Closed coraschefft closed 4 years ago
The standard error specified for the reference arm must be smaller than any of the standard errors for the differences specified within the study.
The standard error of the reference arm squared is the Covariance of each pair of treatment differences, that is Var{A,B} = Var{A} + Var{B} Var{A,C} = Var{A} + Var{C} Quite often if Var_{A} is not known/given, it is imputed by assuming a correlation (usually between 0.3 and 0.7) - you should be able to find examples of this in the literature, e.g. by Salanti.
Thank you, just now saw your reply on this account. Best Cora
Von meinem iPhone gesendet
Am 07.06.2020 um 15:57 schrieb Gert van Valkenhoef notifications@github.com:
The standard error of the reference arm squared is the Covariance of each pair of treatment differences, that is Var{A,B} = Var{A} + Var{B} Var{A,C} = Var{A} + Var{C} Quite often if Var_{A} is not known/given, it is imputed by assuming a correlation (usually between 0.3 and 0.7) - you should be able to find examples of this in the literature, e.g. by Salanti.
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.
Hey Gert,
I'm trying to run an analysis with 50 two arm studies and 8 three arm studies using SMDs. I created the SMDs and respective std.errors by running the pairwise() function of the package meta, applying Hedges' g and then preparing the data for mtc.network. For baseline standard errors in three arm trials I entered the mean of all standard errors of pairwise comparisons of the multiarm trial. --> is that correct at all? If not how do I estimate the baseline se?
When running mtc.model() I get the warning In sqrt(diag(sigma)) : NaNs produced. --> how to deal with this warning? Apparently there are negative numbers in the diagonal of a var/cov Matrix.....