Following the Ants

Abstract

Abstract. Employees’ innovative work is a facet of proactive work behavior that is of increasing interest to industrial and organizational psychologists. As proactive personality and supervisor support are key predictors of innovative work behavior, reliable, and valid employee ratings of these two constructs are crucial for organizations’ planning of personnel development measures. However, the time for assessments is often limited. The present study therefore aimed at constructing reliable short scales of two measures of proactive personality and supervisor support. For this purpose, we compared an innovative approach of item selection, namely Ant Colony Optimization (ACO; Leite, Huang, & Marcoulides, 2008 ) and classical item selection procedures. For proactive personality, the two item selection approaches provided similar results. Both five-item short forms showed a satisfactory reliability and a small, however negligible loss of criterion validity. For a two-dimensional supervisor support scale, ACO found a reliable and valid short form. Psychometric properties of the short version were in accordance with those of the parent form. A manual supervisor support short form revealed a rather poor model fit and a serious loss of validity. We discuss benefits and shortcomings of ACO compared to classical item selection approaches and recommendations for the application of ACO.

A confirmatory examination of age‐associated personality differences: Deriving age‐related measurement‐invariant solutions using ant colony optimization

“…The item selection procedure used in this study was not applied to present an ideal subset of items but to solve the specific problem of questionable unidimensionality of the traits as well as ageassociated measurement invariance. In other contexts, a different set of selection criteria might be more appropriate (for an overview of possible selection criteria, see Janssen et al, 2017;Leite et al, 2008;Schroeders et al, 2016aSchroeders et al, , 2016b.…”

“…This item selection procedure can be applied to improve both the absolute model fit and the relative model fit between different levels of measurement invariance (Schroeders, Wilhelm, & Olaru, 2016a). The usefulness of item selection for improving model fit has been demonstrated in a number of recent articles on short-scale construction (Janssen, Schultze, & Gr€ otsch, 2017;Olaru, Witth€ oft, & Wilhelm, 2015;Schroeders, Wilhelm, & Olaru, 2016a, 2016b. Finding a measurement model that is unidimensional and measurement invariant across age can be seen as a combinatorial problem in which one aims to identify the best item set out of a multitude of possible item sets.…”

“…We thus applied a meta-heuristic algorithm called ant colony optimization (Leite, Huang, & Marcoulides, 2008;Marcoulides & Drezner, 2003), which is capable of tackling this issue by finding an optimal (or close-to-optimal) solution in an iterative process. ACO algorithms were originally based on the foraging behavior of ants (Deneubourg, Aron, Goss, & Pasteels, 1990) and have been applied to derive optimized short scales (Janssen et al, 2017;Leite et al, 2008;Olaru et al, 2015). Similar to the way in which ants use pheromones to attract other ants to the shortest route to a food source, ACO uses virtual pheromones to increase the attractiveness of items that yield better psychometric properties (e.g., model fit).…”

“…Third, in large outcome spaces, the probabilistic approach of ACO is computationally much less demanding than exhaustive algorithms. And finally, ACO can be used to maximize several criteria simultaneously, such as reliability and model fit (see also Janssen et al, 2017;Schroeders et al, 2016aSchroeders et al, , 2016b).…”

“…The optimization function included the CFI and RMSEA (overall fit) as well as the DCFI between the two decisive steps of invariance testing (i.e., Model 3 and Model 2 in Table 1). All three criteria were logit transformed in order to scale the range of the values between 0 and 1 and maximize the differentiation around the critical cut-off values (see also Janssen et al, 2017;Schroeders et al, 2016aSchroeders et al, , 2016b. The logit transformation ensures that the different optimization criteria are scaled on a common metric and thus weighted equally in the overall optimization function.…”