Repeated measures with unequal intervals
Locked
 
|
Hi,
I'm a complete novice to longitudinal analysis and I've muddled my way through some helpful texts but one question I fail to find an answer to is: how do I control for the fact that I have unequal intervals between assessment points for my subjects? It's a longitudinal cohort study where children are assessed at three time points (though a couple have four as they participated in a sub-study), usually at 18/24 months, 36 months and 60 months OR 30 months, 48 months and 60 months (they are recruited after they receive their diagnosis of cer.ebral palsy). There are many missing data points due to difficulty collecting data in this population. How do I code for this? Is it even possible? Currently I used assessment number (1-4) in the repeated box and added age at assessment as a covariate (YR_agecentered).I have no idea if this is correct. I'm trying to find out how fat free mass index (FFMI- continuous) changes over time and if differences in gross motor function (categorical mild (0) vs mod/severe (1)) causes differences in FFMI change, also seeing if sex (0,1) and feeding difficulties (0,1) alters the relationship. When I run the below code I'm told convergence is not achieved. I'm not sure if it is because I need to enter the data a different way or if it is because my data is insufficient. Thank you for any input! Kind regards, Stina MIXED FFMI BY YN_eatdrinkdiff Sex GMFCS_12vs345 WITH YR_agecentered SQ_YR_agecentered /CRITERIA=CIN(95) MXITER(200) MXSTEP(15) SCORING(1) SINGULAR(0.000000000001) HCONVERGE(0, ABSOLUTE) LCONVERGE(0, ABSOLUTE) PCONVERGE(0.000001, ABSOLUTE) /FIXED=GMFCS_12vs345 Sex YN_eatdrinkdiff YR_agecentered SQ_YR_agecentered YR_agecentered*GMFCS_12vs345 YR_agecentered*Sex YR_agecentered*YN_eatdrinkdiff | SSTYPE(3) /METHOD=ML /PRINT=CORB COVB G SOLUTION TESTCOV /RANDOM=INTERCEPT YR_agecentered | SUBJECT(ID) COVTYPE(UN) /REPEATED=AxNumberRank | SUBJECT(ID) COVTYPE(UN). |
Re: Repeated measures with unequal intervals
Administrator
|
"When I run the below code I'm told convergence is not achieved. I'm not sure if it is because I need to enter the data a different way or if it is because my data is insufficient."
Please elaborate (how are the data entered)? How many cases are there? Please tell us how these [YR_agecentered SQ_YR_agecentered] are created? Does the model converge WITHOUT the interactions? --
Please reply to the list and not to my personal email.
Those desiring my consulting or training services please feel free to email me. --- "Nolite dare sanctum canibus neque mittatis margaritas vestras ante porcos ne forte conculcent eas pedibus suis." Cum es damnatorum possederunt porcos iens ut salire off sanguinum cliff in abyssum?" |
|
|
Thanks for looking at my problem!
Data is entered with one line per datapoint (ie one subject ID can have up to 4 lines). There are 150 subjects with 317 cases. I think the problem is that only 51 children have 3 assessments, 53 children have 2 assessments, 41 children only have 1 assessment, then 3 children who have 4 assessments so LOTS of missing data. Is it a problem that e.g. assessment 1 could be 18 or 30 or 48 months and so on? That the age of ax 1-4 varies for each subject? The outcome variable FFMI ranges from 8 to 18, with a mean of 12 and SD of 1.3. YR_agecentered is their age at assessment in years (found by dividing months by 12) and detracting the youngest age found in study (so that child's age YR_agecentered is 0). SQ_YR_agecentered is just (YR_agecentered*YR_agecentered). Without interactions, convergence is still not achieved: MIXED FFMI BY GMFCS_12vs345 Sex YN_eatdrinkdiff WITH YR_agecentered SQ_YR_agecentered /CRITERIA=CIN(95) MXITER(100) MXSTEP(10) SCORING(1) SINGULAR(0.000000000001) HCONVERGE(0, ABSOLUTE) LCONVERGE(0, ABSOLUTE) PCONVERGE(0.000001, ABSOLUTE) /FIXED=GMFCS_12vs345 Sex YN_eatdrinkdiff | SSTYPE(3) /METHOD=ML /PRINT=COVB G SOLUTION /RANDOM=INTERCEPT YR_agecentered | SUBJECT(ID) COVTYPE(UN) /REPEATED=AxNumberRank | SUBJECT(ID) COVTYPE(UN). I'm starting to think I need to change by research question and use a different analysis... Hope I have adequately answered your questions! Thanks again, Stina |
Re: Repeated measures with unequal intervals
|
Administrator
|
Not really my area of expertise, but ponder this: What do subjects with a single assessment contribute to the analysis?
Please reply to the list and not to my personal email.
Those desiring my consulting or training services please feel free to email me. --- "Nolite dare sanctum canibus neque mittatis margaritas vestras ante porcos ne forte conculcent eas pedibus suis." Cum es damnatorum possederunt porcos iens ut salire off sanguinum cliff in abyssum?" |
Re: Repeated measures with unequal intervals
|
|
In reply to this post by stina
This seems more like an issue of how you want to clock time. What would you
expect to be important in your study - age of subject, months since baseline assessment or assessment number? That should be your time variable in the longitudinal study. Unequal time intervals is not something that is problematic. peter -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of stina Sent: Wednesday, January 07, 2015 12:09 PM To: [hidden email] Subject: Re: [SPSSX-L] Repeated measures with unequal intervals Thanks for looking at my problem! Data is entered with one line per datapoint (ie one subject ID can have up to 4 lines). There are 150 subjects with 317 cases. I think the problem is that only 51 children have 3 assessments, 53 children have 2 assessments, 41 children only have 1 assessment, then 3 children who have 4 assessments so LOTS of missing data. Is it a problem that e.g. assessment 1 could be 18 or 30 or 48 months and so on? That the age of ax 1-4 varies for each subject? The outcome variable FFMI ranges from 8 to 18, with a mean of 12 and SD of 1.3. YR_agecentered is their age at assessment in years (found by dividing months by 12) and detracting the youngest age found in study (so that child's age YR_agecentered is 0). SQ_YR_agecentered is just (YR_agecentered*YR_agecentered). Without interactions, convergence is still not achieved: MIXED FFMI BY GMFCS_12vs345 Sex YN_eatdrinkdiff WITH YR_agecentered SQ_YR_agecentered /CRITERIA=CIN(95) MXITER(100) MXSTEP(10) SCORING(1) SINGULAR(0.000000000001) HCONVERGE(0, ABSOLUTE) LCONVERGE(0, ABSOLUTE) PCONVERGE(0.000001, ABSOLUTE) /FIXED=GMFCS_12vs345 Sex YN_eatdrinkdiff | SSTYPE(3) /METHOD=ML /PRINT=COVB G SOLUTION /RANDOM=INTERCEPT YR_agecentered | SUBJECT(ID) COVTYPE(UN) /REPEATED=AxNumberRank | SUBJECT(ID) COVTYPE(UN). I'm starting to think I need to change by research question and use a different analysis... Hope I have adequately answered your questions! Thanks again, Stina -- View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Repeated-measures-with-unequal -intervals-tp5728361p5728374.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 |
|
|
Thanks Peter,
I suppose I have done it correctly then? Age at assessment is what I expect to be important. But since everyone is at a slightly different age at ax I used ax number 1-4 in REPEATED and YR_Agecentered in RANDOM. I guess I was hoping there was a way of making the model work - now off to find a different way of using the data! (cursing study designers....) Thanks for your help! Stina |
Re: Repeated measures with unequal intervals
|
|
I am not very familiar with the MIXED code. But, convergence problems are
almost always related to the complexity of the model, especially the random part and/or the variance of intercept/slope (usually slope). --Try increasing MXITER and MXSTEP. This will allow the model to run more iterations. --Simplify random/repeated statements. COVTYPE(UN) is very general and probably estimates too many parameters for your model given amount of data. Maybe the slope shouldn't be random (variance close to 0)? Maybe you don't need both RANDOM and REPEATED statements? --If age is in fractions of years, you probably have mostly singletons with regards to age - and that will likely be a problem. I don't know MIXED well enough to give you any other specifics with regards to your code, but maybe these give you a few things to consider that may ultimately help with the model. peter -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of stina Sent: Wednesday, January 07, 2015 3:06 PM To: [hidden email] Subject: Re: [SPSSX-L] Repeated measures with unequal intervals Thanks Peter, I suppose I have done it correctly then? Age at assessment is what I expect to be important. But since everyone is at a slightly different age at ax I used ax number 1-4 in REPEATED and YR_Agecentered in RANDOM. I guess I was hoping there was a way of making the model work - now off to find a different way of using the data! (cursing study designers....) Thanks for your help! Stina -- View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Repeated-measures-with-unequal -intervals-tp5728361p5728378.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: Repeated measures with unequal intervals
|
|
In reply to this post by stina
I had written:
--If age is in fractions of years, you probably have mostly singletons with regards to age - and that will likely be a problem. This probably isn't true, at least not with regard to conversion of the model. I was thinking of something different, sorry about that. Just ignore. The others may be useful, though. peter -----Original Message----- From: Peter Link [mailto:[hidden email]] Sent: Wednesday, January 07, 2015 3:23 PM To: '[hidden email]' Subject: RE: [SPSSX-L] Repeated measures with unequal intervals I am not very familiar with the MIXED code. But, convergence problems are almost always related to the complexity of the model, especially the random part and/or the variance of intercept/slope (usually slope). --Try increasing MXITER and MXSTEP. This will allow the model to run more iterations. --Simplify random/repeated statements. COVTYPE(UN) is very general and probably estimates too many parameters for your model given amount of data. Maybe the slope shouldn't be random (variance close to 0)? Maybe you don't need both RANDOM and REPEATED statements? --If age is in fractions of years, you probably have mostly singletons with regards to age - and that will likely be a problem. I don't know MIXED well enough to give you any other specifics with regards to your code, but maybe these give you a few things to consider that may ultimately help with the model. peter -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of stina Sent: Wednesday, January 07, 2015 3:06 PM To: [hidden email] Subject: Re: [SPSSX-L] Repeated measures with unequal intervals Thanks Peter, I suppose I have done it correctly then? Age at assessment is what I expect to be important. But since everyone is at a slightly different age at ax I used ax number 1-4 in REPEATED and YR_Agecentered in RANDOM. I guess I was hoping there was a way of making the model work - now off to find a different way of using the data! (cursing study designers....) Thanks for your help! Stina -- View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Repeated-measures-with-unequal -intervals-tp5728361p5728378.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 |
|
|
In reply to this post by peter link
Thanks,
Running the model without REPEATED, just RANDOM (YR_agecenetered) did allow the model to converge. This is where I get a bit confused about whether or not it recognises that there are repeated measures still? I guess the ID's are still showing that there is... I'll have a bit more of a play with it! Maybe there is hope! Stina |
Re: Repeated measures with unequal intervals
|
|
A final recommendation - UCLA ATS - google this. They have textbook examples
using SPSS. One of these is for a book by Singer and Willett on applied longitudinal analysis (pick this up from your library or buy it - it's a good beginner text, not overly technical but very thorough). It may have examples that are similar to your research. Code and output with some comments are provided. You seem like you are on the right track. Keep going and good luck. peter -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of stina Sent: Wednesday, January 07, 2015 4:20 PM To: [hidden email] Subject: Re: [SPSSX-L] Repeated measures with unequal intervals Thanks, Running the model without REPEATED, just RANDOM (YR_agecenetered) did allow the model to converge. This is where I get a bit confused about whether or not it recognises that there are repeated measures still? I guess the ID's are still showing that there is... I'll have a bit more of a play with it! Maybe there is hope! Stina -- View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Repeated-measures-with-unequal -intervals-tp5728361p5728385.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 |
«
Return to SPSSX Discussion
|
1 view|%1 views
| Free forum by Nabble | Edit this page |