TY - JOUR

T1 - Erratum to

T2 - Robust variance estimation in meta-regression with dependent effect size estimates (Research Synthesis Methods, (2010), 1, 1, (39-65), 10.1002/jrsm.5)

AU - Hedges, Larry V.

AU - Tipton, Laura Elizabeth

AU - Johnson, Matthew C.

N1 - Publisher Copyright:
Copyright © 2010 John Wiley & Sons, Ltd.

PY - 2010/4/1

Y1 - 2010/4/1

N2 - On pages 47–48, under the heading ‘Dependence Introduced by Study Level Effects’, the estimators of τ2 and ω2 given are incorrect. The correct estimators are as follows. Assume a covariance matrix of the form Σ=diag(Σ1, …, Σm), where (Figure presented.) where τ2 is the random between cluster variance, ω2 is the random within cluster between study variance, Ij is a kj × kj identity matrix, Jj is a kj × kj matrix of 1's, and Vj is the diagonal matrix of study level known variances. We let QE be the sum of squares provided in our original paper. That is, (Figure presented.) where Wj=V (Formula presented.). An additional sum of squares can be written (Figure presented.) where. Method of moments estimators of the parameters τ2 and ω2 can therefore be written, (Figure presented.) and (Figure presented.) where (Figure presented.) and p=rank(X), kj is the number of studies in cluster j and V=(X′WX)−1. As suggested in the paper, when either or are negative, their estimators are set equal to 0. Finally, please note that the R functions given in Appendix B contain several typesetting errors (which unfortunately we did not catch). We also did not include an R function for estimating the standard errors in a hierarchical model. In order to avoid new typographical errors, we have included correct (and updated) versions of both functions on our website, http://www.northwestern.edu/ipr/qcenter/RVE-meta-analysis.html.

AB - On pages 47–48, under the heading ‘Dependence Introduced by Study Level Effects’, the estimators of τ2 and ω2 given are incorrect. The correct estimators are as follows. Assume a covariance matrix of the form Σ=diag(Σ1, …, Σm), where (Figure presented.) where τ2 is the random between cluster variance, ω2 is the random within cluster between study variance, Ij is a kj × kj identity matrix, Jj is a kj × kj matrix of 1's, and Vj is the diagonal matrix of study level known variances. We let QE be the sum of squares provided in our original paper. That is, (Figure presented.) where Wj=V (Formula presented.). An additional sum of squares can be written (Figure presented.) where. Method of moments estimators of the parameters τ2 and ω2 can therefore be written, (Figure presented.) and (Figure presented.) where (Figure presented.) and p=rank(X), kj is the number of studies in cluster j and V=(X′WX)−1. As suggested in the paper, when either or are negative, their estimators are set equal to 0. Finally, please note that the R functions given in Appendix B contain several typesetting errors (which unfortunately we did not catch). We also did not include an R function for estimating the standard errors in a hierarchical model. In order to avoid new typographical errors, we have included correct (and updated) versions of both functions on our website, http://www.northwestern.edu/ipr/qcenter/RVE-meta-analysis.html.

UR - http://www.scopus.com/inward/record.url?scp=85016889356&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85016889356&partnerID=8YFLogxK

U2 - 10.1002/jrsm.17

DO - 10.1002/jrsm.17

M3 - Comment/debate

C2 - 26061381

AN - SCOPUS:85016889356

SN - 1759-2879

VL - 1

SP - 164

EP - 165

JO - Research synthesis methods

JF - Research synthesis methods

IS - 2

ER -