[R-meta] Query model structure rma.mv
@ybryn@m@e@ @end|ng |rom gm@||@com
Thu Jul 8 16:27:46 CEST 2021
I have three questions regarding a meta-analysis multivariate model
structure I am using for my dataset that is complex in the following way:
There are multiple sites, and within those sites, multiple years of data
can occur. These years are not always the same, so one site can have data
for eg 2003 and 2004, whereas another site could have data only for 2005.
Another one can have data for 2001 to 2010 every year available.
I propose the following structure for my data :
*metamodel <- rma.mv
<http://rma.mv/>(Hedges_SMD, Hedges_SMD_VARIANCE, random=list(~ 1 |
Site_ID / Observation_ID , ~ Year | Site_ID ),*
I am including the second part of the random structure list to account for
potential autocorrelation and non-independence between the different years
within a same site.
However, I have the following questions when looking at the output and when
trying to visualize the results (estimates + CIs) with a simple forest plot:
1) Does the "*forest*" command also work with complex rma.mv structures -
i.e. for making forest plots? I have an outcome and do not get errors, but
I cannot find anywhere
whether it is appropriate to use for these multivariate objects.
2) When including my CAR structure, the *weights *change when I show them
in the forest plot (using "showweights=TRUE" to visualize them). This I
find strange as the sample
sizes are not changing in any way. Why could this be?
3) In order to *diagnostize *whether the non-independence and
autocorrelation is appropriately taken into account, I thought of running a
variogram with and without the CAR structure in the model, and an acf()
command. However, none of these two tools actually shows me that the
autocorrelation is reduced or removed. What could be the reason for this?
Could I find out in another way whether it is a "better" model with the CAR
Thank you for any assistance/advice!
[[alternative HTML version deleted]]
More information about the R-sig-meta-analysis