The Method of Least Squares and Some Alternatives: Part IV

Conclusions and Recommendations 1. The best choices of measures of central tendency and dispersion and of methods of fitting linear (or non-linear) regression equations depend upon the error law, i.e. the distribution of L1 Double exponential 1 -%l x-1I/() f(x) = _e(Co, + CO) (Laplace's first) L2 Normal-Gaussian 1 _(X)2/2a2 (Laplace's second) f(x) = 2-, 1U UN Uniform-rectangular f(x) = _(--3o,+ 3) 2a 3