Does performance on a visual statistical learning test correlate with L2 spelling? Sample size matters, a lot

Abstract

Purpose: Previous studies reported a significant and positive correlation between performance on a visual statistical learning (VSL) task and reading ability in L1 and L2. In light of these, Mak (2016) investigated if there was any correlation between a person’s ability to detect implicit sequences in a VSL task and their performance on an L2 spelling task. Surprisingly, a negative correlation (r=-0.55, p=.04) was reported; that is, better L2 spellers tended to perform more poorly on the VSL task. This was attributed to a potential trade-off relation between implicit and explicit learning. Built upon this finding, the current study replicated Mak (2016) with a much larger sample and sought to explore if a trade-off exists between implicit VSL and explicit learning. Method: Sixty-four native Cantonese-speaking undergraduates, who also speak English proficiently, participated in the study. They completed a spelling task in which they spelt out highly rare English words, a VSL task (triplet learning), an explicit vocabulary learning task (MLAT-V), and a vocabulary size test. Results: While L2 spelling performance showed a significant positive correlation with L2 vocabulary size, L2 spelling did not correlate with any other measures, suggesting that the finding of Mak (2016) might have been a false positive due to small sample size (N=21). VSL task also showed no relation with the explicit learning task. Conclusions: We believe statistical learning is linked to L2 spelling proficiency, but traditional VSL measures are unable to show such a relation. Our studies also demonstrated that false positives may arise due to small sample size and that the reading research community, especially those who are interested in individual differences, should work towards reproducibility and better science by avoiding unjustifiably small sample sizes.