Comparing the predictability of behavioral act frequencies from a big-five and a maximal-dimensional item set
Elisa Altgassen, Gabriel Olaru and Oliver Wilhelm
Abstract
Personality inventories are predominantly curated using factor analytic approaches. Indicators capturing common and thus redundant variance are preferentially selected, whereas indicators capturing a large proportion of unique variance outside the broad trait domains are omitted from further research. Even recent research dealing with lower-level personality traits such as facets or nuances has invariably relied on inventories founded on this factor analytic approach. However, items can also be selected to ensure low instead of high communality amongst them. The expected predictive power of such item sets is higher compared to those compiled to capitalize on the indicators’ redundancy. To investigate this, we applied Ant Colony Optimization (ACO) to select personality-descriptive adjectives with minimal inter-item correlations. When used to predict the frequency of everyday life behaviors, this ‘crude-grit’ set outperformed a traditional big-five item set and sets of randomly selected adjectives. The size of the predictive advantage of the crude-grit set was generally higher for those behaviors that could also be predicted better by the big-five item set. This study provides a proof-of-concept for an alternative procedure for compiling personality scales, and serves as a starting point for future studies using broader item sets.
Abstract
Personality inventories are predominantly curated using factor analytic approaches. Indicators capturing common and thus redundant variance are preferentially selected, whereas indicators capturing a large proportion of unique variance outside the broad trait domains are omitted from further research. Even recent research dealing with lower-level personality traits such as facets or nuances has invariably relied on inventories founded on this factor analytic approach. However, items can also be selected to ensure low instead of high communality amongst them. The expected predictive power of such item sets is higher compared to those compiled to capitalize on the indicators’ redundancy. To investigate this, we applied Ant Colony Optimization (ACO) to select personality-descriptive adjectives with minimal inter-item correlations. When used to predict the frequency of everyday life behaviors, this ‘crude-grit’ set outperformed a traditional big-five item set and sets of randomly selected adjectives. The size of the predictive advantage of the crude-grit set was generally higher for those behaviors that could also be predicted better by the big-five item set. This study provides a proof-of-concept for an alternative procedure for compiling personality scales, and serves as a starting point for future studies using broader item sets.
-> "We selected the crude-grit set
with the goal to not factorize, cluster, or reduce it in
dimensionality. Therefore, this set will not serve the
purpose of describing personality in the condensed
manner the big-five sets do (Mõttus et al., 2020). Rather,
instead of factors, individual responses are conceptualized as an inventory of trait-like person attributes. In
the end, our findings emphasize the trade-off between
capturing the entire personality trait space or only the
most salient trait domains. While a questionnaire based
on a few broad trait domains will provide a tool for easier
interpretation and communication of personalityoutcome associations, we showed that it may be limited in finding these associations to begin with. In
contrast, a broader personality space coverage can improve the prediction of relevant outcomes, but does so at
the cost of interpretability of the associations."
Keine Kommentare:
Kommentar veröffentlichen