[R-sig-ME] Matched-Pair Cluster Randomized Trials

[Dennis Murphy]

Hi:

On Fri, Jul 15, 2011 at 2:55 PM, Mathieu Maheu-Giroux
<mmaheugi at hsph.harvard.edu> wrote:
> Hi,
>
> I am trying to analyze data from a matched-pair cluster randomized trial.
> The group-level outcomes were analyzed using a random effect meta-analysis
> but I now want to investigate individual-level outcomes. I have used glmer
> to take into account the school clustering but I am unsure about the correct
> way to take into account the matching...

The treatments are assigned within pairs, where a pair can be thought
of as a block, so it would seem to me that you would want pair as a
(random) block effect in the model.  At this level, it's a completely
randomized block structure. After that, we would need to know the
nesting structure to offer any helpful advice. I'm no expert in this
particular application area, but the variation in sample sizes between
matched pairs indicates a weighting issue that needs to be addressed,
perhaps by suitable nesting or some other device, e.g,. average school
enrollment within pair as a covariate. As far as your model is
concerned, I would suggest starting with

glmer(Outcome ~ (1 | Pair) + Trt + (1 | SCHOOL) + <other terms that
correspond to nesting, etc.>, ...)

I assume you either have a pair variable or are capable of producing
one in your data.

(Side note: Blocks in experimental design are analogous to strata in
survey designs, nesting is analogous to subsampling in cluster
designs....just in case.)

HTH,
Dennis
>
> glmer(Outcome ~ Trt + (1|SCHOOL), data=Data, family = binomial)
>
> Is there an option to include the matching as a strata?
>
> (There is a total of 18 schools that were matched on geographic area and
> size. In each pair (9), one school was randomized to an intervention. There
> is a total of 1159 observations (but cluster size varies from 20 to 180).
> The primary outcome is binary.)
>
> Thanks,
> Mathieu MG
>
>        [[alternative HTML version deleted]]
>
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>