Date: Mon, 19 Dec 2011 12:50:10 -0500
Reply-To: Rich Ulrich <rich-ulrich@live.com>
Sender: "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From: Rich Ulrich <rich-ulrich@live.com>
Subject: Re: Can I compare two within-subjects' dependent variables while
holding co-variates?
In-Reply-To: <1324311713513-5086603.post@n5.nabble.com>
Content-Type: multipart/alternative;
If, as you say, you are new at this, then you certainly need to keep
the analyses as simple as possible, because you have plenty of
trouble in laying out simple *hypotheses* from this information and
you don't need to add the difficulty of reading the tests on obscure
contrasts that are only implicit -- Be explicit about what you are
contrasting. See below for further comments.
> Date: Mon, 19 Dec 2011 08:21:53 -0800
> From: noahermesh@gmail.com
> Subject: Re: Can I compare two within-subjects' dependent variables while holding co-variates?
> To: SPSSX-L@LISTSERV.UGA.EDU
>
> Hi,
> Thanks a lot for your replay.
>
> My dependent variables are PTSD measures. I have 3 sub-scales and one total
> measure. These 4 measures refer to two different events, so each participant
> has 8 measures (4 for every event - 4 couples).
Eventually you may do otherwise, but for starters, take *one* measure,
like the Total, and see what sense you can make of it. So now you have
one measure for two events -- and you seem (below) to be interested in the
*difference* of those two. Subtract... and now you have *one* measure
for an analysis. If the average is also of interest, take that average, and
you have another, single, measure to look at in an analysis.
>
> I wanted to compare, for example the degree of avoidance symptoms between
> event 1 and 2, and so on, regarding each couple. Each participant has 2
> marks/grads, for each measure - one for every event.
> But, than I wand to hold as co-variates the time duration that passed since
> each event. I wanted to do that cause there is a significant difference
> between the time duration that passed between the two events.
Here's a problem for testing duration since a traumatic event: "duration" is not
going to be at all equal-interval or linear, so you will get into trouble if you try
to use it as measured merely in days or weeks or months or years. Three years
is not much different from 4 years, in respect to recovery, though the difference
is 365 days. Ten days versus 375 days is an interval that is much longer, for the
difference you would expect.
Maybe you collected the intervals scored into "reasonable" categories. If not,
you need to address that problem. I might apply my own "expert" opinion to
create categories, if I had some idea. Otherwise, if I didn't want to throw out
the detail, I would start by taking the log of the durations, and (possibly)
draw in the extremes (Windsordize). For durations, sometimes the reciprocal
transformation (events per year, instead of years-per-event) is more appropriate.
The reciprocal transformation is stronger that the log, and thus it does more
to emphasize differences that are closer to zero.
> While doing an ANOVA or MANOVA, the two co-variates (duration since each
> event) influence the two dependent variables, and that is no good. I thinks
> there is no point in holding as co-variate the duration since event 1 while
> looking at the symptomolgy after event 2.
> So, I was wondering how can I compare my dependent variables in a better
> way?
But it does make some sense to use both covariates when the criterion is the
difference. I think I see a model here with one outcome and two covariates,
which is not too hard to perform or interpret.
>
> The participants were all volunteers, who agreed to participate in a
> web-survey.
>
>
> I hope my answers are clear.
> Please let me know if not, and I'll be happy to answer.
[snip]
Hope this helps.
--
Rich Ulrich
[text/html]
| 
