How to investigate the association between two series of measurements

[Frank Furter]

I have series of (metric) lab variables measured in a group of about 300 patients, with once daily measurements over 20 days, and I would like to investigate the association between the time courses of two of these variables. What is the preferred statistical procedure for analyzing such associations?

Andreas

[Ryan]

Andreas,

Are you asking how one might test whether there is a significant difference between variable-specific [linear] slopes over time?

Ryan

On May 23, 2013, at 2:38 AM, Andreas Voelp <[hidden email]> wrote:

> I have series of (metric) lab variables measured in a group of about 300
> patients, with once daily measurements over 20 days, and I would like to
> investigate the association between the time courses of two of these
> variables. What is the preferred statistical procedure for analyzing such
> associations?
>
> Andreas
>
>
>
> --
> View this message in context: http://spssx-discussion.1045642.n5.nabble.com/How-to-investigate-the-association-between-two-series-of-measurements-tp5720365.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

[Frank Furter]

Ryan,

I am actually more interested in obtaining some measure of association or covariation (does 'y' tend to increase when 'x' increases and vice versa).

Andreas

[Jo Fennessey]

I will be away from the office until Wednesday 5/29/13, but will check email daily.

[Ryan]

Andreas,

 

You could employ a multivariate linear growth curve model via the MIXED procedure to estimate the covariance (or correlation) between the two slopes. It is worth noting that this model is also referred to as a latent growth curve model.

 

To do so, your dataset should be structured as follows:

 

ID  y_indic   time   y
1       1           0     23
1       2           0     14
1       1           1     24
1       2           1     17
1
1
.
.
1       1          19    52
1       2          19    33
2       1           0     21
2       2           0     17
.
.

Here is the code:

 

 mixed y by y_indic with time
  /fixed= y_indic y_indic*time | noint sstype(3)
  /method=reml
  /print= g solution testcov 
  /random=y_indic y_indic*time | subject(ID) covtype(un).

 

Upon examination of the covariance parameters table, you will find the covariance between the slopes. You could convert the covariance to a correlation by changing covariance structure from UN to UNR.

 

This model assumes a linear change in x and y.

 

Ryan

On Thu, May 23, 2013 at 7:29 AM, Andreas Voelp <[hidden email]> wrote:

Ryan,

I am actually more interested in obtaining some measure of association or
covariation (does 'y' tend to increase when 'x' increases and vice versa).

Andreas



--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/How-to-investigate-the-association-between-two-series-of-measurements-tp5720365p5720369.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

[Frank Furter]

... thanks very much :)