MIXED difference between unstructured and variance component covariance structures?
Locked
MIXED difference between unstructured and variance component covariance structures?
 
|
Dear All,
Am running several (19) mixed logistic regression models and wish to make minimum assumptions about covariance.
So have tried both variance component and unstructured
So what IS the difference?So what IS the difference between unstructured and variance component covariance matrices?
All data sets work fine with variance component
BUT only 8 will run with unstructured
simples has 1 repeated measure, most complex has 1 between (2 level) and 2 within ( 2level each)
complexity does not determine which data sets ‘run’
For data sets that run both, comparison gives:
F-values identical
df1 identical
df2 very large differences (some +, some-), and may be non-integer for either unstructured or variance component or both.
p-null, small differences of necessity as df2 are different - none greater than .008
as p-value differences are so small, one might just say why worry,
but I really would like to know
All help gratefully received
best
Diana
________________________________________
Professor Diana Kornbrot Work University of Hertfordshire College Lane, Hatfield, Hertfordshire AL10 9AB, UK +44 (0) 170 728 4626 [hidden email] http://dianakornbrot.wordpress.com/ http://go.herts.ac.uk/Diana_Kornbrot skype: kornbrotme Home 19 Elmhurst Avenue London N2 0LT, UK +44 (0) 208 444 2081 |
|
|
Diana, Please post your syntax. Ryan On Wed, Feb 8, 2017 at 9:12 AM, Kornbrot, Diana <[hidden email]> wrote:
|
Re: MIXED difference between unstructured and variance component covariance structures?
|
|
Dear Ryan, here is syntax. Data also available
Syntax
variance component
________________________________________
Professor Diana Kornbrot Work University of Hertfordshire College Lane, Hatfield, Hertfordshire AL10 9AB, UK +44 (0) 170 728 4626 [hidden email] http://dianakornbrot.wordpress.com/ http://go.herts.ac.uk/Diana_Kornbrot skype: kornbrotme Home 19 Elmhurst Avenue London N2 0LT, UK +44 (0) 208 444 2081 |
Re: MIXED difference between unstructured and variance component covariance structures?
Administrator
|
In reply to this post by Kornbrot, Diana
Here are some notes that might help. They appear to be geared to PROC MIXED in SAS, but should still be relevant.
https://onlinecourses.science.psu.edu/stat502/node/228 HTH.
--
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/). |
|
|
I see that the dialog help for GENLINMIXED lists the covariance types available but does not define them. You can see the mathematical definitions, however, by looking at the dialog help for MIXED. As you can see, Unstructured requires estimating many more covariance patterns, which may be the source of the problem. On Wed, Feb 8, 2017 at 9:04 AM, Bruce Weaver <[hidden email]> wrote: Here are some notes that might help. They appear to be geared to PROC MIXED |
Re: MIXED difference between unstructured and variance component covariance structures?
|
Administrator
|
Here's an IBM support page that shows the different covariance structures:
http://www.ibm.com/support/knowledgecenter/SSLVMB_20.0.0/com.ibm.spss.statistics.help/covariance_structures.htm It does not show a matrix for Variance Components, but instead refers the reader back to the Scaled Identity example. So with VC, each random effect has "constant variance" and "no correlation between any elements".
--
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/). |
Re: MIXED difference between unstructured and variance component covariance structures?
|
|
Many thanks
Shows everything but variance component! Clearly I need to look at covariance matrix generated in both cases. I now know 1. as I suspected unstructured is what I want -i.e. no constraints 2. df2 is lower for unstructured because more parameters are being estimated 3. in my data sets there appears to be little difference 4. unstructured is at limit of SPSS capability, will need to try other packages 5. unstructured is comparable to multivariate in GLM repeated??? or maybe the different offerings have different constraints on covariance matrix. need to do more reading! but many thanks again for your expert help best Diana > On 8 Feb 2017, at 20:27, Bruce Weaver <[hidden email]> wrote: > > Here's an IBM support page that shows the different covariance structures: > > http://www.ibm.com/support/knowledgecenter/SSLVMB_20.0.0/com.ibm.spss.statistics.help/covariance_structures.htm > > It does not show a matrix for Variance Components, but instead refers the > reader back to the Scaled Identity example. So with VC, each random effect > has "constant variance" and "no correlation between any elements". > > > > Jon Peck wrote >> I see that the dialog help for GENLINMIXED lists the covariance types >> available but does not define them. You can see the mathematical >> definitions, however, by looking at the dialog help for MIXED. As you can >> see, Unstructured requires estimating many more covariance patterns, which >> may be the source of the problem. >> >> On Wed, Feb 8, 2017 at 9:04 AM, Bruce Weaver < > >> bruce.weaver@ > >> > >> wrote: >> >>> Here are some notes that might help. They appear to be geared to PROC >>> MIXED >>> in SAS, but should still be relevant. >>> >>> https://onlinecourses.science.psu.edu/stat502/node/228 >>> >>> HTH. >>> >>> >>> Kornbrot, Diana wrote >>>> Dear All, >>>> >>>> Am running several (19) mixed logistic regression models and wish to >>> make >>>> minimum assumptions about covariance. >>>> So have tried both variance component and unstructured >>>> So what IS the difference?So what IS the difference between >>> unstructured >>>> and variance component covariance matrices? >>>> >>>> All data sets work fine with variance component >>>> BUT only 8 will run with unstructured >>>> simples has 1 repeated measure, most complex has 1 between (2 level) >>> and >>> 2 >>>> within ( 2level each) >>>> complexity does not determine which data sets ‘run’ >>>> >>>> For data sets that run both, comparison gives: >>>> F-values identical >>>> df1 identical >>>> df2 very large differences (some +, some-), and may be non-integer for >>>> either unstructured or variance component or both. >>>> p-null, small differences of necessity as df2 are different - none >>> greater >>>> than .008 >>>> as p-value differences are so small, one might just say why worry, but >>> I >>>> really would like to know >>>> >>>> All help gratefully received >>>> >>>> best >>>> >>>> Diana >>>> ________________________________________ >>>> Professor Diana Kornbrot >>>> Work >>>> University of Hertfordshire >>>> College Lane, Hatfield, Hertfordshire AL10 9AB, UK >>>> +44 (0) 170 728 4626 >>> >>>> d.e.kornbrot@.ac >>> >>>> >> <mailto: >>> >>>> d.e.kornbrot@.ac >>> >>>>> >>>> http://dianakornbrot.wordpress.com/ >>>> http://go.herts.ac.uk/Diana_Kornbrot >>>> skype: kornbrotme >>>> Home >>>> 19 Elmhurst Avenue >>>> London N2 0LT, UK >>>> +44 (0) 208 444 2081 >>>> >>>> >>>> >>>> >>>> ===================== >>>> 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 >>> > >> bweaver@ > >>> 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. >>> >>> -- >>> View this message in context: http://spssx-discussion. >>> 1045642.n5.nabble.com/MIXED-difference-between-unstructured-and-variance- >>> component-covariance-structures-tp5733784p5733791.html >>> Sent from the SPSSX Discussion mailing list archive at 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 >>> >> >> >> >> -- >> Jon K Peck > >> jkpeck@ > >> >> ===================== >> 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. > > -- > View this message in context: http://spssx-discussion.1045642.n5.nabble.com/MIXED-difference-between-unstructured-and-variance-component-covariance-structures-tp5733784p5733795.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 ________________________________________ Professor Diana Kornbrot Work University of Hertfordshire College Lane, Hatfield, Hertfordshire AL10 9AB, UK +44 (0) 170 728 4626 [hidden email] http://dianakornbrot.wordpress.com/ http://go.herts.ac.uk/Diana_Kornbrot skype: kornbrotme Home 19 Elmhurst Avenue London N2 0LT, UK +44 (0) 208 444 2081 ===================== 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 |
|
|
Diana,
The variance component structure specification on the REPEATED statement assumes that the correlation among repeated measurements within the same subject is zero. This is rarely a reasonable assumption. Upon inspection of your genlinmixed code, it looks like you have a doubly repeated measures multivariate model. The UNstructured option with the interaction of both within-subjects variables is appropriate for that type of model. It is possible that two structures modeled simultaneously would make sense as well (e.g. CS to account for correlation of k variables being measured at each time point and AR1 to account for temporal correlation). SPSS does not have that capability as far as I am aware. I am concerned that you are experiencing convergence issues. I would try to diagnose the issue before interpreting the results (e.g. sparse data, separation). Ryan Sent from my iPhone > On Feb 9, 2017, at 5:49 AM, Kornbrot, Diana <[hidden email]> wrote: > > Many thanks > Shows everything but variance component! > Clearly I need to look at covariance matrix generated in both cases. > I now know > 1. as I suspected unstructured is what I want -i.e. no constraints > 2. df2 is lower for unstructured because more parameters are being estimated > 3. in my data sets there appears to be little difference > 4. unstructured is at limit of SPSS capability, will need to try other packages > 5. unstructured is comparable to multivariate in GLM repeated??? or maybe the different offerings have different constraints on covariance matrix. need to do more reading! > > but many thanks again for your expert help > best > Diana >> On 8 Feb 2017, at 20:27, Bruce Weaver <[hidden email]> wrote: >> >> Here's an IBM support page that shows the different covariance structures: >> >> http://www.ibm.com/support/knowledgecenter/SSLVMB_20.0.0/com.ibm.spss.statistics.help/covariance_structures.htm >> >> It does not show a matrix for Variance Components, but instead refers the >> reader back to the Scaled Identity example. So with VC, each random effect >> has "constant variance" and "no correlation between any elements". >> >> >> >> Jon Peck wrote >>> I see that the dialog help for GENLINMIXED lists the covariance types >>> available but does not define them. You can see the mathematical >>> definitions, however, by looking at the dialog help for MIXED. As you can >>> see, Unstructured requires estimating many more covariance patterns, which >>> may be the source of the problem. >>> >>> On Wed, Feb 8, 2017 at 9:04 AM, Bruce Weaver < >> >>> bruce.weaver@ >> >>> > >>> wrote: >>> >>>> Here are some notes that might help. They appear to be geared to PROC >>>> MIXED >>>> in SAS, but should still be relevant. >>>> >>>> https://onlinecourses.science.psu.edu/stat502/node/228 >>>> >>>> HTH. >>>> >>>> >>>> Kornbrot, Diana wrote >>>>> Dear All, >>>>> >>>>> Am running several (19) mixed logistic regression models and wish to >>>> make >>>>> minimum assumptions about covariance. >>>>> So have tried both variance component and unstructured >>>>> So what IS the difference?So what IS the difference between >>>> unstructured >>>>> and variance component covariance matrices? >>>>> >>>>> All data sets work fine with variance component >>>>> BUT only 8 will run with unstructured >>>>> simples has 1 repeated measure, most complex has 1 between (2 level) >>>> and >>>> 2 >>>>> within ( 2level each) >>>>> complexity does not determine which data sets ‘run’ >>>>> >>>>> For data sets that run both, comparison gives: >>>>> F-values identical >>>>> df1 identical >>>>> df2 very large differences (some +, some-), and may be non-integer for >>>>> either unstructured or variance component or both. >>>>> p-null, small differences of necessity as df2 are different - none >>>> greater >>>>> than .008 >>>>> as p-value differences are so small, one might just say why worry, but >>>> I >>>>> really would like to know >>>>> >>>>> All help gratefully received >>>>> >>>>> best >>>>> >>>>> Diana >>>>> ________________________________________ >>>>> Professor Diana Kornbrot >>>>> Work >>>>> University of Hertfordshire >>>>> College Lane, Hatfield, Hertfordshire AL10 9AB, UK >>>>> +44 (0) 170 728 4626 >>>> >>>>> d.e.kornbrot@.ac >>>> >>>>> >>> <mailto: >>>> >>>>> d.e.kornbrot@.ac >>>> >>>>>> >>>>> http://dianakornbrot.wordpress.com/ >>>>> http://go.herts.ac.uk/Diana_Kornbrot >>>>> skype: kornbrotme >>>>> Home >>>>> 19 Elmhurst Avenue >>>>> London N2 0LT, UK >>>>> +44 (0) 208 444 2081 >>>>> >>>>> >>>>> >>>>> >>>>> ===================== >>>>> 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 >>>> >> >>> bweaver@ >> >>>> 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. >>>> >>>> -- >>>> View this message in context: http://spssx-discussion. >>>> 1045642.n5.nabble.com/MIXED-difference-between-unstructured-and-variance- >>>> component-covariance-structures-tp5733784p5733791.html >>>> Sent from the SPSSX Discussion mailing list archive at 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 >>>> >>> >>> >>> >>> -- >>> Jon K Peck >> >>> jkpeck@ >> >>> >>> ===================== >>> 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. >> >> -- >> View this message in context: http://spssx-discussion.1045642.n5.nabble.com/MIXED-difference-between-unstructured-and-variance-component-covariance-structures-tp5733784p5733795.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 > > ________________________________________ > Professor Diana Kornbrot > Work > University of Hertfordshire > College Lane, Hatfield, Hertfordshire AL10 9AB, UK > +44 (0) 170 728 4626 > [hidden email] > http://dianakornbrot.wordpress.com/ > http://go.herts.ac.uk/Diana_Kornbrot > skype: kornbrotme > Home > 19 Elmhurst Avenue > London N2 0LT, UK > +44 (0) 208 444 2081 > > > > > ===================== > 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 |