Deming regression

From Wikipedia, the free encyclopedia

In statistics, Deming regression, named after W. Edwards Deming, is a method of linear regression that finds a line of best fit for a set of related data. It differs from simple linear regression in that it accounts for error in both the x- and the y-axes. The line of regression (or line of best fit) must begin where the x- and y-axes meet (zero).

It is important that the line starts from zero, mainly because one will see the spread of data from zero. It allows one not to skip data, by not starting from zero. For instance, if the x-axis data is known to have no error, but the y data does, (such as a population estimate (y), at a known time (x)), simple linear regression would work.

If both sets of data contained error, for instance the relationship between concentrations of two substances in blood, Deming regression would be more appropriate.

The disadvantage with Deming regression, is that it is mathematically more complex to do. This means doing the calculations, either on paper, or by writing a formula for a spreadsheet, are more difficult. Software such as S-PLUS, Analyse-it, EP Evaluator and MedCalc offer Deming regression.