search

qiime2 / q2-longitudinal

QIIME 2 plugin for paired sample comparisons
BSD 3-Clause "New" or "Revised" License
9 stars 18 forks source link

ENH: control charts and volatility analysis #32

Open nbokulich opened 7 years ago

nbokulich commented 7 years ago

Comments this would be a new visualizer. thanks to @antgonza for the suggestion.

References See compare_trajectories.py

nbokulich commented 6 years ago

Control charts are added in #36 but we still need to find a useful test statistic to compare longitudinal variance/volatility between groups.

Levene/Fligner tests are not appropriate since they assume sample independence.

antgonza commented 6 years ago

Do you have any thoughts about this? In the original conception of that script we didn't have an statistical value for that but perhaps one option will be to do RMS on each trajectory to calculate power and then Levene/Fligner to test independence between trajectories ...

nbokulich commented 6 years ago

@antgonza thanks for discussing!

I had explored using Levene/Fligner tests but discussed with a statistician who informed me that these tests were only appropriate for comparisons at isolated time points (they assume sample independence). So I am still looking for an appropriate test.

RMS looks interesting but I wonder if we would run into the same issues with sample dependence. I will look into this. Thanks for the suggestion!

antgonza commented 6 years ago

RMS, assumes that you have a trajectory (multiple points) and basically calculate the power in that wave form (trajectory). In other words, it will give you one single value per trajectory/signal. IMOO the biggest issue with this method is the Nyquist frequency, which AFAIK is unknown. With this in mind, and having a single value per trajectory, you could use any statistic method to get a p-val and see if the RMS per group/control is significant.