0
$\begingroup$

My 106 respondents answered questionnaires at the start of the study to assess trait and after this they participated in an Experience Sampling Method research design. In this design, participants answered 10 times a day for 7 days questions about their emotions. After the study, the data is stacked. So, for example positive affect has 70 different values in one individual but trait has 70 values that are the same in this individual. Trait doesn't vary within the subject but it does between subjects. Is it possible to run multilevel data analysis with positive affect as the predictor and trait as dependent variable?

$\endgroup$

1 Answer 1

1
$\begingroup$

This is not possible. If you would try to fit such a model (with random intercepts at the subject level), the residual variance is essentially forced to be zero (since there is no variability in the outcome variable within subjects). In essence, the outcome variable must always be a time-varying variable measured at the lowest level.

You have two options:

  1. Use affect as the outcome and the trait variable as the predictor.
  2. Aggregate your data to the subject level (i.e., compute the mean of the affect variable per subject). Then you can run a regular regression with the trait variable as the outcome and the mean affect variable as predictor.
$\endgroup$
0

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.