I am a bit stuck with my model and I wonder if this is even possible using R. Basically I want to use a lagged dependent variable (LDV) in a cross-classified multilevel model (MLM). Following remarks by Allison and others (e.g., https://statisticalhorizons.com/lagged-dependent-variables ) I am assuming that using an LDV in a MLM leads to a bias in which the fixed effect of the LDV is overestimated and other effects underestimated. This seems to be due to a non-zero correlation among the LDV and the random effects in the MLM. Apparently there are ways to model this correlation/dependency explicitly in Stata or when using simpler designs (e.g., two-level MLMs; https://statisticalhorizons.com/wp-content/uploads/AllisonEtAl-Socius2017.pdf ) in R. However, I found no examples on how to do this with more complex designs in R such as mine.

My research question & dataset:

  • I want to predict school grades at time = t + 1 from school grades at time = t in interaction with teacher feedback at time = t.
  • I have 6 measurement times per student (i.e., random intercepts for students)
  • Students are clustered within classes; classes change at each measurement point so are treated as a new class at each measurement point (i.e., random intercepts for class)

I’ve been fitting this MLM using the brm function in R’s brms package (lme4 notation):

grade(t+1) ~ grade*feedback*time + (1 + grade*feedback*time | student) + (1 + grade*feedback | class)

I would be very thankful for any help on how to model the LDV in this MLM. If you could point me to an R code example that would be even greater!


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.