Temporal stability using multilevel modeling

classic Classic list List threaded Threaded
8 messages Options
Reply | Threaded
Open this post in threaded view
|

Temporal stability using multilevel modeling

Oliver
Hi everyone,

I'm interested in examining the temporal stability of a behavior that was
measured at 6 different time points (i.e., once a month for 6 consecutive
months). My sample includes 100 patients, and most of them have data on the
6 time points.  I have three questions:

1) For the whole sample, would it be possible to use AR1 as a measure of
temporal stability of the behavior ? Given that AR1 has often been described
as a measure of "temporal dependency", this seems to make sense. It is
important to note, however, that there are minor between-person variations
in time intervals in-between each of the assessment time points, which is
typical in clinical research. Perhaps AR1 requires a pure "time-series"
design with equally spaced intervals in between assessments ?

2) My behavior of interest is a binary variable (i.e., presence/absence). I
tried to get the AR1 using the GLMM procedure in SPSS, and the AR1 is not
generated. Is it possible to get AR1 with a binary outcome ? If so, is there
any syntax specification needed ? If not, is there an alternative to AR1
with a binary outcome ?

3) The second goal of my study is to examine the within-person (i.e., Level
1) predictors of behavior stability. Therefore, I was wondering whether it
would be possible to derive a within-person indicator (e.g., AR1) of
behavior stability ? If so, I could use this indicator as an outcome
measure, which could then be predicted by other within-person (i.e., Level
1) variables.

Thanks in advance to all of you for your help.
Oliver



--
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
Reply | Threaded
Open this post in threaded view
|

Re: Temporal stability using multilevel modeling

Andy W
I don't have very good answers to your more general question, but just wanted
to comment on the use of AR coefficients as a measure of temporal stability
-- I don't think that is a very good idea. For a few examples, a flat line
would have an AR coefficient of 0 (if the equation had an intercept).
Without an intercept, a flat line would have an AR coefficient of 1. However
an AR coefficient of 1 with an additional error term is a random walk, which
is not mean reverting and so is quite the opposite of stable.

Long story short I don't think AR terms generally map to stability.  



-----
Andy W
[hidden email]
http://andrewpwheeler.wordpress.com/
--
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
Andy W
apwheele@gmail.com
http://andrewpwheeler.wordpress.com/
Reply | Threaded
Open this post in threaded view
|

Re: Temporal stability using multilevel modeling

Rich Ulrich
In reply to this post by Oliver

Like Andy, I'm not speaking to the SPSS questions, either, and I think

AR1 won't be useful. Here are other reasons.  Dichotomous measures

need more cases to match the power achieved by continuous measures.


Six points is an awfully small time series for continuous data for using "time

series methods", I think.  No doubt there are exceptions. You do have 100

parallel series ... which isn't the standard TS problem. My practical experience

with Time series is zero, but, unfortunately for your prospect of getting a better

answer here, we've dealt with very few TS problems, and I suspect that we

have no experts here of that specialty.


The stock markets call their instability "lability" and they measure it from the

standard deviation or variance of stock prices. Variance of a dichotomy is computed

from its mean (i.e., proportion), so I suggest that you get a transparent and obvious

measure of stability/volatility by counting and reporting the number of changes

observed for each ID.   (By the way, if most of your scores for a subject are identical,

at zero or 1, you have a different problem of lack-of-usable-variance for analysis.)


--

Rich Ulrich


From: SPSSX(r) Discussion <[hidden email]> on behalf of Oliver <[hidden email]>
Sent: Saturday, June 23, 2018 9:22:42 AM
To: [hidden email]
Subject: Temporal stability using multilevel modeling
 
Hi everyone,

I'm interested in examining the temporal stability of a behavior that was
measured at 6 different time points (i.e., once a month for 6 consecutive
months). My sample includes 100 patients, and most of them have data on the
6 time points.  I have three questions:

1) For the whole sample, would it be possible to use AR1 as a measure of
temporal stability of the behavior ? Given that AR1 has often been described
as a measure of "temporal dependency", this seems to make sense. It is
important to note, however, that there are minor between-person variations
in time intervals in-between each of the assessment time points, which is
typical in clinical research. Perhaps AR1 requires a pure "time-series"
design with equally spaced intervals in between assessments ?

2) My behavior of interest is a binary variable (i.e., presence/absence). I
tried to get the AR1 using the GLMM procedure in SPSS, and the AR1 is not
generated. Is it possible to get AR1 with a binary outcome ? If so, is there
any syntax specification needed ? If not, is there an alternative to AR1
with a binary outcome ?

3) The second goal of my study is to examine the within-person (i.e., Level
1) predictors of behavior stability. Therefore, I was wondering whether it
would be possible to derive a within-person indicator (e.g., AR1) of
behavior stability ? If so, I could use this indicator as an outcome
measure, which could then be predicted by other within-person (i.e., Level
1) variables.

Thanks in advance to all of you for your help.
Oliver



--
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
Reply | Threaded
Open this post in threaded view
|

Re: Temporal stability using multilevel modeling

Oliver

Rich & Andy,


Thank you for sharing your thoughts about this. That's very useful.


Best,

MO.



From: Rich Ulrich <[hidden email]>
Sent: Saturday, June 23, 2018 8:12 PM
To: [hidden email]; Oliver
Subject: Re: Temporal stability using multilevel modeling
 

Like Andy, I'm not speaking to the SPSS questions, either, and I think

AR1 won't be useful. Here are other reasons.  Dichotomous measures

need more cases to match the power achieved by continuous measures.


Six points is an awfully small time series for continuous data for using "time

series methods", I think.  No doubt there are exceptions. You do have 100

parallel series ... which isn't the standard TS problem. My practical experience

with Time series is zero, but, unfortunately for your prospect of getting a better

answer here, we've dealt with very few TS problems, and I suspect that we

have no experts here of that specialty.


The stock markets call their instability "lability" and they measure it from the

standard deviation or variance of stock prices. Variance of a dichotomy is computed

from its mean (i.e., proportion), so I suggest that you get a transparent and obvious

measure of stability/volatility by counting and reporting the number of changes

observed for each ID.   (By the way, if most of your scores for a subject are identical,

at zero or 1, you have a different problem of lack-of-usable-variance for analysis.)


--

Rich Ulrich


From: SPSSX(r) Discussion <[hidden email]> on behalf of Oliver <[hidden email]>
Sent: Saturday, June 23, 2018 9:22:42 AM
To: [hidden email]
Subject: Temporal stability using multilevel modeling
 
Hi everyone,

I'm interested in examining the temporal stability of a behavior that was
measured at 6 different time points (i.e., once a month for 6 consecutive
months). My sample includes 100 patients, and most of them have data on the
6 time points.  I have three questions:

1) For the whole sample, would it be possible to use AR1 as a measure of
temporal stability of the behavior ? Given that AR1 has often been described
as a measure of "temporal dependency", this seems to make sense. It is
important to note, however, that there are minor between-person variations
in time intervals in-between each of the assessment time points, which is
typical in clinical research. Perhaps AR1 requires a pure "time-series"
design with equally spaced intervals in between assessments ?

2) My behavior of interest is a binary variable (i.e., presence/absence). I
tried to get the AR1 using the GLMM procedure in SPSS, and the AR1 is not
generated. Is it possible to get AR1 with a binary outcome ? If so, is there
any syntax specification needed ? If not, is there an alternative to AR1
with a binary outcome ?

3) The second goal of my study is to examine the within-person (i.e., Level
1) predictors of behavior stability. Therefore, I was wondering whether it
would be possible to derive a within-person indicator (e.g., AR1) of
behavior stability ? If so, I could use this indicator as an outcome
measure, which could then be predicted by other within-person (i.e., Level
1) variables.

Thanks in advance to all of you for your help.
Oliver



--
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
Reply | Threaded
Open this post in threaded view
|

Re: Temporal stability using multilevel modeling

Jon Peck
I have not seen the original post, but if this question is really about stationarity, then the AR coefficients would be very relevant, and there are necessary and sufficient conditions for this based on them.

On Sun, Jun 24, 2018 at 8:17 PM Marc O Martel <[hidden email]> wrote:

Rich & Andy,


Thank you for sharing your thoughts about this. That's very useful.


Best,

MO.



From: Rich Ulrich <[hidden email]>
Sent: Saturday, June 23, 2018 8:12 PM
To: [hidden email]; Oliver
Subject: Re: Temporal stability using multilevel modeling
 

Like Andy, I'm not speaking to the SPSS questions, either, and I think

AR1 won't be useful. Here are other reasons.  Dichotomous measures

need more cases to match the power achieved by continuous measures.


Six points is an awfully small time series for continuous data for using "time

series methods", I think.  No doubt there are exceptions. You do have 100

parallel series ... which isn't the standard TS problem. My practical experience

with Time series is zero, but, unfortunately for your prospect of getting a better

answer here, we've dealt with very few TS problems, and I suspect that we

have no experts here of that specialty.


The stock markets call their instability "lability" and they measure it from the

standard deviation or variance of stock prices. Variance of a dichotomy is computed

from its mean (i.e., proportion), so I suggest that you get a transparent and obvious

measure of stability/volatility by counting and reporting the number of changes

observed for each ID.   (By the way, if most of your scores for a subject are identical,

at zero or 1, you have a different problem of lack-of-usable-variance for analysis.)


--

Rich Ulrich


From: SPSSX(r) Discussion <[hidden email]> on behalf of Oliver <[hidden email]>
Sent: Saturday, June 23, 2018 9:22:42 AM
To: [hidden email]
Subject: Temporal stability using multilevel modeling
 
Hi everyone,

I'm interested in examining the temporal stability of a behavior that was
measured at 6 different time points (i.e., once a month for 6 consecutive
months). My sample includes 100 patients, and most of them have data on the
6 time points.  I have three questions:

1) For the whole sample, would it be possible to use AR1 as a measure of
temporal stability of the behavior ? Given that AR1 has often been described
as a measure of "temporal dependency", this seems to make sense. It is
important to note, however, that there are minor between-person variations
in time intervals in-between each of the assessment time points, which is
typical in clinical research. Perhaps AR1 requires a pure "time-series"
design with equally spaced intervals in between assessments ?

2) My behavior of interest is a binary variable (i.e., presence/absence). I
tried to get the AR1 using the GLMM procedure in SPSS, and the AR1 is not
generated. Is it possible to get AR1 with a binary outcome ? If so, is there
any syntax specification needed ? If not, is there an alternative to AR1
with a binary outcome ?

3) The second goal of my study is to examine the within-person (i.e., Level
1) predictors of behavior stability. Therefore, I was wondering whether it
would be possible to derive a within-person indicator (e.g., AR1) of
behavior stability ? If so, I could use this indicator as an outcome
measure, which could then be predicted by other within-person (i.e., Level
1) variables.

Thanks in advance to all of you for your help.
Oliver



--
Sent from: http://spssx-discussion.1045642.n5.nabble.com/
This forum is an archive for the mailing list [hidden email] (more options) Messages posted here will be sent to this mailing list.


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


--
Jon K Peck
[hidden email]

===================== 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
Reply | Threaded
Open this post in threaded view
|

Re: Temporal stability using multilevel modeling

Bruce Weaver
Administrator
Perhaps Oliver posted via Nabble without first joining the actual mailing
list?  Oliver, if you have not joined the mailing list, click the "more
options" link at the top of the Nabble archive page.  

Jon, you can see Oliver's original post and the rest of the thread here:

http://spssx-discussion.1045642.n5.nabble.com/Temporal-stability-using-multilevel-modeling-td5736274.html

HTH.


Jon Peck wrote
> I have not seen the original post, but if this question is really about
> stationarity, then the AR coefficients would be very relevant, and there
> are necessary and sufficient conditions for this based on them.
>
> On Sun, Jun 24, 2018 at 8:17 PM Marc O Martel <

> momartel.bwh.harvard@

>> wrote:
>
>> Rich & Andy,
>>
>>
>> Thank you for sharing your thoughts about this. That's very useful.
>>
>>
>> Best,
>>
>> MO.
>>
>> ------------------------------
>> *From:* Rich Ulrich &lt;

> rich-ulrich@

> &gt;
>> *Sent:* Saturday, June 23, 2018 8:12 PM
>> *To:*

> SPSSX-L@.UGA

> ; Oliver
>> *Subject:* Re: Temporal stability using multilevel modeling
>>
>>
>> Like Andy, I'm not speaking to the SPSS questions, either, and I think
>>
>> AR1 won't be useful. Here are other reasons.  Dichotomous measures
>>
>> need more cases to match the power achieved by continuous measures.
>>
>>
>> Six points is an awfully small time series for continuous data for using
>> "time
>>
>> series methods", I think.  No doubt there are exceptions. You do have 100
>>
>> parallel series ... which isn't the standard TS problem. My practical
>> experience
>>
>> with Time series is zero, but, unfortunately for your prospect of getting
>> a better
>>
>> answer here, we've dealt with very few TS problems, and I suspect that we
>>
>> have no experts here of that specialty.
>>
>> The stock markets call their instability "lability" and they measure it
>> from the
>>
>> standard deviation or variance of stock prices. Variance of a dichotomy
>> is
>> computed
>>
>> from its mean (i.e., proportion), so I suggest that you get a transparent
>> and obvious
>>
>> measure of stability/volatility by counting and reporting the number of
>> changes
>>
>> observed for each ID.   (By the way, if most of your scores for a subject
>> are identical,
>>
>> at zero or 1, you have a different problem of lack-of-usable-variance for
>> analysis.)
>>
>>
>> --
>>
>> Rich Ulrich
>> ------------------------------
>> *From:* SPSSX(r) Discussion &lt;

> SPSSX-L@.UGA

> &gt; on behalf of
>> Oliver &lt;

> momartel.bwh.harvard@

> &gt;
>> *Sent:* Saturday, June 23, 2018 9:22:42 AM
>> *To:*

> SPSSX-L@.UGA

>> *Subject:* Temporal stability using multilevel modeling
>>
>> Hi everyone,
>>
>> I'm interested in examining the temporal stability of a behavior that was
>> measured at 6 different time points (i.e., once a month for 6 consecutive
>> months). My sample includes 100 patients, and most of them have data on
>> the
>> 6 time points.  I have three questions:
>>
>> 1) For the whole sample, would it be possible to use AR1 as a measure of
>> temporal stability of the behavior ? Given that AR1 has often been
>> described
>> as a measure of "temporal dependency", this seems to make sense. It is
>> important to note, however, that there are minor between-person
>> variations
>> in time intervals in-between each of the assessment time points, which is
>> typical in clinical research. Perhaps AR1 requires a pure "time-series"
>> design with equally spaced intervals in between assessments ?
>>
>> 2) My behavior of interest is a binary variable (i.e., presence/absence).
>> I
>> tried to get the AR1 using the GLMM procedure in SPSS, and the AR1 is not
>> generated. Is it possible to get AR1 with a binary outcome ? If so, is
>> there
>> any syntax specification needed ? If not, is there an alternative to AR1
>> with a binary outcome ?
>>
>> 3) The second goal of my study is to examine the within-person (i.e.,
>> Level
>> 1) predictors of behavior stability. Therefore, I was wondering whether
>> it
>> would be possible to derive a within-person indicator (e.g., AR1) of
>> behavior stability ? If so, I could use this indicator as an outcome
>> measure, which could then be predicted by other within-person (i.e.,
>> Level
>> 1) variables.
>>
>> Thanks in advance to all of you for your help.
>> Oliver
>>
>>
>>
>> --
>> Sent from: http://spssx-discussion.1045642.n5.nabble.com/
>> SPSSX Discussion | Mailing List Archive
>> &lt;http://spssx-discussion.1045642.n5.nabble.com/&gt;
>> spssx-discussion.1045642.n5.nabble.com
>> This forum is an archive for the mailing list

> spssx-l@.uga

>> (more options) Messages posted here will be sent to this mailing list.
>>
>>
>> =====================
>> 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
>
>
>
> --
> 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.

--
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."

NOTE: My Hotmail account is not monitored regularly.
To send me an e-mail, please use the address shown above.
Reply | Threaded
Open this post in threaded view
|

Re: Temporal stability using multilevel modeling

Rich Ulrich

Bruce, and Jon,

I'm reading from the List, not from Nabble, and I saw Oliver's message on Saturday.

So there seems to be no problem with his distribution to the List.

Was there a problem with the distribution by the List to Nabble?


--

Rich Ulrich


From: SPSSX(r) Discussion <[hidden email]> on behalf of Bruce Weaver <[hidden email]>
Sent: Monday, June 25, 2018 8:28 AM
To: [hidden email]
Subject: Re: Temporal stability using multilevel modeling
 
Perhaps Oliver posted via Nabble without first joining the actual mailing
list?  Oliver, if you have not joined the mailing list, click the "more
options" link at the top of the Nabble archive page. 

Jon, you can see Oliver's original post and the rest of the thread here:

http://spssx-discussion.1045642.n5.nabble.com/Temporal-stability-using-multilevel-modeling-td5736274.html

HTH.


===================== 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
Reply | Threaded
Open this post in threaded view
|

Re: Temporal stability using multilevel modeling

Maguin, Eugene
In reply to this post by Oliver
Ignoring the number of observations per person, I visualize you wanting to run a two level logistic regression model in which the only level 1 predicto, other than the intercept, is the prior observation. And, you'd like to have both a random intercept and a random slope, which I construe to be your AR(1) term, because you want to relate variation in the level 1 slope to person characteristics. In reality, I don't think you can get that equation to solve with only six observations and a random slope. I think you could get an intercept only model to solve but's not quite what you want. So, what if you tabulated the number of reversals in the observation string. Example: (101) has two reversals; (100) as one. The drawback is that (010) and (101) have the same number of reversals so perhaps another variable to use is the initial value.
Perhaps people that have worked with data like this have better suggestions/methods.

Gene Maguin



-----Original Message-----
From: SPSSX(r) Discussion <[hidden email]> On Behalf Of Oliver
Sent: Saturday, June 23, 2018 9:23 AM
To: [hidden email]
Subject: Temporal stability using multilevel modeling

Hi everyone,

I'm interested in examining the temporal stability of a behavior that was measured at 6 different time points (i.e., once a month for 6 consecutive months). My sample includes 100 patients, and most of them have data on the
6 time points.  I have three questions:

1) For the whole sample, would it be possible to use AR1 as a measure of temporal stability of the behavior ? Given that AR1 has often been described as a measure of "temporal dependency", this seems to make sense. It is important to note, however, that there are minor between-person variations in time intervals in-between each of the assessment time points, which is typical in clinical research. Perhaps AR1 requires a pure "time-series"
design with equally spaced intervals in between assessments ?

2) My behavior of interest is a binary variable (i.e., presence/absence). I tried to get the AR1 using the GLMM procedure in SPSS, and the AR1 is not generated. Is it possible to get AR1 with a binary outcome ? If so, is there any syntax specification needed ? If not, is there an alternative to AR1 with a binary outcome ?

3) The second goal of my study is to examine the within-person (i.e., Level
1) predictors of behavior stability. Therefore, I was wondering whether it would be possible to derive a within-person indicator (e.g., AR1) of behavior stability ? If so, I could use this indicator as an outcome measure, which could then be predicted by other within-person (i.e., Level
1) variables.

Thanks in advance to all of you for your help.
Oliver



--
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