Hi Pier,

> What is the additional uncertainty introduced by your correction
> procedure
> as described under 3)? For certain such a procedure makes the
> measurement
> less precise...

I don't think this will introduce an additional error.
Of course one have to make sure not to change the raw values
arbitrarily in a non-traceable way by adding random or standard
values. This will indeed introduce a larger error.

The idea behind (3) is to bring the first (two?) value(s) "in line"
with the constant value measured after the memory effect is gone. The
memory effect is gone in our system usually after one or two
injections, but this changes from day to day and system set up. So
you can not use a
standard value for the correction determined once. Instead you have
to do this in every run, every day. The value(s) used in the memory
correction is used to adjust the whole sequence. You must not use a
new value for every "jump", of course!

The raw data of the third value (which is not any more affected by
the memory) is virtually not changed by this procedure. Samples with
precursors similar in the isotopic composition (+/- 20%o in Deuterium)
will also not be affected, since the correction approaches zero. So it
is a good idea to arrange the samples in ascending/descending order
and use a drift standard with a similar composition to minimize this
effect. You may also refer to [1], where this is described in a bit 
more detail.

Post run correction procedures, if carefully applied, correct the
raw data for real effects during the measurement and does not
manipulate the data or make it "less precise". The opposite is true.
It gives you better idea about the real value of your sample (from a
statistical point of view) since your mean value relies on two or more
measured values instead of ony one value. It would be even better, of
course, to analyze the sample in more than one sequence.   

Also, you must monitor your long term external reproducibility by
measuring the same standard in every sequence with the post run
corrections applied to these values, too. In case the procedure
will introduce a larger error to values as compared to the
raw/uncorrected data this will become obvious and you have to
revise your procedure (or skip it and use method 1 or 2).

Long text. I definitely didn't want to start a discussion about post
run corrections...


Regards,
Robert


[1] Roland A. Werner and Willi A. Brand, Referencing Strategies and
Techniques in Stable Isotope Ratio Analysis, Rapid Comm. Mass
Spectrom. 15 (2001) 501-519 

-- 
Dr. Robert van Geldern
Section 3: Geochronology and Isotope Hydrology
Leibniz Institute for Applied Geosciences (GGA-Institute)

GEOZENTRUM HANNOVER
Stilleweg 2
30655 Hannover, Germany

Phone: +49-(0)511-643-2313
Fax: +49-(0)511-643-3665
mailto:[log in to unmask]