I completely agree with you. I forgot to care about resolution,
and I was thinking only on the problem of determining a representative
*value* of a magnitude. In this sense, something like 3.45+-0.03 I
interpret it as the confidence interval where I'd expect to find
the *real* value with a given probality (usually araound 68% if
the data can be assumed to have a gaussian distribution). And for
this to be so, 0.03 must be the standard error of the mean.
Regards,
Miguel A. Santos
msantos2etse.urv.es
On Tue, 10 Oct 2000, Michael Edmiston wrote:
So what are you trying to tell the reader? Are you trying to demonstrate
that your new instrument/method has good resolution? If so, report the
standard deviation. On the other hand, if you are trying to indicate how
well you think you have pinned down the mean, you should report the
standard error of the mean.