<color><param>0100,0100,0100</param>Dear Allstaters
As promised, here are the collated replies to my time series query
(at the bottom of this message).
Many thanks to all those who replied - it's been very helpful.
Francesca
<FontFamily><param>Times New Roman</param>Try cross corelation function.
Also try to consruct the time series without the group mean.<smaller>
<bigger>References
Box, Jenkins and Reinsel "Time series analysis, forecasting and
control",Prentice-Hall
and
Shumway and Stoffer "Time series analysis and its Applications", Springer<smaller>
<bigger>I worked with functional data some time ago calculating medians for f.
data but I think some more work has been done by Ricardo Fraiman and
Graciela Boente who have produced anovas for f. data. You might be
able to use his algorithms to solve your problem.
Good luck and let me know if you have any answers!
Has he/she only one group of patients? SOme sort of repeated
measures would be what you would need, but it depends on what question
you
are asking. The first step would be to produce a set of patient-specific
plots to get an idea of what model might be appropriate for the data.
I would analyse the correlation between them.
This sounds to me more like a 'repeated measures' model than time
series.
The analysis chosen destroys information on 'between patient
variability' which is presumably important.
I am sure there are people who know more about 'repeated measures'
analysis than I do. There are a number of options and it is not clear
which is best in your case.
Your description is slightly ambiguous, in suggesting "four variables" but
then taking a mean.
I looked at something similar, in a protocol that had each patient
"resting" for some minutes, then a minor intervention, then continued
measurements. Although the measurements were at nominal one minute
intervals, there was some unknown variation in the gaps, and the time to
conduct the intervention would also vary. Hence it did not seem prima
facie reasonable to take the average at each nominal time. It is also not
clear at this stage (in either analysis) what the form of difference might
be, so any mechanical process would have dubious power.
In the end, I simply plotted each time series as a line using two colours
(control and experimental), and it was very plain that the two groups
intermingled thoroughly until the intervention time, then one group jumped
to a higher level before subsiding into more or less the mixed range.
This plot then justified a simplistic interpretation of t-values to answer
the central question which was that the experimental intervention had a
real effect in the desired direction.
The plot was easily achieved using Stata: sort on patient and time, then
use connect(L) to join lines.<smaller>
<FontFamily><param>Arial</param><bigger>------- Forwarded message follows -------
</color>From: <color><param>0000,0000,8000</param>Francesca Chappell <<[log in to unmask]></color>
To: <color><param>0000,0000,8000</param>[log in to unmask]</color>
<bold>Subject: <color><param>0000,0000,8000</param>time series query</bold></color>
Date sent: <color><param>0000,0000,8000</param>Wed, 19 Nov 2003 16:20:15 +0000</color>
Dear ALLstaters
A colleague of mine has data from a group of patients. Four
continuous variables were measured twelve times on each patient.
At each time point, the mean of the group was calculated for each
variable. These group means were used to produce four time
series with twelve data points each. He would like to know if the
series are signficantly different. Does anybody know of a test or
reference that could help me?
I will collate replies and send them to the list.
Thank you.
Francesca
<color><param>0100,0100,0100</param>------- End of forwarded message -------
<nofill>
Francesca Chappell
Medical Statistician
University of Edinburgh
Clinical Neurosciences
Western General Hospital
Crewe Road
Edinburgh
EH4 2XU
Tel: 0131 537 2932
|