<html><body><div>Hi Wolfgang,</div><div><br data-mce-bogus="1"></div><div>Thank you for your quick response.</div><div><br data-mce-bogus="1"></div><div>I have never constructed a V matrix. Based on the Berkey tutorial, I have tried to apply the following</div><div></div><div><br data-mce-bogus="1"></div><div>dat$v2i <- dat$NPSD^2 / (dat$NPN*dat$NP^2)<br>V <- bldiag(lapply(split(dat[,c("var", "v2i")], dat$exp), as.matrix))<br>res <-rma.mv(yi=lnR, V=V, random=~1|Site/exp/obs, mods=~Predator, data=dat)<br></div><div><br data-mce-bogus="1"></div><div>But it gives me an error that 'V' must be a square matrix. Is this because the split in V has to be done at the "obs" level? I used "exp" because that indicates the level at which all effect sizes use the same control.</div><div><br data-mce-bogus="1"></div><div>Thank you</div><div>Cesar</div><div><br data-mce-bogus="1"></div><div><br>On March 25, 2020 at 7:44 AM, "Viechtbauer, Wolfgang (SP)" <wolfgang.viechtbauer@maastrichtuniversity.nl> wrote:<br><br></div><div><blockquote type="cite"><div class="msg-quote"><div class="_stretch"><span class="body-text-content">Hi Cesar,<br><br>When the same group is used to compute multiple estimates (i.e., a common control), then this also induces dependency on the sampling errors. For the log response ratio, the covariance is then:<br><br>SD_C^2 / (n_C*M_C^2)<br><br>where M_C and SD_C are the mean and SD of the common control group and n_C the control group size.<br><br>So, you ideally should construct a proper V matrix that includes these covariances and that you can then pass to rma.mv().<br><br>But yes, random=~1|Site/exp/obs would be sensible.<br><br>Best,<br>Wolfgang<br><br>-----Original Message-----<br>From: R-sig-meta-analysis [mailto:r-sig-meta-analysis-bounces@r-project.org] On Behalf Of César Terrer<br>Sent: Wednesday, 25 March, 2020 15:34<br>To: <a href="mailto:r-sig-meta-analysis@r-project.org" data-mce-href="mailto:r-sig-meta-analysis@r-project.org">r-sig-meta-analysis@r-project.org</a><br>Subject: [R-meta] Random effects structure<br><br>Dear community,<br><br>I am conducting a meta-analysis to study the growth rate of bacterial predators as compared to their prey, using the log response ratio. Furthermore, I want to study if this effect varies across different predators. The dataset has the following structure, here showing a subset:<br><br>Site CommonControl exp obs Predator lnR var<br>A Alaska 155 1 1 Bdello -0.6713152 0.03785708<br>A Alaska 155 1 2 Cytoph -0.0702467 0.05763364<br>A Alaska 155 1 3 Myxo -0.148982 0.00748768<br>A Alaska 1510 2 4 Bdello -0.4926361 0.01691187<br>A Alaska 1510 2 5 Cytoph -0.213787 0.01045785<br>B Andesite1controlWeek1 9 6 Bdello 0.27873598 0.14129722<br>B Andesite1controlWeek1 9 7 Cytoph -0.3243682 0.01466085<br>B Andesite1controlWeek1 9 8 Lyso 1.18302506 0.11663149<br>B Andesite1controlWeek6 11 9 Bdello -0.8465128 0.03701618<br>B Andesite1controlWeek6 11 10 Cytoph -0.1559056 0.0283173<br>B Andesite1controlWeek6 11 11 Lyso -0.8039415 0.04926915<br><br>1. There are different sites, thus a potential source of non-independency<br>2. Within each site, we use the value for preys in the denominator multiple times. I guess rows of data using the same denominator (CommonControl) are also potentially correlated and should be also added as a random-effect.<br><br>Based on 1., 2., and what I have understood from the Konstantopoulos (2011) tutorial, I think I should use the following model:<br><br>res <-rma.mv(yi=lnR, V=var, random=~1|Site/exp/obs, mods=~Predator, data=data)<br><br>Could you please let me know if the structure of random effects seems appropriate, and help me understand why I need to include "obs"?<br>Thank you.<br>Cesar<br></span></div></div></blockquote></div></body></html>