No evidence for visual statistical learning in standard reaction time measures

Himberger, K. 1 , Finn, A. 2 & Honey, C. 1

1 Johns Hopkins University, Department of Psychological & Brain Sciences
2 University of Toronto, Department of Psychology

Standard visual statistical learning (VSL) paradigms in the temporal domain entail exposing participants to regularities and then assessing learning with direct measures like forced-choice recognition tests (Siegelman et al., 2017). However, because direct measures may be subject to influence from explicit knowledge (Bertels et al., 2015), it has been proposed that genuinely implicit aspects of VSL should be assessed indirectly (Turk-Browne et al., 2005).
Across several experiments, we assess learning indirectly by calculating response time differences between perfectly predictable stimuli and pseudo-random stimuli. We found that predictable stimuli were not detected faster than non-predictable stimuli. At the same time, we identified a key measurement confound in standard reaction time paradigms which had shown speeding for predictable items. This confound is sufficient to account for previously reported reaction time facilitation effects, and may be necessary for obtaining such effects. While finding no speeding for predictable items, we did find reliable evidence of VSL using two direct measures (forced-choice and recall). These results undermine the evidence that VSL can occur without any explicit knowledge. Additionally, we suggest that prior VSL work using reaction time measures should be re-interpreted.