Hi everybody, I'm working on my PhD thesis and I need help with data analysis of longitudinal data...
To make the long story short, it's about hospital treatment of AUD. I have 3 crucial variables:

1. Self-forgiveness trait (scale)
2. Alcohol craving (scale) - longitudinal data - 3 measures/week during treatment
3. Follow up (90 days after discharge) drinking status (nominal) - 0=relapse, 1=abstinence (this one could also be measured as number of days of abstinence after discharge measured in that time point)

And I have 3 problems/hypothesis

1. Self-forgiveness trait is a predictor of changes in alcohol craving during treatment
2. Self-forgiveness trait is a predictor of drinking status 90 days after treatment
2. Changes in alcohol craving during treatment are predictor of drinking status 90 days after treatment

As you can see, my longitudinal data serves me both as criterion and predictor variable.
Is HLM best/possible way for testing my hypothesis? Should I test follow-up data with some kind of survival analysis?

Thanks in advance,

Nikola

On Thu, 29 Mar 2018 09:13:36 -0700 (PDT), Nikola Babi?
<nikobab@gmail.com> wrote:

>Hi everybody, I'm working on my PhD thesis and I need help with data analysis of longitudinal data...
>To make the long story short, it's about hospital treatment of AUD. I have 3 crucial variables:
>
>1. Self-forgiveness trait (scale)
>2. Alcohol craving (scale) - longitudinal data - 3 measures/week during treatment
>3. Follow up (90 days after discharge) drinking status (nominal) - 0=relapse, 1=abstinence (this one could also be measured as number of days of abstinence after discharge measured in that time point)
>
>And I have 3 problems/hypothesis
>
>1. Self-forgiveness trait is a predictor of changes in alcohol craving during treatment
>2. Self-forgiveness trait is a predictor of drinking status 90 days after treatment
>2. Changes in alcohol craving during treatment are predictor of drinking status 90 days after treatment
>
>As you can see, my longitudinal data serves me both as criterion and predictor variable.
>Is HLM best/possible way for testing my hypothesis? Should I test follow-up data with some kind of survival analysis?
>

Instead of answers, I have preliminary questions about the data.

For (1) and (2) - does "Scale" indicate that there are a number of
items which are counted (0/1) or averaged (0-k) ? - Otherwise,
I assume it must be a single score, rated 0-k for some k.

For (3) - Do the comments indicate that you effectively have "duration
without drinking" which is number of days, censored at 90?

You have 3 measures/week during treatment: How many weeks of
treatment? (Is this a fixed number?)

The chance you have for finding a within-person correlation between
forgiveness and craving depends on having some variability, and
on having a sufficient number of cases. How many cases? What
can you say about variation, for each of the measures?
- Do you want to use /all/ the treatment data together, or is there
an odd baseline before the subject de-toxifies, etc.? - In my
experience with clinical psychiatric research, the first week of an
inpatient stay started out "very ill", especially for psychotic
patients or those with drug episodes. Rating scales were typically
/noise/ to be discarded, so far as most research questions on
subsequent treatment were concerned.

--
Rich Ulrich