I want to estimate a polynomial regression to test the effect of self and follower-perception of leadership behavior on some outcome (e.g., followers' job satisfaction). Hence, I have a multilevel data structure with followers' assessments of their supervisors' leadership ($X_1$) as well as followers' job satisfaction ($Y$) on level 1 and supervisors' self-assessment of their leadership behavior ($X_2$) on level 2.

Ignoring the multilevel structure of the data, I would fit the following model:

$$ Job Satisfaction = b_0 + b_1SelfRating + b_2FollowerRating + b_3SelfRaring^2 + b_4SelfRating \times FollowerRating + b_5FollowerRating + e $$

My question: Can I fit a similar model as a hierarchical linear model? And if yes, are there any things, like mandatory random effects, I have to consider?

In my mind the final model would look (something) like this:

$$ Z_{ij} = \gamma_{00_j} + \gamma_{01_j}Y_j + \gamma_{02_j}Y_j^2 + \gamma_{10_j} X_{ij} + \gamma_{11_j}X_{ij}Y_j + \gamma_{20_j}X_{ij}^2 + u_{0_j}+ u_{1_j} + u_{2_j} + e_{ij} $$

Thanks for your help!

  • $\begingroup$ The necessity of random effect or hierarchical structure depends on the dependent variable (jobsatisfaction). If dependent variable is independent , then do not need random effect. Your description is not clear enough to judge if the dependent variable is independent or not. $\endgroup$
    – user158565
    Commented Dec 12, 2018 at 3:47
  • $\begingroup$ I am not sure if I get your right. What do you mean with independent? Independent of the nested structure of the data? $\endgroup$ Commented Dec 20, 2018 at 8:44

1 Answer 1


I found an article by Nestler, Humberg and Schönbrodt forthcoming in Psychological Methods titled "Response surface analysis with multilevel data - Illustration for the case of congruence hypotheses". it is not yet published but you can find the preprint here.

For a case with two level-1 independent variables, they argue for including fixed effects for group-mean centered first-order terms, group-mean centered higher-order terms, and the group means. They also include all possible random effects.

For a case with one level-1 and one level-2 independet variable they advise "to group-mean center the Level 1 congruence variable and including the group-mean-centered Level 1 congruence variable, the Level 2 congruence variable, and the resulting group means in the model to disentangle the Level 1 and Level 2 effects in the response surface analysis."


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.