtrl`

is a list of named arguments to be passed on to the`control`

argument of the`glmer`

function, and`intCtrl`

is a list of named arguments (i.e.,`rel.tol`

and`subdivisions`

) to be passed on to the`integrate`

function.`hessianCtrl`

is a list of named arguments to be passed on to the`method.args`

argument of the`hessian`

function. For some borderline cases, it may be necessary to bump up the`r`

argument to a higher number to get sufficient accuracy when approximating the Hessian numerically (the default is`control=list(hessianCtrl=list(r=16))`

).

Also, for `glmer`

, the `nAGQ`

argument is used to specify the number of quadrature points. The default value is 7, which should provide sufficient accuracy in the evaluation of the log-likelihood in most cases, but at the expense of speed. Setting this to 1 corresponds to the Laplacian approximation (which is faster, but less accurate).

Information on the progress of the various algorithms can be obtained by setting `verbose=TRUE`

. Since fitting the various models can be computationally expensive, this option is useful to determine how the model fitting is progressing. One can also set `verbose`

to an integer (`verbose=2`

yields even more information and `verbose=3`

also sets `option(warn=1)`

temporarily).

For `model="CM.EL"`

and `measure="OR"`

, optimization involves repeated calculation of the density of the non-central hypergeometric distribution. When `method="ML"`

, this also requires integration over the same density. This is currently implemented in a rather brute-force manner and may not be numerically stable, especially when models with moderators are fitted. Stability can be improved by scaling the moderators in a similar manner (i.e., don't use a moderator that is coded 0 and 1, while another uses values in the 1000s). For models with an intercept and moderators, the function actually rescales (non-dummy) variables to z-scores during the model fitting (results are given after back-scaling, so this should be transparent to the user). For models without an intercept, this is not done, so sensitivity analyses are highly recommended here (to ensure that the results do not depend on the scaling of the moderators).

Agresti, A. (2002). *Categorical data analysis* (2nd. ed). Hoboken, NJ: Wiley.

Bagos, P. G., & Nikolopoulos, G. K. (2009). Mixed-effects Poisson regression models for meta-analysis of follow-up studies with constant or varying durations. *The International Journal of Biostatistics*, **5**(1). https://doi.org/10.2202/1557-4679.1168

van Houwelingen, H. C., Zwinderman, K. H., & Stijnen, T. (1993). A bivariate approach to meta-analysis. *Statistics in Medicine*, **12**(24), 2273–2284. https://doi.org/10.1002/sim.4780122405

Jackson, D., Law, M., Stijnen, T., Viechtbauer, W., & White, I. R. (2018). A comparison of seven random-effects models for meta-analyses that estimate the summary odds ratio. *Statistics in Medicine*, **37**(7), 1059-1085. https://doi.org/10.1002/sim.7588

Liao, J. G., & Rosen, O. (2001). Fast and stable algorithms for computing and sampling from the noncentral hypergeometric distribution. *American Statistician*, **55**(4), 366–369. https://doi.org/10.1198/000313001753272547

Simmonds, M. C., & Higgins, J. P. T. (2016). A general framework for the use of logistic regression models in meta-analysis. *Statistical Methods in Medical Research*, **25**(6), 2858–2877. https://doi.org/10.1177/0962280214534409

Stijnen, T., Hamza, T. H., & Ozdemir, P. (2010). Random effects meta-analysis of event outcome in the framework of the generalized linear mixed model with applications in sparse data. *Statistics in Medicine*, **29**(29), 3046–3067. https://doi.org/10.1002/sim.4040

Turner, R. M., Omar, R. Z., Yang, M., Goldstein, H., & Thompson, S. G. (2000). A multilevel model framework for meta-analysis of clinical trials with binary outcomes. *Statistics in Medicine*, **19**(24), 3417–3432. https://doi.org/10.1002/1097-0258(20001230)19:24<3417::aid-sim614>3.0.co;2-l

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

`rma.uni`

, `rma.mh`

, `rma.peto`

, and `rma.mv`

for other model fitting functions.

`dat.nielweise2007`

, `dat.nielweise2008`

, `dat.collins1985a`

, and `dat.pritz1997`

for further examples of the use of the `rma.glmm`

function.

```
### random-effects model using rma.uni() (standard RE model analysis)
rma(measure="OR", ai=tpos, bi=tneg, ci=cpos, di=cneg, data=dat.bcg, method="ML")
#>
#> Random-Effects Model (k = 13; tau^2 estimator: ML)
#>
#> tau^2 (estimated amount of total heterogeneity): 0.3025 (SE = 0.1549)
#> tau (square root of estimated tau^2 value): 0.5500
#> I^2 (total heterogeneity / total variability): 91.23%
#> H^2 (total variability / sampling variability): 11.40
#>
#> Test for Heterogeneity:
#> Q(df = 12) = 163.1649, p-val < .0001
#>
#> Model Results:
#>
#> estimate se zval pval ci.lb ci.ub
#> -0.7420 0.1780 -4.1694 <.0001 -1.0907 -0.3932 ***
#>
#> ---
#> Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
#>
### random-effects models using rma.glmm() (require 'lme4' package)
### unconditional model with fixed study effects
# \dontrun{
rma.glmm(measure="OR", ai=tpos, bi=tneg, ci=cpos, di=cneg, data=dat.bcg, model="UM.FS")# }
#>
#> Random-Effects Model (k = 13; tau^2 estimator: ML)
#> Model Type: Unconditional Model with Fixed Study Effects
#>
#> tau^2 (estimated amount of total heterogeneity): 0.2949
#> tau (square root of estimated tau^2 value): 0.5430
#> I^2 (total heterogeneity / total variability): 91.0235%
#> H^2 (total variability / sampling variability): 11.1402
#>
#> Tests for Heterogeneity:
#> Wld(df = 12) = 163.1649, p-val < .0001
#> LRT(df = 12) = 176.9544, p-val < .0001
#>
#> Model Results:
#>
#> estimate se zval pval ci.lb ci.ub
#> -0.7450 0.1756 -4.2435 <.0001 -1.0891 -0.4009 ***
#>
#> ---
#> Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
#>
### unconditional model with random study effects
# \dontrun{
rma.glmm(measure="OR", ai=tpos, bi=tneg, ci=cpos, di=cneg, data=dat.bcg, model="UM.RS")# }
#> Warning: Currently not possible to fit RE/ME model='UM.RS' with nAGQ > 1. nAGQ automatically set to 1.
#>
#> Random-Effects Model (k = 13; tau^2 estimator: ML)
#> Model Type: Unconditional Model with Random Study Effects
#>
#> tau^2 (estimated amount of total heterogeneity): 0.3198
#> tau (square root of estimated tau^2 value): 0.5655
#> I^2 (total heterogeneity / total variability): 91.6652%
#> H^2 (total variability / sampling variability): 11.9979
#>
#> sigma^2 (estimated amount of study level variability): 1.8616
#> sigma (square root of estimated sigma^2 value): 1.3644
#>
#> Tests for Heterogeneity:
#> Wld(df = 12) = 161.1955, p-val < .0001
#> LRT(df = 12) = 174.1317, p-val < .0001
#>
#> Model Results:
#>
#> estimate se zval pval ci.lb ci.ub
#> -0.7616 0.1815 -4.1969 <.0001 -1.1172 -0.4059 ***
#>
#> ---
#> Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
#>
### conditional model with approximate likelihood
# \dontrun{
rma.glmm(measure="OR", ai=tpos, bi=tneg, ci=cpos, di=cneg, data=dat.bcg, model="CM.AL")# }
#>
#> Random-Effects Model (k = 13; tau^2 estimator: ML)
#> Model Type: Conditional Model with Approximate Likelihood
#>
#> tau^2 (estimated amount of total heterogeneity): 0.2887
#> tau (square root of estimated tau^2 value): 0.5373
#> I^2 (total heterogeneity / total variability): 90.8488%
#> H^2 (total variability / sampling variability): 10.9276
#>
#> Tests for Heterogeneity:
#> Wld(df = 12) = 147.9061, p-val < .0001
#> LRT(df = 12) = 161.8210, p-val < .0001
#>
#> Model Results:
#>
#> estimate se zval pval ci.lb ci.ub
#> -0.7221 0.1744 -4.1405 <.0001 -1.0639 -0.3803 ***
#>
#> ---
#> Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
#>
### conditional model with exact likelihood
### note: fitting this model may take a bit of time, so be patient
# \dontrun{
rma.glmm(measure="OR", ai=tpos, bi=tneg, ci=cpos, di=cneg, data=dat.bcg, model="CM.EL")# }
#>
#> Random-Effects Model (k = 13; tau^2 estimator: ML)
#> Model Type: Conditional Model with Exact Likelihood
#>
#> tau^2 (estimated amount of total heterogeneity): 0.3116 (SE = 0.1612)
#> tau (square root of estimated tau^2 value): 0.5582
#> I^2 (total heterogeneity / total variability): 91.4646%
#> H^2 (total variability / sampling variability): 11.7159
#>
#> Tests for Heterogeneity:
#> Wld(df = 12) = 268.4283, p-val < .0001
#> LRT(df = 12) = 176.8738, p-val < .0001
#>
#> Model Results:
#>
#> estimate se zval pval ci.lb ci.ub
#> -0.7538 0.1801 -4.1860 <.0001 -1.1068 -0.4009 ***
#>
#> ---
#> Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
#>
```