144 DIMENSIONS 1, 2, 3, . . . , ∞
Besides, there are home-made problems that are generated by choices made in the initial step.
An example of home-made problems can be illustrated for linear models. The interpretation of
the Gauss-Markov estimator as a linear projection shows that seemingly only the coefficients for
single regressors are estimated. In fact, a vector in the vector space spanned by the regressors
is estimated; the attribution to individual regressors is mere linear algebra. The allocation to
individual components does not depend on the influence of the individual regressor, but on
the joint geometry of all the regressors. There is a direct interpretation of the coefficients only
if the regressors form an orthogonal basis. For example if ...