Help with Pre-Post Analyses (only 2 trials and no between subjects effects)

#1
I need some help for my thesis - am not getting much from my stats committee member.

I am looking at change in scores across treatment on a particular measure. I am using data kept in a forensic facility and I don't have any latitude to do more than what they do (i.e., pre and post testing). I also don't have any control group comparisons. Sample size is 150.

To look at change, I did paired samples t. Is this the right analysis? Seems so basic, but I don't have groups really. I guess I could look at responders vs. nonresponders or high and low scorers...not sure.

To see what predicted change, I looked at some predictors of change based on theory and correlations and formulated a HLR model. The change score method and regressor or lagged variable method come with same results, but the former is recommended based on the literature.

Here are my primary questions:

I had included time in treatment in my predictors, but I was wondering about controlling for nonrandom assignment of treatment. Treatment is mandatory for this program. Is it ok to include time in there?

I also was wondering about the different time points for individuals since the time for post depends on when they are done. Do I need to do some type of survival analyses after the regression? I didn't think so, but I wasn't sure.

Thanks for any thoughts.
 
#3
I think you have 2 options.

(1) make change scores and have your variables of interest--predictors of change (including time in txmt).

(2) use a auto-regression approach with time 2 as DV and time 1 as IV. Then you can include variables in model as predictors of change (including time in txmt).

A survival analysis has a binary outcome (death vs living) and the amount of time associated with the event occurring (i.e., time to death).

So years including time as a predictor of change is acceptable.

The different points of time for for post test will be reflected in your "time" variable.

Jason

http://thecsionline.org