Skip to content

[R-meta] potential shortcomings of combining meta-analysis

2 messages · Diego Grados Bedoya, James Pustejovsky

Diego Grados Bedoya
# 
Dear all,

I am struggling about the implications of combining two meta-analyses (the
potential shortcomings of conducting a second-order meta-analysis):

Meta-analysis1 was achieved with the ln(RR1)[based on variable x1]

Meta-analysis2 was achieved with the ln(RR2) [based on variable X2]

There is an underlying relation between the units of X1 (u1) and X2 (u2):

u3=u1*3k ; u3=u2*300k

I wonder if it is possible to perform such analysis just including for
instance as a moderator the categorical variable for units (units1,units2)?
Am I missing something in the implications of the units that would affect
the correct comparison of ln(RR1) and ln (RR2)? I have consulted the work
of Schmidt and Oh (2013), Castellanos and Verd? (2012), Tamin (2011), and
other articles related to effects sizes calculations but I am not pretty
sure if I am not considering some kind of previous correction of the effect
sizes of the first order meta-analyses (or if it is not possible at all).

Maybe you can point me out in some direction about it.

Thank you in advance,

Diego
James Pustejovsky
# 
Hi Diego,

The log response ratio is a scale-free metric, so linear transformations of
the units of measurement will not change the value of lnRR. From what you
have described, it seems that u1 is a linear transformation of u2 (u1 = 100
* u2), and so in principle lnRR measures calculated from measurements on
either scale should be directly comparable.

James

On Thu, Dec 10, 2020 at 4:15 PM Diego Grados Bedoya <diegogradosb at gmail.com>
wrote: