09-30-2011 11:50 AM
I'm planning to collect data on an ecological momentary assessment predicting assessing two psychologicla dimensions. Basically, I will be collecting eight random data collection per day for a period of a week (a total of 56 datapoints). I wanted to use all of the previous information to predict the current data point. I am familiar with multilevel modeling using SAS' Proc Mixed, but to use all of the information from one psychological variable to predict the other, I would need 55 lagged variables. Additionally, this approach just didn't seem right to me (alpha criterion for significance would need to be adjusted to .05/56 = 8.928571428571429e-4 giving me almost no power).
After doing some searching, I realized with this study I hope to use a time-varying coviariate (psychological dimension 1) to predict a repeated dependent variable (psychological dimension 2). Accordingly, the only analyses that I was able to find to perform this investigation was a Cox proportional hazards regression.
Does cox proportional hazards regression allow you to use all of the previous data, or the most recent data point? If not, is there a particular statistical method that would accomplish what I would like? It's important to me to analyze all of the time data as the temporal relationship between the two psychologicla dimensions in such a short time intervals is entirely unprecedented.
I would appreciate any and all help you can provide.
10-30-2011 01:54 PM
I'm further approaching the time when I will need to conduct the analysis and I hadn't heard a response from anyone.
If anyone has any input I would greatly appreciate it!
10-30-2011 02:45 PM
I'm not a pshcyologist, so my knowledge of EMA is just from Googling it just now.
My guess is that Proportional Hazards is not appropriate because it requires a binary outcome measure. You didn't state the units of your data but referenced MIXED, which generally uses a continuous response.
10-30-2011 02:51 PM
Hi Dr. Muhlbaier,
Thanks for getting back to me. You're correct that I would have to person-center and dichomoize the continuous data. It would be great to be able to make use of the full data in this case, I'm just uncertain if such a procedure exists. If you have any suggestions, please let me know.
01-20-2012 03:46 PM
It's been a few months since I originally asked this question. Unfortunately, I still haven't been able to find an answer. Does anyone else have any suggestions?