Mixed Model Repeated Measures
Locked
 
|
I am analyzing some data from a home visiting program. Because there are repeated measures, and some of those measures are missing at random, the literature suggests that I use a mixed model analysis in spss. I have some questions since I haven't done this before.
On the first slide (after analyze-mixed models-linear), I put in client ID as the subject variable and time as the repeated measure. What would be the appropriate "repeated covariance type" on this slide? I assume the repeated measure (nurturing assessment) would be correlated since the same assessment was given over time. On the second slide I have put in time, primary language, and ethnicity as fixed factors. What would I put in, if anything, for random effects? I also want to study dosage (# of home visits per client) and how it effects nurturing---but it is a continuous variable. How can I do this? I've thought since the # varies from 2 to 72 home visits that I could categorize the visits to low, medium, and high. Would that be appropriate to enter as a fixed variable? I do have number of months enrolled in the program and can enter that number in as a covariate. Thanks for any help that can be offered. |
Re: Mixed Model Repeated Measures
|
|
Haven't you posted on this before?
A better project/data structure description would be helpful. Employing some ESP, let's say you have a single group design, N unknown, in which a somebody or somebodies visited people in their homes. One or more measures were supposed to be collected at each visit but sometimes they weren't. The number of visits per home varied for various reasons but ranged between 2 and 72. Just out of curiosity, what does the distribution of visits look like? High right skew? How many have 72 visits? I may be wrong about this because I haven't run a repeated measures analysis with mixed on a dataset with missing data but my bet is that cases with missing data are going to be excluded. Your repeated statement will be something exactly like Repeated visit | subject(people) covtype(??). Thus each people will be expected to have 72 visit data points. My suggestion is to consider a growth curve model. As Bruce suggested to somebody else, I think it would be useful to dig Singer and Willet, Applied Longitudinal Data Analysis, out of the MN library and see if a growth model might work. Gene Maguin -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of mdaugher Sent: Tuesday, May 13, 2014 4:51 PM To: [hidden email] Subject: Mixed Model Repeated Measures I am analyzing some data from a home visiting program. Because there are repeated measures, and some of those measures are missing at random, the literature suggests that I use a mixed model analysis in spss. I have some questions since I haven't done this before. On the first slide (after analyze-mixed models-linear), I put in client ID as the subject variable and time as the repeated measure. What would be the appropriate "repeated covariance type" on this slide? I assume the repeated measure (nurturing assessment) would be correlated since the same assessment was given over time. On the second slide I have put in time, primary language, and ethnicity as fixed factors. What would I put in, if anything, for random effects? I also want to study dosage (# of home visits per client) and how it effects nurturing---but it is a continuous variable. How can I do this? I've thought since the # varies from 2 to 72 home visits that I could categorize the visits to low, medium, and high. Would that be appropriate to enter as a fixed variable? I do have number of months enrolled in the program and can enter that number in as a covariate. Thanks for any help that can be offered. -- View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Mixed-Model-Repeated-Measures-tp5726021.html Sent from the SPSSX Discussion mailing list archive at 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 ===================== 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 |
Re: Mixed Model Repeated Measures
|
|
This statement:
" I may be wrong about this because I haven't run a repeated measures analysis with mixed on a dataset with missing data but my bet is that cases with missing data are going to be excluded. Your repeated statement will be something exactly like Repeated visit | subject(people) covtype(??). Thus each people will be expected to have 72 visit data points. " Would be true when running GLM, but not MIXED. MIXED and GENLINMIXED use all available visit data points. Alex |
Re: Mixed Model Repeated Measures
|
|
Alex,
Thank you for setting me straight. A followup question. Does this imply that the within vcv matrix may have different Ns for different elements? I understand (or think I do) the computation implications of ‘all available data points’ in a growth curve model, but I don’t for a repeated measures.
Thanks, Gene Maguin From: SPSSX(r) Discussion [mailto:[hidden email]]
On Behalf Of Alex Reutter This statement:
|
Re: Mixed Model Repeated Measures
|
|
Right, so if you have time periods with
high numbers of missing values, then in the "Covariance Parameters"
table, you'll see higher standard errors for those estimates, perhaps to
the point where you can't really trust those estimates because you have
too many missing values.
Though when there are potentially 72 visits for each patient, I'm not sure that trying to use visit as a REPEATED variable will return anything helpful -- that's trying to estimate a 72x72 R matrix. Alex |
|
|
In reply to this post by Maguin, Eugene
Gene,
I am using the techniques discussed in Singer and Willet, Applied Longitudinal Data Analysis. I'm using a mixed model-linear analysis. I just don't know how best to enter # of home visits and program duration (# of months in the program) to see if they relate to the dependent variable of nurturing quality. Since both are continuous variables, do I code them and put them in as a fixed effect or should they be random effects or covariance effects? When I run the analysis using just time (1st assessment, 2nd assessment, etc), primary language and ethnicity I get significance with one primary language and one ethnicity. However, when I put in # of home visits and program duration, I get an error message and there is no significance. Also, it appears that in order to add #of home visits as a random effect, it also has to be entered as a fixed effect. As I said, I'm new to this type of analysis. |
|
|
In reply to this post by Maguin, Eugene
Gene, Run the code below and you will see that the main-diagonal elements of the residual covariance matrix are estimated using all possible data for each group by comparing the residual variance estimates from MIXED to the group-specific variances from MEANS.
HTH. Ryan -- *Generate Data. set seed 65923454. new file. input program. compute subject=-99. compute group = -99. leave subject to group. loop subject= 1 to 1000. loop group = 1 to 3. compute y = rv.normal(0,1). end case. end loop. end loop. end file. end input program. execute. USE ALL. COMPUTE filter_$=(uniform(1)<=.95). VARIABLE LABELS filter_$ 'Approximately 95% of the cases (SAMPLE)'. FORMATS filter_$ (f1.0). FILTER BY filter_$. EXECUTE. IF (filter_$=0) y=$SYSMIS. EXECUTE. delete variables filter_$. MIXED y BY group /FIXED=group | SSTYPE(3) /METHOD=REML /PRINT=R /REPEATED=group | SUBJECT(subject) COVTYPE(diag).
MEANS TABLES=y BY group /CELLS=VAR COUNT. On Wed, May 14, 2014 at 10:48 AM, Maguin, Eugene <[hidden email]> wrote:
|
|
|
In reply to this post by mdaugher
Since this is thread on mixed models with repeated measures, i figured i'd
place this here versus starting a new one I'm running a repeated measure analysis using mixed models. There are lots of missing data so mixed models makes the most sense. The study is a randomized trial of group (placebo = 0 and drug = 1) with a centered baseline DV as a covariate and followups for this DV each year for 5 years (evenly timed). MIXED DV with cen_BaseDV group time timesq /FIXED= cen_BaseDV group time time*cen_BaseDV time*group timesq | SSTYPE(3) /METHOD=REML /PRINT=G SOLUTION TESTCOV /Repeated = time | SUBJECT (subjectid) COVTYPE (AR1). The above appears to make the most sense for the data including the covariance structure. But, my question and dilemma is the whether a random intercept needs to be included and how does one determine the covariance structure for this? This statement: /RANDOM intercept time | SUBJECT(subjectid) COVType(AR1) ...with the above does not converge but some other covariance structures, like diagonal, do converge and lower the akaike compared with the model without the /random intercept. But, i can't quite make the decision as to whether or not it needs to be included and if so, how to determine the covariance structure. For the repeated factor, it was very clear from correlations and covariances over time that the AR1 makes sense. can someone recommend some diagnostics, pictures, or tables that might help me make these decisions? Thanks Carol -- 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 |
Re: Mixed Model Repeated Measures
|
|
hello, in my understanding, you either do /repeated or /random, not both. Random is better if the time is more of a continuous variable whereas /repeated is better when time is more of a categorical variable.
my two cents. HTH. bozena ________________________________________ From: SPSSX(r) Discussion [[hidden email]] on behalf of parisec [[hidden email]] Sent: Tuesday, September 25, 2018 1:02 PM To: [hidden email] Subject: Re: Mixed Model Repeated Measures Since this is thread on mixed models with repeated measures, i figured i'd place this here versus starting a new one I'm running a repeated measure analysis using mixed models. There are lots of missing data so mixed models makes the most sense. The study is a randomized trial of group (placebo = 0 and drug = 1) with a centered baseline DV as a covariate and followups for this DV each year for 5 years (evenly timed). MIXED DV with cen_BaseDV group time timesq /FIXED= cen_BaseDV group time time*cen_BaseDV time*group timesq | SSTYPE(3) /METHOD=REML /PRINT=G SOLUTION TESTCOV /Repeated = time | SUBJECT (subjectid) COVTYPE (AR1). The above appears to make the most sense for the data including the covariance structure. But, my question and dilemma is the whether a random intercept needs to be included and how does one determine the covariance structure for this? This statement: /RANDOM intercept time | SUBJECT(subjectid) COVType(AR1) ...with the above does not converge but some other covariance structures, like diagonal, do converge and lower the akaike compared with the model without the /random intercept. But, i can't quite make the decision as to whether or not it needs to be included and if so, how to determine the covariance structure. For the repeated factor, it was very clear from correlations and covariances over time that the AR1 makes sense. can someone recommend some diagnostics, pictures, or tables that might help me make these decisions? Thanks Carol -- 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 ===================== 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 |
|
|
Thank you. I've been using the book "multilevel longitudinal modeling with
IBM spss" by Heck, Thomas, and Tabata. It uses both in the example i'm following which is why i'm questioning. The /repeated alone model is easier for me to wrap my head around. -- 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 |
|
|
In reply to this post by parisec
Carol,
It is possible to model correlated residuals which follow a first-order autoregressive structure, conditional on subject-specific intercepts. You would capture this by specifying REPEATED and RANDOM statements as follows: /RANDOM intercept SUBJECT(subjectid) /REPEATED = time | SUBJECT(subjectid) COVTYPE(AR1) See: Littell, RC; Pendergast, J; Natarajan, R (2000) Modelling covariance structure in the analysis of repeated measures data. Statistics in Medicine 19:1793-1819. Ryan > On Sep 25, 2018, at 4:02 PM, parisec <[hidden email]> wrote: > > Since this is thread on mixed models with repeated measures, i figured i'd > place this here versus starting a new one > > I'm running a repeated measure analysis using mixed models. There are lots > of missing data so mixed models makes the most sense. The study is a > randomized trial of group > (placebo = 0 and drug = 1) with a centered baseline DV as a covariate and > followups for this DV each year for 5 years (evenly timed). > > > MIXED DV with cen_BaseDV group time timesq > /FIXED= cen_BaseDV group time time*cen_BaseDV time*group timesq | > SSTYPE(3) > /METHOD=REML > /PRINT=G SOLUTION TESTCOV > /Repeated = time | SUBJECT (subjectid) COVTYPE (AR1). > > The above appears to make the most sense for the data including the > covariance structure. But, my question and dilemma is the whether a random > intercept needs to be included and how does one determine the covariance > structure for this? > > This statement: > /RANDOM intercept time | SUBJECT(subjectid) COVType(AR1) > > ...with the above does not converge but some other covariance structures, > like diagonal, do converge and lower the akaike compared with the model > without the /random intercept. > > But, i can't quite make the decision as to whether or not it needs to be > included and if so, how to determine the covariance structure. For the > repeated factor, it was very clear from correlations and covariances over > time that the AR1 makes sense. > > can someone recommend some diagnostics, pictures, or tables that might help > me make these decisions? > > Thanks > Carol > > > > > > > > > -- > 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 ===================== 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 |
Re: Mixed Model Repeated Measures
Administrator
|
In reply to this post by parisec
Ryan has given a good answer to your question. This is an aside. Note that
when you are using MIXED, there is no need to compute a timesq variable, because you express it as time*time on the FIXED sub-cmmand. MIXED DV with cen_BaseDV group time /FIXED= cen_BaseDV group time time*cen_BaseDV time*group time*time | SSTYPE(3) HTH. parisec wrote > Since this is thread on mixed models with repeated measures, i figured i'd > place this here versus starting a new one > > I'm running a repeated measure analysis using mixed models. There are lots > of missing data so mixed models makes the most sense. The study is a > randomized trial of group > (placebo = 0 and drug = 1) with a centered baseline DV as a covariate and > followups for this DV each year for 5 years (evenly timed). > > > MIXED DV with cen_BaseDV group time timesq > /FIXED= cen_BaseDV group time time*cen_BaseDV time*group timesq | > SSTYPE(3) > /METHOD=REML > /PRINT=G SOLUTION TESTCOV > /Repeated = time | SUBJECT (subjectid) COVTYPE (AR1). > > The above appears to make the most sense for the data including the > covariance structure. But, my question and dilemma is the whether a random > intercept needs to be included and how does one determine the covariance > structure for this? > > This statement: > /RANDOM intercept time | SUBJECT(subjectid) COVType(AR1) > > ...with the above does not converge but some other covariance structures, > like diagonal, do converge and lower the akaike compared with the model > without the /random intercept. > > But, i can't quite make the decision as to whether or not it needs to be > included and if so, how to determine the covariance structure. For the > repeated factor, it was very clear from correlations and covariances over > time that the AR1 makes sense. > > can someone recommend some diagnostics, pictures, or tables that might > help > me make these decisions? > > Thanks > Carol > > > > > > > > > -- > Sent from: http://spssx-discussion.1045642.n5.nabble.com/ > > ===================== > 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
--
Bruce Weaver bweaver@lakeheadu.ca http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." PLEASE NOTE THE FOLLOWING: 1. My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. 2. The SPSSX Discussion forum on Nabble is no longer linked to the SPSSX-L listserv administered by UGA (https://listserv.uga.edu/). |
|
|
In reply to this post by Ryan
I just tried that option and got a non-convergence warning. I can change the
AR1 and make it work but i i'm wondering if I really need this in the model. Is there a definitive diagnostic i'm missing here? thanks. Ryan Black wrote > Carol, > > It is possible to model correlated residuals which follow a first-order > autoregressive structure, conditional on subject-specific intercepts. You > would capture this by specifying REPEATED and RANDOM statements as > follows: > > /RANDOM intercept SUBJECT(subjectid) > /REPEATED = time | SUBJECT(subjectid) COVTYPE(AR1) > > See: > > Littell, RC; Pendergast, J; Natarajan, R (2000) Modelling covariance > structure in the analysis of repeated measures data. Statistics in > Medicine > 19:1793-1819. > > Ryan -- 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 |
|
|
In reply to this post by Bruce Weaver
nice! one less compute is always a good thing.
thanks! Bruce Weaver wrote > Ryan has given a good answer to your question. This is an aside. Note > that > when you are using MIXED, there is no need to compute a timesq variable, > because you express it as time*time on the FIXED sub-cmmand. > > MIXED DV with cen_BaseDV group time > /FIXED= cen_BaseDV group time time*cen_BaseDV time*group time*time > | > SSTYPE(3) > > HTH. -- 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 |
|
|
In reply to this post by parisec
Carol,
The short answer is no—You do not always need to specify random intercepts with repeated measures data. The long answer deals with the need to model dependencies in your data through the random cov matrix and/or the residual cov matrix. I cannot determine why you are receiving non-convergence from here, but it could very well be that you are specifying a model that is too complex for the data at hand. Ryan Sent from my iPhone > On Sep 25, 2018, at 11:33 PM, parisec <[hidden email]> wrote: > > I just tried that option and got a non-convergence warning. I can change the > AR1 and make it work but i i'm wondering if I really need this in the > model. Is there a definitive diagnostic i'm missing here? > > thanks. > > > Ryan Black wrote >> Carol, >> >> It is possible to model correlated residuals which follow a first-order >> autoregressive structure, conditional on subject-specific intercepts. You >> would capture this by specifying REPEATED and RANDOM statements as >> follows: >> >> /RANDOM intercept SUBJECT(subjectid) >> /REPEATED = time | SUBJECT(subjectid) COVTYPE(AR1) >> >> See: >> >> Littell, RC; Pendergast, J; Natarajan, R (2000) Modelling covariance >> structure in the analysis of repeated measures data. Statistics in >> Medicine >> 19:1793-1819. >> >> Ryan > > > > > > -- > 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 ===================== 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 think i may have figured out why the model won't run with the random
intercept and the AR1 for the repeated parameter. My DV was measured 6 times. When i use the baseline value as the covariate, with the other 5 times as a repeated factor, i get the error. If i leave out the baseline DV, i don'tt get the error. Is it logical to think that because i'm adjusting for the baseline of the DV, i no longer have random intercepts since i'm adjusting the model to account for these differences at baseline? -- 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 |
|
|
Carol:
It’s difficult to help from here but I’ll try. 1. Are the DV values consecutive whole numbers starting at zero as follows?: 0,1,2,3,4,5 2. Did you remove baseline scores from the DV when you decided to treat them as a covariate? If you did remove them, do the DV values now range from “1” to “5”? Ryan > On Oct 1, 2018, at 5:48 PM, parisec <[hidden email]> wrote: > > I think i may have figured out why the model won't run with the random > intercept and the AR1 for the repeated parameter. > > My DV was measured 6 times. When i use the baseline value as the covariate, > with the other 5 times as a repeated factor, i get the error. If i leave out > the baseline DV, i don'tt get the error. > > Is it logical to think that because i'm adjusting for the baseline of the > DV, i no longer have random intercepts since i'm adjusting the model to > account for these differences at baseline? > > > > -- > 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 ===================== 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 Black wrote
> Carol: > > It’s difficult to help from here but I’ll try. > > 1. Are the DV values consecutive whole numbers starting at zero as > follows?: 0,1,2,3,4,5 > > Yes. They were taken at baseline (0) and years 1,2,3,4,5 > > 2. Did you remove baseline scores from the DV when you decided to treat > them as a covariate? If you did remove them, do the DV values now range > from “1” to “5”? > > Yes. When i created the DV using the values from years 1 through 5 and > kept 0 as the covariate. > > > Ryan > >> On Oct 1, 2018, at 5:48 PM, parisec < > PariseC@ > > wrote: >> >> I think i may have figured out why the model won't run with the random >> intercept and the AR1 for the repeated parameter. >> >> My DV was measured 6 times. When i use the baseline value as the >> covariate, >> with the other 5 times as a repeated factor, i get the error. If i leave >> out >> the baseline DV, i don'tt get the error. >> >> Is it logical to think that because i'm adjusting for the baseline of the >> DV, i no longer have random intercepts since i'm adjusting the model to >> account for these differences at baseline? >> >> >> >> -- >> Sent from: http://spssx-discussion.1045642.n5.nabble.com/ >> >> ===================== >> 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 > > ===================== > 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 -- 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 |
|
|
Carol,
I don’t think treating baseline scores as a fixed effect covariate in the way you have described should impact estimating subject-specific intercepts or other parameter estimates, from a mathematical perspective without knowing the nuances of your data. I have estimated many a model with random intercepts in which fixed effects covariates were included. But maybe there is something unique to your data of which I am clearly unaware. I would consider making the first time-point in your DV = “0” and increasing sequentially to the final time point (“4”). This will impact interpretation of the fixed intercept term, of course. Ryan > On Oct 2, 2018, at 4:25 PM, parisec <[hidden email]> wrote: > > Ryan Black wrote >> Carol: >> >> It’s difficult to help from here but I’ll try. >> >> 1. Are the DV values consecutive whole numbers starting at zero as >> follows?: 0,1,2,3,4,5 >> >> Yes. They were taken at baseline (0) and years 1,2,3,4,5 >> >> 2. Did you remove baseline scores from the DV when you decided to treat >> them as a covariate? If you did remove them, do the DV values now range >> from “1” to “5”? >> >> Yes. When i created the DV using the values from years 1 through 5 and >> kept 0 as the covariate. >> >> >> Ryan >> >>> On Oct 1, 2018, at 5:48 PM, parisec < > >> PariseC@ > >> > wrote: >>> >>> I think i may have figured out why the model won't run with the random >>> intercept and the AR1 for the repeated parameter. >>> >>> My DV was measured 6 times. When i use the baseline value as the >>> covariate, >>> with the other 5 times as a repeated factor, i get the error. If i leave >>> out >>> the baseline DV, i don'tt get the error. >>> >>> Is it logical to think that because i'm adjusting for the baseline of the >>> DV, i no longer have random intercepts since i'm adjusting the model to >>> account for these differences at baseline? >>> >>> >>> >>> -- >>> Sent from: http://spssx-discussion.1045642.n5.nabble.com/ >>> >>> ===================== >>> 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 >> >> ===================== >> 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 > > > > > > -- > 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 ===================== 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 |
«
Return to SPSSX Discussion
|
1 view|%1 views
| Free forum by Nabble | Edit this page |