Multilevel Modeling question

[Monika Stojek]

Hello,

I am using MLM to examine the levels of craving across eight time points and
two conditions. In the final step of the model, which includes both
condition and time, I found main effects of time and medication, but no
interaction between the two. When I ran t-tests, the two conditions did not
differ significantly on their levels of craving. Therefore, I am not quite
sure how to interpret this finding. Please let me know if you have any
suggestions or ideas. I am obviously happy to answer any follow-up questions.

Kind regards,

Monika

=====================
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]

Interpreting results so far:  "There is an overall
difference between craving between groups averaged
across all periods.  There is a difference across time. 
The overall interaction with 7 d.f. is not significant."

Other comments:  You do not give enough detail.  So far,
it sounds like a repeated measures design, not Multilevel.
What are the elements, and what are their Ns?  How strong are
those overall effects -- 0.001?  barely 0.05?

It is usually poor form (because it offers relatively little power) to
analyze 8 time periods without considering useful contrasts instead
of relying on that test across 8 periods.   A 1 d.f.  contrast that
contains most of the trend or other differences will have far more
power than the 7 d.f.  test among periods.  Testing the specific contrast
also answers a specific question, like, "Is there a trend?" -- instead of
giving a relatively useless answer like, "There are differences."

You say you "ran t-tests."  I have to guess, so I guess that you ran
t-tests at the separate times.  The lack of "significant" tests here,
whereas the overall test was significant, shows that you do have
marginal power and that you cannot afford to waste power by
testing the non-specific hypotheses.


Do you expect a linear time trend?  Was there a pre-intervention
Baseline, which should be contrasted to the next period, to test an
Intervention effect?  - An Intervention design might deserve a t-test
between 1 and 2 (assuming intervention after 1), to detect a jolt, with
a separate test for 2 through 8 that tests for trend that might either
return towards Baseline, move away, or stay constant.

--
Rich Ulrich

> Date: Thu, 27 Jun 2013 09:20:19 -0400

> From: [hidden email]
> Subject: Multilevel Modeling question
> To: [hidden email]
>
> Hello,
>
> I am using MLM to examine the levels of craving across eight time points and
> two conditions. In the final step of the model, which includes both
> condition and time, I found main effects of time and medication, but no
> interaction between the two. When I ran t-tests, the two conditions did not
> differ significantly on their levels of craving. Therefore, I am not quite
> sure how to interpret this finding. Please let me know if you have any
> suggestions or ideas. I am obviously happy to answer any follow-up questions.
>

[Ryan]

Monika,

Please post your syntax, and tell us *exactly* what your research question(s) is/are.

Ryan

Sent from my iPhone

On Jun 27, 2013, at 9:20 AM, Monika Stojek <[hidden email]> wrote:

> Hello,
>
> I am using MLM to examine the levels of craving across eight time points and
> two conditions. In the final step of the model, which includes both
> condition and time, I found main effects of time and medication, but no
> interaction between the two. When I ran t-tests, the two conditions did not
> differ significantly on their levels of craving. Therefore, I am not quite
> sure how to interpret this finding. Please let me know if you have any
> suggestions or ideas. I am obviously happy to answer any follow-up questions.
>
> Kind regards,
>
> Monika
>
> =====================
> 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

=====================
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