Signed regression: Principle and user guide

To our knowledge, there is no linear regression method that systematically ensures that the sign of the coefficients agrees with that of the simple correlation between the dependent variable and each regressor and whose value does not conform to the constraint imposed on them. But this twofold imperative – a coefficient that is positive and sufficiently different from zero for each variable – is often a prerequisite for practitioners working, for example, in the field of consumer satisfaction, where statistical collinearity is the norm. The purpose of this article is to provide them with a formal method that addresses this need and produces results comparable to those of statistically based methods, with strict and systematic conformity to the associated sign