ABSTRACT
The Dual Pathway to Creativity Model theorizes that creative performance arises from cognitive flexibility, persistence, or a combination of the two. Existing research has only examined these pathways independently, limiting our understanding of the model and of how people generate ideas. We extend the DPCM model by predicting that creative output arises from the generation of response sets containing both highly related (persistence pathway) and unrelated (flexibility) ideas, an idea generation strategy we call Dual Pathway Divergent Thinking (DPDT). In a study of 147 subjects, we propose a serial mediation model: people higher in Openness to Ideas use DPDT as an idea generation strategy which leads to higher nascent creativity which, in turn, leads to higher final creativity. The results provide initial evidence that the two pathways may operate simultaneously as a single idea generation strategy. Theoretical and practical implications are discussed.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Notes
1. Fritz and MacKinnon (Citation2007) provided minimum sample size estimates for achieving .80 power in mediation analyses. A sample size of 77 is sufficient when using bias-corrected bootstrap methods and assuming medium population effect sizes. Prior research suggests medium effects are reasonable estimates for the paths in our models (Berg, Citation2014; De Dreu et al., Citation2008; Fleischhauer et al., Citation2010; McCrae, Citation1987; Medeiros, Steele, Watts, & Mumford, Citation2018; Mussel, Citation2010; Nijstad et al., Citation2010; Vincent et al., Citation2002). However, a sample size of 148 is listed as sufficient if the population effect sizes are assumed to be slightly smaller. Moreover, Bayesian methods have been found to increase power even above bootstrapping methods and especially in samples lower than 200 (Miočević, MacKinnon, & Levy, Citation2017). We repeated our analyses using a Bayesian estimator in Mplus and found that results were essentially identical.
2. Because our reliability estimate fell just below the recommended minimum of .70 we examined the impact of this measure’s reliability on our findings. Alpha can underestimate reliability when a scale has few items or when items are heterogenous (Tavakol & Dennick, Citation2011). We identified one item likely to cause heterogeneity, as it had a notably lower item-total correlation (r = .24 vs. mean r = .48 across all other items). Removing this item, the only reverse-coded item, increased alpha to .71. Results computed with a 4-item Openness to Ideas measure (by excluding this item) were essentially identical to results with the original 5-item measure. We report results using the original 5-item measure validated by Soto and John (Citation2009).
3. This produces the same estimate as computing the analyses with standardized variables, allowing for the estimation of bias-corrected bootstrapped confidence intervals.
4. Average semantic distance was not included as a control in any analyses per Becker et al.’s (Citation2016) impotence rule. Analyses repeated with average semantic distance included were essentially identical.