multilevel growth model questions

[Maguin, Eugene]

This is the model statement I have been using:

mixed motivate with wave txgroupn il2018/fixed=wave wave*wave txgroupn

   wave*txgroupn il2018 wave*il2018/

   print solution/random intercept | subject(rid) covtype(id).

 

What I’d like to see are comparisons between values of txgroupn at values of wave. After looking at emmeans (and trying it out), I can’t use it unless the focal variables are as factors. I’m assuming no work-around for this. I then changed the model this:

mixed motivate by wave txgroupn il2018/fixed=wave wave*wave wavesq txgroupn

   wave*txgroupn il2018 wave*il2018/

   print solution/random intercept | subject(rid) covtype(id).

 

This wasn’t liked at all and the problem seemed to the wave*wave term. Ok, easy to fix, just compute wavesq (wave squared). The new model:

mixed motivate by wave wavesq txgroupn il2018/fixed=wave wavesq txgroupn

   wave*txgroupn il2018 wave*il2018/

   print solution/random intercept | subject(rid) covtype(id).

 

This executed but yielded different results because the wavesq contrasts were marked as redundant. This is  wrong. I had thought could get around this by naming wavesq using a with subcommand. However, this term was also marked as redundant. Again, wrong, seriously wrong.

 

What is, is. The Test command seems to be the only alternative; however, in terms of examples, can covariates be accommodated in the Test command? Or, more generally, how do I get the comparisons I’m interested in?

 

Thanks, Gene Maguin

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

[Bruce Weaver]

Hi Gene.  Several years ago, I wrote a pair of macros (!OLScomp and !MLEcomp)
for comparing any two fitted values, because I assumed that the LMATRIX
sub-command of UNIANOVA could only handle factor variables.  I assumed that
because that's what all the examples showed.  I got my macros published in a
good journal, as it turned out, so it was not a total waste of time.  But
well after the fact, I realized that LMATRIX (UNIANOVA) and TEST (NOMREG)
could handle factor variables, and so could make all the same fitted value
comparisons as my macros.  I bet the TEST sub-command for MIXED can handle
covariates too.  

You can find some relevant documents on this web-page:

https://sites.google.com/a/lakeheadu.ca/bweaver/Home/statistics/spss/olscomp_mlecomp

See especially the PDF document linked in the second to last paragraph.  The
LMATRIX examples there are likely fairly similar to what you need for TEST
in MIXED.  

Cheers,
Bruce



Maguin, Eugene wrote

> This is the model statement I have been using:
> mixed motivate with wave txgroupn il2018/fixed=wave wave*wave txgroupn
>    wave*txgroupn il2018 wave*il2018/
>    print solution/random intercept | subject(rid) covtype(id).
>
> What I'd like to see are comparisons between values of txgroupn at values
> of wave. After looking at emmeans (and trying it out), I can't use it
> unless the focal variables are as factors. I'm assuming no work-around for
> this. I then changed the model this:
> mixed motivate by wave txgroupn il2018/fixed=wave wave*wave wavesq
> txgroupn
>    wave*txgroupn il2018 wave*il2018/
>    print solution/random intercept | subject(rid) covtype(id).
>
> This wasn't liked at all and the problem seemed to the wave*wave term. Ok,
> easy to fix, just compute wavesq (wave squared). The new model:
> mixed motivate by wave wavesq txgroupn il2018/fixed=wave wavesq txgroupn
>    wave*txgroupn il2018 wave*il2018/
>    print solution/random intercept | subject(rid) covtype(id).
>
> This executed but yielded different results because the wavesq contrasts
> were marked as redundant. This is  wrong. I had thought could get around
> this by naming wavesq using a with subcommand. However, this term was also
> marked as redundant. Again, wrong, seriously wrong.
>
> What is, is. The Test command seems to be the only alternative; however,
> in terms of examples, can covariates be accommodated in the Test command?
> Or, more generally, how do I get the comparisons I'm interested in?
>
> Thanks, Gene Maguin
>
> =====================
> To manage your subscription to SPSSX-L, send a message to

> LISTSERV@.UGA

>  (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





-----
--
Bruce Weaver
[hidden email]
http://sites.google.com/a/lakeheadu.ca/bweaver/

"When all else fails, RTFM."

NOTE: My Hotmail account is not monitored regularly.
To send me an e-mail, please use the address shown above.

--
Sent from: http://spssx-discussion.1045642.n5.nabble.com/

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

[Ryan]

Yes, the TEST subcommand (based on the coefficient matrix L) can be used to construct the contrast(s) of interest. 

Sent from my iPhone

On Aug 20, 2020, at 4:16 PM, Maguin, Eugene <[hidden email]> wrote:



This is the model statement I have been using:

mixed motivate with wave txgroupn il2018/fixed=wave wave*wave txgroupn

   wave*txgroupn il2018 wave*il2018/

   print solution/random intercept | subject(rid) covtype(id).

 

What I’d like to see are comparisons between values of txgroupn at values of wave. After looking at emmeans (and trying it out), I can’t use it unless the focal variables are as factors. I’m assuming no work-around for this. I then changed the model this:

mixed motivate by wave txgroupn il2018/fixed=wave wave*wave wavesq txgroupn

   wave*txgroupn il2018 wave*il2018/

   print solution/random intercept | subject(rid) covtype(id).

 

This wasn’t liked at all and the problem seemed to the wave*wave term. Ok, easy to fix, just compute wavesq (wave squared). The new model:

mixed motivate by wave wavesq txgroupn il2018/fixed=wave wavesq txgroupn

   wave*txgroupn il2018 wave*il2018/

   print solution/random intercept | subject(rid) covtype(id).

 

This executed but yielded different results because the wavesq contrasts were marked as redundant. This is  wrong. I had thought could get around this by naming wavesq using a with subcommand. However, this term was also marked as redundant. Again, wrong, seriously wrong.

 

What is, is. The Test command seems to be the only alternative; however, in terms of examples, can covariates be accommodated in the Test command? Or, more generally, how do I get the comparisons I’m interested in?

 

Thanks, Gene Maguin

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