Academic journal article Psychological Test and Assessment Modeling

A Primer on Relative Importance Analysis: Illustrations of Its Utility for Psychological Research

Academic journal article Psychological Test and Assessment Modeling

A Primer on Relative Importance Analysis: Illustrations of Its Utility for Psychological Research

Article excerpt

(ProQuest: ... denotes formulae omitted.)

Researchers often have the dual goals of both trying to predict valued criteria and trying to understand the relative importance of the variables used to predict these criteria. The statistical method usually employed for this purpose is multiple regression, and researchers interested in evaluating predictors used in multiple regression analyses mostly rely on straightforward statistical indices, such as standardized regression coefficients, squared correlations, zero-order correlations, and semi-partial correlations (Johnson, 2001; Johnson & LeBreton, 2004). Unfortunately, when multiple predictors are correlated with one another, as is nearly always the case, these simpler measures for evaluating the relative importance of predictors can be problematic because they fail to properly partition variance to the different predictors (Darlington, 1968). Since psychological research often measures constructs consisting of different but correlated facets, a more elaborate research methodology is needed. Recent work on relative importance analyses (Azen & Budescu, 2003, Tonidandel & LeBreton, 2011) offers easily accessible ways of determining the importance of multiple predictors while accounting for the correlations between them.

Our aim in this primer is to demonstrate for psychological researchers the potential that comes with using relative importance analyses in addition to classical regression methods. We do this by using data from two already published papers to illustrate two related methods and their respective benefits. Our criteria were that these needed to be from different fields of research so that our examples were accessible to a broad range of readers, and that they must have multiple, correlated predictors - since this best illustrates the added benefits of relative importance analysis. Hence we re-analysed data from an study conducted by Cooper-Thomas, van Vianen, and Anderson (2004) on the impact of socialization tactics on newcomer adjustment. For our second example, we reanalysed data from an organizational governance investigation conducted by Lui and Ngo (2012) on the drivers and outcomes of long-term orientation in co-operative relationships between organisations.

Throughout this primer, we focussed only on the theoretical aspects of the relative importance approach to the extent that it is useful as a context for highlighting the utility of this approach (for a detailed discussion, see Tonidandel & LeBreton, 2011). We next present an overview on relative importance analyses, their theoretical basis and application, before presenting the two examples in more detail.

Relative importance

Various approaches have been proposed for investigating the relative importance of each of a set of correlated predictors. One of the latest and most useful is relative weights analysis. First we present traditional data analysis approaches of correlations and multiple regression analysis, with associated advantages and shortcomings, before elaborating on the mechanics and additional benefits of relative weights analysis.

Correlation and multiple regression analyses

Correlation analyses compare one pair of variables at a time and provide indicators that reflect the direction (positive or negative) and size (-1 to +1) of that specific linear relationship. While this can be useful, a downside is that it ignores the relations that each variable in the pair has with other variables that may be of interest in predicting the target variable. Multiple regression offers an advantage over correlation analysis, in that multiple predictor variables can be considered in predicting a single criterion variable and yielding a single prediction equation, while considering the predictors' collinearity. An overall R2 is reported which reflects the amount of variance that these predictors jointly explain.

In multiple regression, ideally each predictor variable contributes substantially and independently to the prediction of the variability in the criterion variable (Tabachnick & Fidell, 2001). …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.