Hi, I have some repeated measures data, one measurement a day for three days in a row, and the measured variable looks normally distributed. I have two groups, the "really ill" and the "not ill after all", and want to use ROC curves to see if it's a good idea to use my measured variable as a test for this illness. Does anyone know if there is there an ROC curve setting for my repeated measures data or do I have to just produce three ROC curves, one for each day? thanks
$\begingroup$
$\endgroup$
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
1
Without any further information, there's no way to specify a "canonical" way to combine your three measurements into a single quantity (which is what you need to do to create a single ROC curve that takes into account all the data simultaneously). If you have no further theoretical insight into the nature of your measurement and how repeated measurements behave, then you can experiment with different combining rules to see which one gives the best ROC curve.
For example, you could try taking the average of your measurements (mean, median, or root mean square for example) or the max or the min. Taking the average might make sense if you suspect that the variation from day to day is chiefly due to noise. Taking the max might make sense if you're trying to detect something that is difficult to detect and want to give yourself three opportunities to do so.
answered May 2, 2013 at 14:30
-
$\begingroup$ Thanks for your answer. I was basically wondering if there is a "canonical" way to combine them. From what I understood from speaking with the doctor whose data this is, there is unfortunately no way to combine these three measurements to one, at least not a way that makes sense from a medical point of view. Especially given the large amount of missing data that we have, it would be best, i guess, to just look at the three ROC curves separately. $\endgroup$– IoannaMay 13, 2013 at 13:37