[R-meta] Effect size measure for count data

James Pustejovsky jepu@to @end|ng |rom gm@||@com
Tue Aug 17 20:43:34 CEST 2021


One way to approach your question would be to consider what effect size
measure is implied by the model you're using for the IPD meta-analysis.
>From what you've described, it sounds like you may be using a generalized
linear mixed model with a Poisson (or possibly negative-binomial)
distribution. Typically, such models use a natural log link function, which
corresponds to using log response ratio  ("ROM" in metafor::escalc) as the
effect size measure.


On Tue, Aug 17, 2021 at 12:25 PM Sarah Schaefer <sarah.schaefer using lir-mainz.de>

> Dear colleagues,
> We are currently planning an individual-patient-data meta-analysis on the
> effect of post-trauma sleep on symptoms of posttraumatic stress disorder
> (i.e., number of intrusions). For this purpose, we have data of seven
> analogue studies that assess intrusion frequency as a count variable using
> intrusion diaries or experimental triggering tasks. Number of intrusions
> per study are our primary outcome. Events (intrusions) regularly occur more
> than ones per participant. Our data follows a Poisson or a
> binomial-negative distribution (as we are still waiting for one dataset to
> finally check dispersion). For our IPD analysis we use a multilevel model
> which accounts for between-study heterogeneity on level 2. However, there
> is at least one study for which we do not have IPD. Therefore, we want to
> present both, a one-step approach IPD (using the multilevel model) and an
> analysis based on aggregated data (i.e., traditional meta-analysis). Now we
> a struggling finding an appropriate effect size mea
>  sure for the latter analysis. We aim to compare sleep and wake groups but
> I do not think that standardized mean differences would be an appropriate
> measure as they assume data to follow a normal distribution. However,
> looking into effect size measures for count data (within metafor), these
> are related to specific time intervals or group sizes. Due to different
> experimental designs (intrusion diary and intrusion triggering task) and
> the occurrence of multiple events per participant these are not applicable
> in our case. Therefore, we wanted to ask if anyone has an idea how to solve
> this issue (within metafor or other meta-analysis R packages). We would
> need an effect size measure for group differences in count data that allows
> for multiple events per participant (number of events > number of
> participants per group).
> Kind regards and thank you very much on advance,
> Sarah
> Aufsichtsratsvorsitzende: Dr. Carola Zimmermann,
> Geschaeftsfuehrer: Prof. Dr. Klaus Lieb, Prof. Dr. Beat Lutz
> (wissenschaftlich), Dr. Thorsten Mundi (kaufmaennisch).
> HRB 48032, Amtsgericht Mainz
> Diese E-Mail enthaelt vertrauliche und/oder rechtlich geschuetzte
> Informationen. Wenn Sie nicht der richtige Adressat sind oder diese E-Mail
> irrtuemlich erhalten haben, informieren Sie bitte sofort den Absender und
> loeschen Sie diese E-Mail. Das Kopieren sowie die Weitergabe dieser E-Mail
> und der darin enthaltenen Informationen ist nicht gestattet.
> This email may contain confidential and/or privileged ...{{dropped:20}}

More information about the R-sig-meta-analysis mailing list