How to take to account matching in cluster randomised controlled trial analysis?

[Mari TO]

Hello,

I am analysing data from a matched-pair cluster randomised controlled trial. Pairs were matched according to schools’ and classes’ (5th grades) size and spacial proximity.

Final match was the following:
24 pairs – 1 + 1
4 pairs – 2 + 1 (e.g. one very big school had the same number of children in 5th grades as two big schools)
6 „empty“ pairs – 1 + 0 (e.g. the school declined their participation just before the study)

So far I have not taken to account of matching when doing multilevel analysis.

As an example:

xtmelogit initiation intervention || school_no:, variance or

where initiation refers to alcohol use initiation by the participant (yes/no), intervention refers to intervention condition (yes/no) and school_no refers to a number each school has (altogether there are 66 schools, 34 intervention and 32 control).

I am unsure of how to take account of matching. Any ideas?

Many thanks,
Mariliis Tael-Öeren

[Clyde Schechter]

Create a variable that identifies each pair (or triplet). The pairs and schools form a multiple-membership model which you can model as if they were crossed random effects. Or, perhaps simpler, keep the random effects part of the model as is and just add i.pair_identifier to the fixed effects level.

[Mari TO]

Thank you. I will try that.