SPSSX Discussion

two times versus four times in mediation analysis

Classic

List

Threaded

5 messages Options

Zdaniuk, Bozena-3

two times versus four times in mediation analysis

hello everyone, it is more of a theoretical question and I would love some feedback.

I am analyzing an intervention (8 week group therapy) study that measures outcomes at pre-intervention (t1), immediately post-intervention (t2), 6 months later (t3) and 12 months later (t4). I am looking at the changes in potential mediators predicting changes in the main outcomes. I am doing it by using a person-centered approach and looking at the difference between two time points - t1 and t4. Some of my group members believe I should find a way to look at all 4 points in my mediation analysis. Would using all 4 times be better than using just t1 - t4 difference?

thanks so much for any thoughts on this!

bozena

===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD

Rich Ulrich

Re: two times versus four times in mediation analysis

You don't say enough to suggest whether there is a strong, immediate impact, or

whether there is nothing at first. I'm sure you have some ideas that should shape

your hypotheses.

When I looked at sleep studies where meds were given to depressed patients,

there were HUGE effects on many sleep variables, greater that 1 SD. (How much

delay before sleep, before REM sleep, how much time in the first/second/third REM

periods, how much sleep in stage 1, etc.)

The question for the next few weeks of treatment was the shape of the followup

curve for each variable. The standard testing paradigm we adopted was to do a

t-test (each variable) to measure the initial impact, T1 to T2 (huge, moderate, none);

and then look at the slope of the changes across T2 to T6 - which might fall back to

the pre-meds level, or it might start changing or continue changing. We had time

intervals such that we expected equal amounts of change, so the linear trend across

the observed points, treating them as equal intervals, worked fine.

Rich Ulrich

From: SPSSX(r) Discussion <[hidden email]> on behalf of Zdaniuk, Bozena <[hidden email]>
Sent: Wednesday, November 14, 2018 11:58 AM
To: [hidden email]
Subject: two times versus four times in mediation analysis

hello everyone, it is more of a theoretical question and I would love some feedback.

thanks so much for any thoughts on this!

bozena

Maguin, Eugene

Re: two times versus four times in mediation analysis

In reply to this post by Zdaniuk, Bozena-3

To clarify.

Deltay12=y2-y1.

Deltam12=m2-m1.

Correlation deltam12 with deltay12.

Main question.

Is M measured at each time point? If not, then I don’t see that you can go beyond the (1,2) time point pair, unless more is going on. Also, if M and Y are analyzed for the same time point wouldn’t deltaY causes deltaM be an alternative explanation for deltam with deltay?

Gene Maguin

From: SPSSX(r) Discussion <[hidden email]> On Behalf Of Zdaniuk, Bozena
Sent: Wednesday, November 14, 2018 11:58 AM
To: [hidden email]
Subject: two times versus four times in mediation analysis

hello everyone, it is more of a theoretical question and I would love some feedback.

thanks so much for any thoughts on this!

bozena

Zdaniuk, Bozena-3

Re: two times versus four times in mediation analysis

In reply to this post by Rich Ulrich

We have quite a big t1 - t2 (pre- to post-intervention) effect and then it continues to improve with time or at least the gains are maintained. This is why I wanted to use t1 - t4 to look at the maximum observed changes.

And, yes, I can look at the shape of the post intervention changes (t1, t3, and t4) but it becomes much more complicated when I try to analyze whether changes in the mediator predict changes in the outcome. With two time points, the mediation analysis seems (at least to me) manageable...

bozena

From: Rich Ulrich [[hidden email]]
Sent: Wednesday, November 14, 2018 10:03 AM
To: [hidden email]; Zdaniuk, Bozena
Subject: Re: two times versus four times in mediation analysis

You don't say enough to suggest whether there is a strong, immediate impact, or

whether there is nothing at first. I'm sure you have some ideas that should shape

your hypotheses.

When I looked at sleep studies where meds were given to depressed patients,

there were HUGE effects on many sleep variables, greater that 1 SD. (How much

delay before sleep, before REM sleep, how much time in the first/second/third REM

periods, how much sleep in stage 1, etc.)

The question for the next few weeks of treatment was the shape of the followup

curve for each variable. The standard testing paradigm we adopted was to do a

t-test (each variable) to measure the initial impact, T1 to T2 (huge, moderate, none);

and then look at the slope of the changes across T2 to T6 - which might fall back to

the pre-meds level, or it might start changing or continue changing. We had time

intervals such that we expected equal amounts of change, so the linear trend across

the observed points, treating them as equal intervals, worked fine.

Rich Ulrich

hello everyone, it is more of a theoretical question and I would love some feedback.

thanks so much for any thoughts on this!

bozena

Rich Ulrich

Re: two times versus four times in mediation analysis

I've never been comfortable with the term, "mediator" - I like "predictors".

I never touched a Latent Structure Analysis, partly because I have very rarely

seen one that impressed me... and that's where I think of mediators. So I'm

going to add more comments only on Outcome.

You have a big effect, t1-t2, similar to t1-t4. For the sleep studies I mentioned,

we had dozens of variables, with up/same/down being three trajectories we

observed, after the initial big effect; you have something different.

Without knowing your subject, I can only guess that your audience will be

interested in knowing that the large effect is immediate, and lasts without much

further change. A simple demonstration would test the means, with paired t-test,

to confirm a similar, large change for t1-t2 and t1-t4. I might want to show the

tininess of the effect for t2-t4, where the thoroughly non-significant linear trend

might serve. Oneway ANCOVA, with t1 as covariate for t2 or t4 could be used to

add in tests for other predictors from time 1.

(I don't see where the "change in a mediator" will even fit in.)

Rich Ulrich

From: SPSSX(r) Discussion <[hidden email]> on behalf of Zdaniuk, Bozena <[hidden email]>
Sent: Thursday, November 15, 2018 8:00 PM
To: [hidden email]
Subject: Re: two times versus four times in mediation analysis

bozena

From: Rich Ulrich [[hidden email]]
Sent: Wednesday, November 14, 2018 10:03 AM
To: [hidden email]; Zdaniuk, Bozena
Subject: Re: two times versus four times in mediation analysis

You don't say enough to suggest whether there is a strong, immediate impact, or

whether there is nothing at first. I'm sure you have some ideas that should shape

your hypotheses.

When I looked at sleep studies where meds were given to depressed patients,

there were HUGE effects on many sleep variables, greater that 1 SD. (How much

delay before sleep, before REM sleep, how much time in the first/second/third REM

periods, how much sleep in stage 1, etc.)

The question for the next few weeks of treatment was the shape of the followup

curve for each variable. The standard testing paradigm we adopted was to do a

t-test (each variable) to measure the initial impact, T1 to T2 (huge, moderate, none);

and then look at the slope of the changes across T2 to T6 - which might fall back to

the pre-meds level, or it might start changing or continue changing. We had time

intervals such that we expected equal amounts of change, so the linear trend across

the observed points, treating them as equal intervals, worked fine.

Rich Ulrich

hello everyone, it is more of a theoretical question and I would love some feedback.

thanks so much for any thoughts on this!

bozena