Two Aspects of Bias in Multivariate Studies: Mixing Specific with General Concepts and “Comparing Apples and Oranges”

Abstract

This paper presents two types of bias that occur relatively often when using multivariate analysis. For both types of bias, it is characteristic that the number and choice of different types of variables are not balanced by application of clear methodological rules. Following the interpretation of broader theoretical positions, which include "confirmation bias" ( of initial hypothesis) and "mis¬specification bias", a description of two types of bias characteristic of multivariate analysis are given: "mixed-level bias" (in terms of specificity - generality) and "mixed-constructs bias" . Both types of bias further enhance the disparity in the number and ratio of different types of variables in the same multivariate analysis. Details of situations, when these two types of bias appear, are presented and displayed in four different examples. Several strategies are proposed as to how these types of bias can try to be avoided, during the preparation of studies, during the statistical analyses and their interpretation.

Keywords

• Mixed-constructs bias
• Mixed-level bias
• Multivariate analysis