Commentary on Two Articles by C. A. Los

Recently, C. A. Los has recommended that all previous methods for estimating linear relations from data with unknown errors be scrapped, and proposes to nd an `exact objective mathematical solution' that depends only on the data. We discuss this with the conclusions that (1) his criticisms of previous methods, while overly severe, make some important points worth noting; (2) however, his solution ignores relevant information and does not exist except in very special cases; (3) the Los solution, when it exists, may be characterized as the one which would be correct if we had an in nite amount of data which led to the same data correlation matrix; (4) methods for dealing with the problem which are optimal in all cases have been known for 20 years.