# Idea of statistical analysis for a particular design

#### Dita

##### New Member
Hi all, I really need your expertise!

Here a set of data obtained from a study: there is 16 participants. Each participant had 3 successive lab visits. During each visit, they consume one drink : either the Drink 1, Drink 2 or Drink 3 (control). The order of drink administration is randomized, in the end they all tested the 3 drinks. We took a physiological measure continually 30 min before consumption (divided in 8 time-points) and 60 min after consumption (divided in 16 time-points). The physiological measure is the intervals between heart beats expressed in milliseconds.

I would like to compare my list of heart beats intervals [mms] between each product on the same time point.
ie: is there a difference between the 3 drinks at the first time point after consumption ? After the second time point after consumption? etc.

I would like to compare all drinks: 1 with 2; 1 with 3; and 2 with 3.

Hope it is clear, do you have any suggestions of statistical analysis please?
If any, what are the test assumptions, and what do I do if my dataset doesn't fit assumptions?
Is is possible to include the baseline of each product (before consumption)?

Thanks a lot,

Dita

#### Karabiner

##### TS Contributor
Looks like a repeated-measures analysis of variance with 3 repeated-measures factors,
i.e. drink (3 levels), period (2 levels: before/after), and time point (8 levels; if time points
9 to 16 after drink are not included in this analysis).
This will provide some global tests on the effects associated with the respective factors.
If you additionally want to analyse the data time point-wise, then you could follow-up
the analysis with 8 two-factorial repeated-measures analyses (only factors drink
type & period), one for each time point.

If time point is not suitable for 1:1 matching (before/after), and/or you want to include
all 16 time points after drink, then you could perform a global analysis with just two
repeated-measures factors, "drink" and "time point" (16 levels), and in addition the
8 baseline measurements as covariates. As follow-up, you could repeat that anaysis
separately for each of the 16 time points, although that looks a bit awkward, and
maybe there is a more elegant solution.

Maybe the time intervals should be transformed before analysis, for example by taking
the logarithm, but this depends on substantial considerations and on what is common in
this field of study.

With kind regards

Karabiner

Last edited: