2.4. Behavioral Data Analysis

For participants in the Divided Attention condition, performance on the 3-back task was quantified using d′, with hit rate quantified as the proportion of 3-back matches that were classified correctly and false alarm rate quantified as the proportion of non-matches that were incorrectly classified as hits. This measure was used rather than overall accuracy as there were more non-match trials than match trials, and thus simply responding “non-match” to all trials would lead to a performance of 75% accuracy.

On the familiarity-rating task, ratings were analyzed using a repeated-measures ANOVA with word category (word, part-word, non-word) as a within-participants factor and attention condition (full, divided) as a between-participants factor. For correlational analyses, performance was also quantified by subtracting the average rating to foil items (both part-words and non-words) from the average rating to words, for each participant. Perfect sensitivity on this “familiarity rating score” would be a score of 3, with values above 0 providing evidence of learning. As an additional measure of performance on this task, RTs were analyzed using a second repeated-measures ANOVA with the same factors as above. Median RTs were computed within each word category and participant to reduce the influence of outliers.

For the target-detection task, responses that occurred within 1200 ms after a target were considered to be hits, whereas responses that occurred anytime other than within 0 – 1200 ms of a target were considered to be false alarms; this is the same criterion used in all our past studies (Batterink et al., 2015; Batterink & Paller, 2015, 2017). Mean RTs to detected targets (hits) were calculated for each syllable condition (word-initial, word-medial, and word-final) for each participant; mean rather than median was used as a measure of central tendency in this analysis given that RTs longer than 1200 ms were already excluded according to our “hit” criterion. RTs were analyzed using a repeated-measures ANOVA with syllable position (initial, medial, final) as a within-participants factor, and attention condition (full, divided) as a between-participants factor. Planned contrasts were used to examine whether RTs decreased linearly as a function of syllable position. Performance was further quantified through an “RT prediction effect,” computed as the proportion of RT decrease to third position targets relative to initial position targets [(RT₁ − RT₃)/RT₁). Because decreases in RTs are not independent of the overall speed of response (cf. Siegelman et al., 2017), this computation adjusts for potential differences in baseline RTs between individuals, allowing us to compare statistical learning across individuals with different RT baselines. Larger positive values on the RT prediction effect indicate greater facilitation.