On 4 Mar 2004 08:28:34 -0800, [EMAIL PROTECTED] (Euh) wrote:

> Hi all,
> 
> I'm following the time course of a certain variable by using three
> independent measurement techniques (for which I can estimate the
> variance associated with the measurment)
> 
> I can't really assume any function to describ the evolution of the
> variable over time (linear, quadratic, etc). Is there a way to compare
> the results and assess if all the methods give similar results ?
> 
> ANOVA analysis at each time ?
> 
> An example would be:
> 
>       Method 1        Method 2         Method 3       
>       mean    stdev   mean    stdev   mean    stdev
> t1    24.6    6.4     31.0    13.8    15.5    5.7
> t2    123.5   87.6    155.2   71.5    62.3    20.0
> t3    174.0   33.8    142.7   46.4    75.8    18.9
> t4    210.6   91.2    113.6   26.8    101.9   45.1
> t5    396.4   25.4    263.9   16.5    209.3   11.7
> t6    303.7   66.7    271.9   156.6   216.9   172.1
> t7    153.6   93.4    261.0   225.6   76.0    41.7
> t8    250.9   140.7   289.5   122.7   93.7    28.1

Preliminary points:  

There is an enormous range in means
with an apparent zero-minimum, and a huge increase in
variance for larger numbers.  This strongly suggests the
matter of a reliability model would be helped, most likely,
by taking the log of each of the scores.

The huge variation of variances also implies (I hope) that
those variances are based on 2 or 3 observations at most;
the numbers are not individually precise.

Here are evolutions over time:  Is there any component
of time-series correlation for the multiple measurements
of each method?  Is there a *strong*  component?

Did someone say these are animal experiments?
- that's what comes to my mind, and it gives me something
concrete to discuss -

You should try to use the within-animal correlation, if
there is one, to improve the estimates, or to improve the
knowledge of the *variance*  of the estimates.

I mean, there are at least two measures of Method-One at
t4, t5, and t6, and they show an increase, then decrease, in 
means. Is this describing a biological reality of a *trend*, or 
is this 'sampling variability' between different specimens?  
 - *antibody* levels may be tapped week after week, and there
may be strong correlation within-animal.  However, if measures
require 'sacrificing' an animal in order to measure some organ, 
the weekly scores do not have this component.  Do the scores
reflect 8weeks times 3 times 3 = 72 animals, or do they reflect
just 3 times 3 animals, each measured 8 times?

By the way, *are*  those similar results, according to clinical 
judgment?

-- 
Rich Ulrich, [EMAIL PROTECTED]
http://www.pitt.edu/~wpilib/index.html
 - I need a new job, after March 31.  Openings? -
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to