Global Data Science Forum

 View Only
  • 1.  linear mixed models - estimating R square

    Posted Fri May 20, 2022 05:41 AM
    I have the following problem with my data.
    I have two significant interaction terms in the model which accounts for the random effect of a group (small group interactions). However, R square, calculated as:
    (random effects residuals - full model residuals)/ random effect residuals has horribly low values (.003 for the model with one interaction term, .01 for the model with two interaction terms). When I add these terms in a regular regression, both the total r-square and r-square changes are much higher. how can it be? 

    I understand that regular regression does not separate random effects from fixed effects, however, the estimates are quite high.

    Maybe my calculation of R square is not right? Any help? 

    I've downloaded both the dataset and the results.

    Karolina Ziembowicz


  • 2.  RE: linear mixed models - estimating R square

    Posted Mon May 23, 2022 10:15 AM

    Just want to make sure I understand.  So you the random effects are not nearly as strong in terms of R-squared that you'd expect?  When you add random effects to a model with fixed effects, they do little to increase the R-Squared?


    Shad Griffin

  • 3.  RE: linear mixed models - estimating R square

    Posted Mon May 23, 2022 10:38 PM
    I looked through the output.  Everything looks right, granted I don't have much context.  Unfortunately, your r-squared is low and there isn't much difference between the groups in the data set.  Did you get a different result with a standard regression?  I didn't see output for that.

    Shad Griffin