trends in mixed models
Locked
 
|
Hello, I will be analyzing repeated measures data that are nested using mixed model (HLM) approach (the same measures collected 4 times for each student - pre-intervention and each year for three years post intervention - students nested in classroom nested in schools). Can I test the trends (linear or non-linear) in the mixed model analysis? I know I can do it in the regular RM ANOVA but i am unsure whether it can be done in mixed models...
Thanks so much, 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 |
|
|
Hi Bozena,
It seems to me that it can be handled through model evaluation and selection: 1) Specifying a model with a linear tread, and 2) specifying a second model with a non-linear trend. Then you can compare the two models through one of the approaches below: 1) Use of information model selection criteria: AIC, BIC, CAIC, etc. 2) Use of deviance difference test, assuming it is the maximum likelihood estimation that you use 3) Test of significance on the nonlinear term Hongwei "Patrick" Yang, University of Kentucky -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Zdaniuk, Bozena Sent: Wednesday, September 04, 2013 4:17 AM To: [hidden email] Subject: trends in mixed models Hello, I will be analyzing repeated measures data that are nested using mixed model (HLM) approach (the same measures collected 4 times for each student - pre-intervention and each year for three years post intervention - students nested in classroom nested in schools). Can I test the trends (linear or non-linear) in the mixed model analysis? I know I can do it in the regular RM ANOVA but i am unsure whether it can be done in mixed models... Thanks so much, 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 ===================== 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 |
|
|
In reply to this post by Zdaniuk, Bozena-3
Are you modeling the data as repeated measures or as growth curve? If the latter, you'd see a coefficient and SE and if linear time were random a variance and a SE (plus a covariance).
Gene Maguin -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Zdaniuk, Bozena Sent: Wednesday, September 04, 2013 4:17 AM To: [hidden email] Subject: trends in mixed models Hello, I will be analyzing repeated measures data that are nested using mixed model (HLM) approach (the same measures collected 4 times for each student - pre-intervention and each year for three years post intervention - students nested in classroom nested in schools). Can I test the trends (linear or non-linear) in the mixed model analysis? I know I can do it in the regular RM ANOVA but i am unsure whether it can be done in mixed models... Thanks so much, 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 ===================== 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 |
|
|
Well, it is a bit more complicated. We expect the students in the treatment condition to have a big jump in learning after the intervention (time 1 to t2) and then a slower growth (t2-3 difference will be smaller than t1-2 difference). We expect students in control condition to have a linear growth with scores getting bigger over time but at all three points lower than the treatment group. However, the biggest difference between treatment and control will be at t2 (post-test) and the difference between the two groups will become smaller at t3. I know how to test all this with post hoc tests in the regular GM Repeated Measures MANOVA but is there a way in MIXED MODELS to test those specific contrasts as being different from each other? Since I have students nested within the schools, i veer towards using MIXED MODELS...
Thanks so much for any help. Bozena -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Maguin, Eugene Sent: September 4, 2013 9:58 AM To: [hidden email] Subject: Re: trends in mixed models Are you modeling the data as repeated measures or as growth curve? If the latter, you'd see a coefficient and SE and if linear time were random a variance and a SE (plus a covariance). Gene Maguin -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Zdaniuk, Bozena Sent: Wednesday, September 04, 2013 4:17 AM To: [hidden email] Subject: trends in mixed models Hello, I will be analyzing repeated measures data that are nested using mixed model (HLM) approach (the same measures collected 4 times for each student - pre-intervention and each year for three years post intervention - students nested in classroom nested in schools). Can I test the trends (linear or non-linear) in the mixed model analysis? I know I can do it in the regular RM ANOVA but i am unsure whether it can be done in mixed models... Thanks so much, 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 ===================== 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 |
|
|
I really don't have any experience with multi-level repeated measures models using mixed. I have done analyses using a piecewise two group growth curve analysis using Mplus but not with the added complication of the school level. I looked back and found a message from Ryan Black, who among others, would be really qualified to respond to your question. I'd suggest you search the archives for messages from him and on the more general topic. His message follows. You may be able to adapt his reply to your needs although I think you are probably also interested in planned contrasts. Again, he has posted on that topic also. Gene Maguin
A random intercept model is mathematically equivalent to a model with a repeated statement with a specified compound symmetric variance-covariance matrix. Consequently, the following two pieces of code will yield identical results: MIXED y BY time /FIXED=time | SSTYPE(3) /METHOD=REML /RANDOM=INTERCEPT | SUBJECT(person) COVTYPE(VC). MIXED y BY time /FIXED=time | SSTYPE(3) /METHOD=REML /REPEATED=time | SUBJECT(person) COVTYPE(CS). Ryan Gene Maguin wrote: > > I have an odd question. A repeated measures design can be analyzed in > Mixed using the Repeated subcommand. The documentation has an example. > > My question is this. Can a repeated measures design be analyzed using > Mixed but WITHOUT the Repeated subcommand such that the same results > as an analysis using the Repeated subcommand? I'm assuming the answer > is yes and my followup questions are what is the command syntax and > what is the underlying alebraic model. -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Zdaniuk, Bozena Sent: Friday, September 06, 2013 12:46 AM To: [hidden email] Subject: Re: trends in mixed models Well, it is a bit more complicated. We expect the students in the treatment condition to have a big jump in learning after the intervention (time 1 to t2) and then a slower growth (t2-3 difference will be smaller than t1-2 difference). We expect students in control condition to have a linear growth with scores getting bigger over time but at all three points lower than the treatment group. However, the biggest difference between treatment and control will be at t2 (post-test) and the difference between the two groups will become smaller at t3. I know how to test all this with post hoc tests in the regular GM Repeated Measures MANOVA but is there a way in MIXED MODELS to test those specific contrasts as being different from each other? Since I have students nested within the schools, i veer towards using MIXED MODELS... Thanks so much for any help. Bozena -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Maguin, Eugene Sent: September 4, 2013 9:58 AM To: [hidden email] Subject: Re: trends in mixed models Are you modeling the data as repeated measures or as growth curve? If the latter, you'd see a coefficient and SE and if linear time were random a variance and a SE (plus a covariance). Gene Maguin -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Zdaniuk, Bozena Sent: Wednesday, September 04, 2013 4:17 AM To: [hidden email] Subject: trends in mixed models Hello, I will be analyzing repeated measures data that are nested using mixed model (HLM) approach (the same measures collected 4 times for each student - pre-intervention and each year for three years post intervention - students nested in classroom nested in schools). Can I test the trends (linear or non-linear) in the mixed model analysis? I know I can do it in the regular RM ANOVA but i am unsure whether it can be done in mixed models... Thanks so much, 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 ===================== 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 ===================== 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 |
|
|
In reply to this post by Zdaniuk, Bozena-3
I don't know which procedure works better, but I think you
will be well served if you run separate analyses for the separate hypotheses - so you get clearer tests and clearer measurements. The first question is the change from 1 to 2. Test it, and measure it. The second question is what happens from 2 to 4. That is a little delicate because the starting-levels are no longer equal, but the basic question is whether there are linear trends in the two groups, and if the trends are equal. The outcome is ambiguous if the Intervention group shows much regression-to-the-mean while the control group keeps improving. -- Rich Ulrich > Date: Fri, 6 Sep 2013 04:45:45 +0000 > From: [hidden email] > Subject: Re: trends in mixed models > To: [hidden email] > > Well, it is a bit more complicated. We expect the students in the treatment condition to have a big jump in learning after the intervention (time 1 to t2) and then a slower growth (t2-3 difference will be smaller than t1-2 difference). We expect students in control condition to have a linear growth with scores getting bigger over time but at all three points lower than the treatment group. However, the biggest difference between treatment and control will be at t2 (post-test) and the difference between the two groups will become smaller at t3. I know how to test all this with post hoc tests in the regular GM Repeated Measures MANOVA but is there a way in MIXED MODELS to test those specific contrasts as being different from each other? Since I have students nested within the schools, i veer towards using MIXED MODELS... > Thanks so much for any help. > Bozena > > -----Original Message----- > From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Maguin, Eugene > Sent: September 4, 2013 9:58 AM > To: [hidden email] > Subject: Re: trends in mixed models > > Are you modeling the data as repeated measures or as growth curve? If the latter, you'd see a coefficient and SE and if linear time were random a variance and a SE (plus a covariance). > Gene Maguin > > -----Original Message----- > From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Zdaniuk, Bozena > Sent: Wednesday, September 04, 2013 4:17 AM > To: [hidden email] > Subject: trends in mixed models > > Hello, I will be analyzing repeated measures data that are nested using mixed model (HLM) approach (the same measures collected 4 times for each student - pre-intervention and each year for three years post intervention - students nested in classroom nested in schools). Can I test the trends (linear or non-linear) in the mixed model analysis? I know I can do it in the regular RM ANOVA but i am unsure whether it can be done in mixed models... > Thanks so much, > Bozena > |
«
Return to SPSSX Discussion
|
1 view|%1 views
| Free forum by Nabble | Edit this page |