Level identification for repeated measures

[JosipKaruc]

Hi,

I wonder if there is a
I am currently working on study that aims to investigate how physical activity level (predictor) is related to sleep duration (outcome) in adolescents (n=92). We have tracked students for seven days (i.e. we have 7 measurements for each subject for every variable) which makes this study a study with a repeated measures design. Also, students are clustered within the classes (n=68) within the schools (n=4). Physical activity level, sleep duration, and other confounding variables are continuous.

My questions are following:

1. Do I have 4-level structured data (i.e. measurements at L-1; students at L-2; classes at L-3; and schools at L-4) ?

2. Can I statistically compare models at level-1 (which includes measurements) with models at level-2 (wich includes measurements and students) to see where is more clustering?

3. Does L-1 variable (i.e. measurements) is set in mlwin in the same way as other ''typical'' variable of interest (e.g. student at L-1)?

4. Which 'statistical approach' is best for these kind of structured data?

Thanks

Josip

[billb]

Hi Josip,
You appear to have a 4 level structure but in practice you would fit this as 3-levels with school just included as fixed effect dummies as you only have 4 schools. I am not sure I understand your 2nd question but fitting it as 3 levels would give you variances for the 3 levels so you could see the relative variability at levels 1 and 2. So yes to question 3 you can set measurement at level 1 within student at level 2.
Question 4 is a bit too vague - for repeated measures the choice is to fit the 3 level model as described or alternatively to fit a multivariate model for the time points with a 2 level structure above (students in classes). I think which is better depends on your research question.
Best wishes,
Bill.