Would you prefer your coefficients with a little bias, or rather with a lot of variance?

In this editorial we have reminded ourselves and hopefully also our readers how psychological assessment often has a very practical aim: predicting human behavior in the real world. We aimed to illustrate how bias in regression estimates – which may theoretically sound undesirable – may be practically beneficial. When the aim is to predict human behavior, we may prefer an algorithm that aims to optimize predictive accuracy and variable selection instead of traditional unbiased methods, which aim to recover the “true” values of the parameters of a stochastic model that generated the data. A simulation experiment partly inspired by a study from Aluja and Blanch (2004) is presented, along with the results. (PsycINFO Database Record (c) 2018 APA, all rights reserved)
Source: European Journal of Psychological Assessment - Category: Psychiatry & Psychology Source Type: research