1
$\begingroup$

I have a question regarding a mixed model analyses in SPSS. I have 2 groups (shift workers and non shift workers), each individual has different number of observations (depending on the frequency of the onset of a disease). But it is clear that shift workers have higher frequency of the disease than non-shift workers. Depending on the onset of a disease, absenteeism was measured. I know that mixed model can deal with missing data. But does it also takes into account that shift workers has more observations than non-shift workers?

Because depending on that, my interpretation of the results would be different.

$\endgroup$
2
$\begingroup$

In general, mixed models provide valid inferences under the missing at random assumption provided that the mean and variance-covariance structures are appropriately specified. This protective effect also carries over to the visiting process, namely the mechanism that describes when subjects provide measurements. In particular, in mixed models will provide valid results when the visiting process is a random process, i.e., in a longitudinal study when the decision that a subject needs to come earlier than originally planned is based on the previously observed measurements of that subject.

$\endgroup$
  • $\begingroup$ Hi Dimitris. Thank you for your reply. I don't understand it fully yet. Becasue do you know ifmixed models takes into account the differences in number of observations between the 2 groups? Is mixed models 'smart' enough for that? $\endgroup$ – Anna May 6 '19 at 12:03
  • $\begingroup$ For example. My outcome is that shift workers and non-shift workers do not differences much in absenteeism. However, shift workers have higher disease episodes. So if mixed model does not take the number of disease episodes into account, and the absenteeism rate is similar, then your interpetation would be differ a lot if it takes the number of disease episodes into account. :) $\endgroup$ – Anna May 6 '19 at 12:05
  • 1
    $\begingroup$ Mixed models can handle unbalanced data, and therefore deferences between the number of measurements in the two groups. However, as I mentioned in my answer, the important consideration is why these differences exist, and whether it is an informative process. $\endgroup$ – Dimitris Rizopoulos May 6 '19 at 13:43

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.