Establishing determinant relevance using CIBER: an introduction and tutorial

When developing behavior change interventions, it is important to target the most important determinants of behavior (i.e. psychological constructs that predict behavior). This is challenging for two reasons. First, determinant selection requires integrating multiple information sources: determinants' associations with either behavior or with determinant that mediate their effect on behavior (i.e. effect sizes), as well as how much room for improvement there is in the population (i.e. means and spread). Second, only information from samples is normally available, and point estimates obtained from samples vary from sample to sample, and therefore cannot be interpreted without information about how much they can be expected to vary over samples. In practice, determinant studies often present multivariate regression analyses, but this is problematic because by default, shared covariance is removed from the equation (literally), compromising operationalisations' validity and affecting effect sizes (i.e., the results of such analyses cannot be used as a first source of information regarding each determinant's association to behavior). In the present contribution, we will briefly explain these points in more detail, after which we will introduce a solution: confidence interval based estimation of relevance (CIBER). We will then present a brief tutorial as to how to generate CIBER plots and how to interpret them. This is a more detailed explanation and introduction: originally, CIBER was published in Crutzen, Peters & Noijen (2017).