Hello everyone,
I'm working on data from the World Value Survey Longitudinal Dataset and a time-constant variable of heterogeneity at the country level. I would like to perform a multilevel analysis with cross-level interaction effect, such as:
Yij=f(Xij, Zj, Xij*Zj)
Yij is a 10-point Likert scale variable;
Xij is a constructed dichotomous variable from a Likert scale;
Zj is a continuous variable, ranges between 0 and 1 and is time-constant;
When running a random-intercept random-slope multilevel model in Stata, the coefficient of Zj and interaction term is not significant.
I was wondering if there is a way to assess if these results depend on the fact that the variable is time-constant and varies only across countries and how to take this characteristic into account.
Thank you for any advice and suggested literature,
AM
You could do some power calculations to see what size of effect could be detected given your data - eg using MLpowSim - if you do not have many countries and there is little variation between them in the predictor variable you may need a very large effect to be able to detect anything at conventional significance levels
http://www.bristol.ac.uk/cmm/software/mlpowsim/
http://www.bristol.ac.uk/cmm/media/software/mlwin/downloads/manuals/2-33/manual-web.pdf