0
$\begingroup$

I have read through most of the threads on this forum and several must-read materials on HLM/mixed effect models. Most empirical settings where people use this method focused on education (i.e. student-> class -> school) or natural grouping process as a result of social-economic segregation (i.e. residents -> county -> nation). What about cases where level-2 (organization) exerts considerable somewhat more coercive influence on level-1 outcomes (individual), i.e. facilities nested under/owned by parent firm or employees working for the same company? Can HLM/mixed effect be applied in the same way or are there any caveats in doing so, statistically?

P.S. My data is cross-classified: each facility (level-1) is owned by a parent firm (level-2a) and resides in a geographical location (level-2b). My worry here is whether level-2a is at all reasonable statistically since it's a "ownership" relationship. My # of groups as well as variations on either level in the dataset are both quite large.

Anyone could help? Leads on similar empirical evidence in prior research would also be extremely useful. Thank you!

$\endgroup$
1
  • $\begingroup$ If your question has been answered to your satisfaction, you can accept an answer by clicking the check mark under the voting arrows. $\endgroup$ Commented Jul 18, 2017 at 16:20

2 Answers 2

1
$\begingroup$

Mixed models are totally appropriate for this problem. Just a note about what it is mixed models/HLMs do (among other things): They partition variance accounted for in the outcome into the levels you specify according to the nested structure of your data. So whether one level accounts for more or less variability in the outcome is not something that is detrimental to a mixed model.

Take a simpler two-level model in which you wanted to know whether age predicted employee satisfaction, but also wanted to account for the nested structure of your data. So you run a mixed effects model in which employees are nested within work site. If a large proportion of variance in the outcome is accounted for by the level 2 variable (work site), you'll be glad you ran your mixed effects model to obtain unbiased parameter estimates of the relation between age and satisfaction. If work site accounted for little to no variability in the outcome or variability in the relation between age and employee satisfaction, then the results would essentially approximate the results you would obtain from a linear regression. However, even in this instance, it may be more theoretically defensible to employ a mixed model, even if the empirical results suggest that it buys you little compared with a linear a regression.

$\endgroup$
0
$\begingroup$

What about cases where level-2 (organization) exerts considerable somewhat more coercive influence on level-1 outcomes (individual)

I don't see why what you've already heard about mixed models would apply any less to this situation than to students nested within schools. You seem to be concerned about the effect size, but a mixed model places no restrictions on the sizes of the random effects. What is the statistical relevance of an "ownership" or "coercive" relationship?

$\endgroup$

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.