jjr> I don't understand Bruce's remark. Incorporating prior knowledge jjr> into statistics is completely proper according to modern jjr> statistical theory. If two quantities are correlated and one jjr> knows one, then naturally the other is better determined. Well, leaving aside the question of whether one can call a Levenberg- Marquardt fit "modern", I agree that one can set a value and call it part of the model. One does that all the time in any fitting problem. But in that case you are asserting a value with no error bar. There is no situation where you can assert a value AND an error bar. My concern was your statement that you somehow knew the value of the error bar on S02. If it floats, you have to use the error bar the fitting algorithm reports. But the most important thing in my comment and the most important thing in the discussion on the topic of amplitude terms in exafs that we have seen in the last week on this list is that, in a difficult fitting problem (i.e. not copper), one rarely has reliable a priori knowledge of the *product* of S02, coordination, and the systematic uncertainties. Thus one rarely has a reliable prior in the Baysian sense. B -- Bruce Ravel ----------------------------------- ravel@phys.washington.edu Code 6134, Building 3, Room 222 Naval Research Laboratory phone: (1) 202 767 5947 Washington DC 20375, USA fax: (1) 202 767 1697 NRL Synchrotron Radiation Consortium (NRL-SRC) Beamlines X11a, X11b, X23b, X24c, U4b National Synchrotron Light Source Brookhaven National Laboratory, Upton, NY 11973 My homepage: http://feff.phys.washington.edu/~ravel EXAFS software: http://feff.phys.washington.edu/~ravel/software/exafs/