Note that when t2 = 0, i.e. where the heterogeneity statistic Q is as small as or smaller than its degrees of freedom (k - 1), the weights reduce to those given by the inverse variance method.
If the estimate of t2 is greater than zero then the weights in random-effects models (w' = 1/(SE(^)2 + t2)) will be smaller and more similar to each other than the weights in fixed effect models (w. = 1/SE(^)2). This means than random-effects meta-analyses will be more conservative (the confidence intervals will be wider) than fixed effect analyses22 since the variance of the pooled effect is the inverse of the sum of the weights. It also means that random effects models give relatively more weight to smaller studies than the fixed effect model. This may not always be desirable (see Chapter 11).
The DerSimonian and Laird method has the same wide applicability as the inverse variance method, and can be used to combine any type of estimates provided standard errors are available (see Stata commands meta and metan in Chapter 18).
Was this article helpful?