2
$\begingroup$

I am working on a dataset of longitudinal (baseline and control) data about some values in blood in two treatment groups

I performed a propensity score matching with the matchthem methodology in R (first imputing the data with mice function, and then matching with the matchthem package). After that the package suggest using svyglm for computing the tests on a dichotomic variable as outcome. However I am trying to compute a linear mixed model, adjusted by id as random variable and time and treatment in this matched dataset, but I can't find the way of doing it.

Any ideas?

$\endgroup$
2
  • $\begingroup$ Hi! Is your treatment time varying? If it is, did you specify a time-varying propensity score? $\endgroup$
    – jmarkov
    Feb 23, 2023 at 9:39
  • $\begingroup$ Tretatment is not time varying. It's constant $\endgroup$ Feb 24, 2023 at 10:02

1 Answer 1

0
$\begingroup$

All you need to do is fit each outcome model within each imputed dataset. This is what MatchThem facilitates. With mixed models, you may not be able to do so using MatchThem's built-in functionality, but you don't need to. Once you have fit your models, you can combine them with Rubin's rules as usual.

Note that fitting mixed models to matched data is not straightforward in the sense that the properties of such an estimator are not well studied. It is generally recommended you include a random effect for matched pair membership. Note that mixed effects models can be used easily with matching weights, so make sure your matched sample is self-weighting (i.e., all units in each treatment group have the same weight).

$\endgroup$
2
  • $\begingroup$ So you suggest to fit a linear mixed model in each time and then pool the results? If you could provide some example code to fit a lmer in a Matchthem object it would be great. Thank you in any case $\endgroup$ Feb 24, 2023 at 10:37
  • $\begingroup$ Matching is highly inefficient. $\endgroup$ Oct 20, 2023 at 12:58

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.