I'm really struggling to find a way to conduct suitable statistical tests for the following research I'm doing:
I'm taking 120 drug addicts through a 4-week treatment program.
50 have completed the treatment, 70 have dropped out.
I've used two psychological tests,
Measure A - provides a single score (Ai)
Measure B - has three component scores (Bi, Bii, Biii)
- & I'd really like to include B(total) if possible...?
I measured everyone at the start (Start)
But only treatment completers at the end (End):
My required alpha level is .05 (before adjustments) and beta is .90, the effect size I'm anticipating is moderate (~0.5).
I have two hypotheses I want to test:
1. People who drop out will have scored significantly higher on measures Ai, Bi, Bii, and Biii on starting the treatment
2. People completing treatment will score significantly lower on measures Ai, Bi, Bii, and Biii than when starting the treatment
...If it helps, previous research has found Measure A to not significantly correlate with Measure B. However, the subscales (Bi, Bii, and Biii) are all significantly and highly positively correlated.
I would be most gratefully for any help you can give me on this. Please let me know if anything I wrote doesn't make sense or you need more information
I'm thinking 2 MANOVAs might do it -
1. Start: completers vs. non-completers for Ai, Bi, Bii, and Biii
2. completers: start vs. end for Ai, Bi, Bii, and Biii
... However, would this have a fair chance of getting a significant result, and would I have room for looking at individual tests?
... Would I be better off simply running 2 lots of 4 t-tests?