2
$\begingroup$

I want to calculate the proportion of total variance in a multi-level meta-analysis with 2+ random effect terms according to pp. 1261 & equation (24) in Nakagawa & Santos 2012 https://www.researchgate.net/profile/Shinichi_Nakagawa2/publication/233341316_Methodological_issues_and_advances_in_biological_meta-analysis/links/00b495157aec0585c0000000.pdf

I have the following output from function rma.mv in R package metafor with the model structure

test.meta = rma.mv(d, Var.d., random = list(~1|Study, ~1|Order.class), data=data). 

How can I obtain the total heterogeneity (variance) by which to divide each of the sigma^2 terms?

Multivariate Meta-Analysis Model (k = 274; method: REML)

    logLik    Deviance         AIC         BIC        AICc
-1264.4201   2528.8403   2534.8403   2545.6687   2534.9295  

Variance Components:

            estim    sqrt  nlvls  fixed       factor
sigma^2.1  1.1223  1.0594     65     no        Study
sigma^2.2  0.4060  0.6372      7     no  Order.class

Test for Heterogeneity:

Q(df = 273) = 2660.6043, p-val < .0001

Model Results:

estimate       se     zval     pval    ci.lb    ci.ub          
 -0.6585   0.3125  -2.1074   0.0351  -1.2710  -0.0461        * 

---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 

A similar example with only one random-effect term is here http://www.metafor-project.org/doku.php/tips:rma.uni_vs_rma.mv under "Random effects model."

$\endgroup$
3
$\begingroup$

First, you need to compute what is denoted $\sigma^2_m$, the 'typical' within-study variance, as defined by Higgins and Thompson (2002). You can do this with:

wi <- 1/test.meta$vi
k <- length(wi)
s2m <- (k-1)*sum(wi)/(sum(wi)^2 - sum(wi^2))

Then the total variance, $\sigma^2_t$, as defined in Nakagawa and Santos (2012), can be computed with:

s2t <- s2m + sum(test.meta$sigma2)

And now you can compute $I^2$-like measures with:

test.meta$sigma2 / s2t
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.