Dear all, I am working with the rma.uni function to fit a multivariate random effects model. I see that the W argument is an optional argument to specify a user-defined weight matrix. I was curious if any weight is applied if I don?t specify this argument. Specifically, I am wondering if the inverse variance weight is applied ?automatically?, considering I am specifying the V argument in the model itself. In terms of my model, I am wondering about the difference between these: res5 <- rma.mv(EffectSize_NEW, Variance_New, method="REML",random = list(~ 1 | EffectSize_ID_NEW, ~ 1 | Study), data=Data) res5 res6 <- rma.mv(EffectSize_NEW, Variance_New, W=weight, method="REML",random = list(~ 1 | EffectSize_ID_NEW, ~ 1 | Study), data=Data) res6 Note: the argument weight in my dataset is simply: 1/Variance_New Hope this is all clear. Many thanks and kind regards, Tina
[R-meta] Weight argument in rma.uni models
4 messages · Tina Dudenhöffer, Wolfgang Viechtbauer
Dear Tina, I assume you mean the rma.mv() function, not rma.uni(). If you don't specify W, then weights are applied automatically. See: https://wviechtb.github.io/metafor/reference/rma.mv.html and search for "Together with the variance-covariance matrix of the sampling errors". However, for the model you show, the model implied marginal variance-covariance matrix of the observed outcomes (M) is not diagonal, but also has off-diagonal elements. So, there is actually an entire weight matrix. Also, in 'res5', M (and hence the weight matrix W = M^(-1)) will incorporate the estimates of the two variance components (for EffectSize_ID_NEW and Study), while 'res6' only uses 1/Variance_New as the weights and doesn't consider the two variance components. Best, Wolfgang -----Original Message----- From: R-sig-meta-analysis [mailto:r-sig-meta-analysis-bounces at r-project.org] On Behalf Of Tina Dudenh?ffer Sent: Wednesday, 02 October, 2019 13:48 To: r-sig-meta-analysis at r-project.org Subject: [R-meta] Weight argument in rma.uni models Dear all, I am working with the rma.uni function to fit a multivariate random effects model. I see that the W argument is an optional argument to specify a user-defined weight matrix. I was curious if any weight is applied if I don?t specify this argument. Specifically, I am wondering if the inverse variance weight is applied ?automatically?, considering I am specifying the V argument in the model itself. In terms of my model, I am wondering about the difference between these: res5 <- rma.mv(EffectSize_NEW, Variance_New, method="REML",random = list(~ 1 | EffectSize_ID_NEW, ~ 1 | Study), data=Data) res5 res6 <- rma.mv(EffectSize_NEW, Variance_New, W=weight, method="REML",random = list(~ 1 | EffectSize_ID_NEW, ~ 1 | Study), data=Data) res6 Note: the argument weight in my dataset is simply: 1/Variance_New Hope this is all clear. Many thanks and kind regards, Tina
Dear Wolfgang, thanks for your quick response. I, indeed, mean rma.mv()! Considering the default is the inverse variance weight, would you suggest I stick to my res5 version?
res5 <- rma.mv(EffectSize_NEW, Variance_New, method="REML",random = list(~ 1 | EffectSize_ID_NEW, ~ 1 | Study), data=Data) res5
I saw somewhere else the argument ?weighted=TRUE? added - but when looking the documentation, I don?t see it. I suppose it?s not necessary or at least not in my case? Thank you so much! Tina
On Oct 2, 2019, at 2:49 PM, Viechtbauer, Wolfgang (SP) <wolfgang.viechtbauer at maastrichtuniversity.nl> wrote: Dear Tina, I assume you mean the rma.mv() function, not rma.uni(). If you don't specify W, then weights are applied automatically. See: https://wviechtb.github.io/metafor/reference/rma.mv.html and search for "Together with the variance-covariance matrix of the sampling errors". However, for the model you show, the model implied marginal variance-covariance matrix of the observed outcomes (M) is not diagonal, but also has off-diagonal elements. So, there is actually an entire weight matrix. Also, in 'res5', M (and hence the weight matrix W = M^(-1)) will incorporate the estimates of the two variance components (for EffectSize_ID_NEW and Study), while 'res6' only uses 1/Variance_New as the weights and doesn't consider the two variance components. Best, Wolfgang -----Original Message----- From: R-sig-meta-analysis [mailto:r-sig-meta-analysis-bounces at r-project.org] On Behalf Of Tina Dudenh?ffer Sent: Wednesday, 02 October, 2019 13:48 To: r-sig-meta-analysis at r-project.org Subject: [R-meta] Weight argument in rma.uni models Dear all, I am working with the rma.uni function to fit a multivariate random effects model. I see that the W argument is an optional argument to specify a user-defined weight matrix. I was curious if any weight is applied if I don?t specify this argument. Specifically, I am wondering if the inverse variance weight is applied ?automatically?, considering I am specifying the V argument in the model itself. In terms of my model, I am wondering about the difference between these: res5 <- rma.mv(EffectSize_NEW, Variance_New, method="REML",random = list(~ 1 | EffectSize_ID_NEW, ~ 1 | Study), data=Data) res5 res6 <- rma.mv(EffectSize_NEW, Variance_New, W=weight, method="REML",random = list(~ 1 | EffectSize_ID_NEW, ~ 1 | Study), data=Data) res6 Note: the argument weight in my dataset is simply: 1/Variance_New Hope this is all clear. Many thanks and kind regards, Tina
rma.uni() has a 'weighted' argument (which is TRUE by default): https://wviechtb.github.io/metafor/reference/rma.uni.html Best, Wolfgang -----Original Message----- From: Tina Dudenh?ffer [mailto:tina.dudenhoeffer at gmail.com] Sent: Wednesday, 02 October, 2019 15:00 To: Viechtbauer, Wolfgang (SP) Cc: r-sig-meta-analysis at r-project.org Subject: Re: [R-meta] Weight argument in rma.uni models Dear Wolfgang, thanks for your quick response. I, indeed, mean rma.mv()! Considering the default is the inverse variance weight, would you suggest I stick to my res5 version?
res5 <- rma.mv(EffectSize_NEW, Variance_New, method="REML",random = list(~ 1 | EffectSize_ID_NEW, ~ 1 | Study), data=Data) res5
I saw somewhere else the argument ?weighted=TRUE? added - but when looking the documentation, I don?t see it. I suppose it?s not necessary or at least not in my case? Thank you so much! Tina
On Oct 2, 2019, at 2:49 PM, Viechtbauer, Wolfgang (SP) <wolfgang.viechtbauer at maastrichtuniversity.nl> wrote: Dear Tina, I assume you mean the rma.mv() function, not rma.uni(). If you don't specify W, then weights are applied automatically. See: https://wviechtb.github.io/metafor/reference/rma.mv.html and search for "Together with the variance-covariance matrix of the sampling errors". However, for the model you show, the model implied marginal variance-covariance matrix of the observed outcomes (M) is not diagonal, but also has off-diagonal elements. So, there is actually an entire weight matrix. Also, in 'res5', M (and hence the weight matrix W = M^(-1)) will incorporate the estimates of the two variance components (for EffectSize_ID_NEW and Study), while 'res6' only uses 1/Variance_New as the weights and doesn't consider the two variance components. Best, Wolfgang -----Original Message----- From: R-sig-meta-analysis [mailto:r-sig-meta-analysis-bounces at r-project.org] On Behalf Of Tina Dudenh?ffer Sent: Wednesday, 02 October, 2019 13:48 To: r-sig-meta-analysis at r-project.org Subject: [R-meta] Weight argument in rma.uni models Dear all, I am working with the rma.uni function to fit a multivariate random effects model. I see that the W argument is an optional argument to specify a user-defined weight matrix. I was curious if any weight is applied if I don?t specify this argument. Specifically, I am wondering if the inverse variance weight is applied ?automatically?, considering I am specifying the V argument in the model itself. In terms of my model, I am wondering about the difference between these: res5 <- rma.mv(EffectSize_NEW, Variance_New, method="REML",random = list(~ 1 | EffectSize_ID_NEW, ~ 1 | Study), data=Data) res5 res6 <- rma.mv(EffectSize_NEW, Variance_New, W=weight, method="REML",random = list(~ 1 | EffectSize_ID_NEW, ~ 1 | Study), data=Data) res6 Note: the argument weight in my dataset is simply: 1/Variance_New Hope this is all clear. Many thanks and kind regards, Tina