Orthogonal regression

This topic has come up a few times recently, so it must be 'in the wind'

these days. Depending on what approach you take and what area you are
coming 
from, it goes under the names Orthogonal [Distance] Regression, Total
Least 
Squares, Errors in Variables, Deming Regression. The middle two
encompass 
much more in statistics terms (and I don't know what Deming did for
sure.)

If you have just two variables and you assume they are observed without 
error, then the regression slope is just the ratio of the standard 
deviations with the sign of the correlation coefficient, and the line
passes 
through the means. There are assumptions about the errors that will
produce 
the same result as a 'best linear fit,' but these assumptions about the 
errors may not fit your situation. Do some reading in the TLS / EIV
arena to 
see if you really want bald ODR.

HTH,
David L. Reiner