I am tasked with analyzing a preexisting dataset. The experiment is fairly simple, but I'm finding that analyzing the data is challenging due to the fact that the experimental design is not ideal for analysis....

The data example (wide format):

Subj  Depend_1  Depend_2  Depend_3  Covar_1  Covar_2  Age  Group
 A       10       15        16        2         3     20     1
 B       11       13        NA        3         NA    31     1
 C       7        13        15        4         4     37     2

Subj = Subject

Depend_1 thru Depend_3 = repeated within-subjects variable. Depend_1 was pre-treatment, Depend_2 was post treatment, Depend_3 was 1 month post treatment.

Covar_1 and 2 = participant scores on a different test, Covar_1 was taken at pre treatment, Covar_2 was taken 1 month post treatment, there was no Covar test taken at post treatment. I'm interested in using this variable as a covariate.

Age = Subject age at test(s), I would also like to use this as a covariate.

Group = Between-subjects variable, where participants were randomly assigned to a treatment or control group.

I would like to run either an MLM or ANCOVA with this data. But I'm not sure how to format it for analysis. I have 3 DV measurements for each subject, 2 measurements for Covar, and 1 for age...

Would it be sensible to use Depend_1 as a covariate, since it is a pre-treatment variable? Meaning that, if I were to format this in long format, it would look like:

Subj  Time  DV  Covar  PreTest  Age  Group
 A      1   15    2       10    20     1
 A      2   16    3       10    20     1
 B      1   13    3       11    31     1
 C      1   13    4       7     37     2
 C      2   15    4       7     37     2

Or is there another way I should handle the mismatched measurements in the data?

Hopefully my explanation is clear, and any help or referrals will be greatly appreciated!

Thank you!


1 Answer 1


The way the data are set up it looks like a repeated measures model structured hierarchically - subjects at the higher level along along with related measures like age, pretest, and group, and times at the lower level ("within" each subject), measured with variables like DV and Covar.

I'd say that this is a problem best handled by the MIXED procedure in SPSS Statistics, or Generalized Linear Mixed Models depending on your assumptions about the distribution of the errors for the dependent measure.

  • $\begingroup$ Thank you! I will look into this. $\endgroup$
    – Tom
    Dec 4, 2019 at 16:10

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.