Age of acquisition predicts naming and lexical-decision performance above and beyond 22 other predictor variables: An analysis of 2,342 words

Abstract

Age of acquisition (AoA) ratings were obtained and were used in hierarchical regression analyses to predict naming and lexical-decision performance for 2,342 words (from Balota, Cortese, Sergent-Marshall, Spieler, & Yap, 2004). In the analyses, AoA was included in addition to the set of predictors used by Balota et al. (2004). AoA significantly predicted latency performance on both tasks above and beyond the standard predictor set. However, AoA was more strongly related to lexical-decision performance than to naming performance. Finally, the previously reported effect of imageability on naming latencies by Balota et al. was not significant with AoA included as a factor. These results are consistent with the idea either that AoA has a semantic/lexical locus or that AoA effects emerge primarily in situations in which the input–output mapping is arbitrary.

Notes

¹ We thank Andy Ellis for pointing out this line of reasoning.

² Based on a recent paper by Baayen, Feldman, and Schreuder (2006), log frequency's effect on processing speed seems to level off at the very high end of the frequency scale. We addressed this issue by replacing our log frequency value, log (frequency + 1), with [log (frequency + 1)]². This modification did not change the results reported here in any meaningful way. Specifically, in all cases, R ² was either the same or higher with the original measure of log frequency. More importantly, this change did not change the predictive power of AoA at all, and only in one case did the predictive power of another factor change. That is, in the re-analysis of naming accuracy, subjective frequency was now a significant predictor (β = .073, p < .05). Also, we note that frequency trajectory correlates with AoA, r  =  –.50, p < .01. Therefore, it is possible that the variance attributable to AoA in the first set of analyses would be eliminated once frequency trajectory was added to the model. We conducted another set of analyses in which the objective frequency measure employed by Balota et al. (2004) was replaced by the cumulative frequency measure and frequency trajectory measure defined by Zevin and Seidenberg (2004). There were 2,046 words in our data set for which there were values available for this new set of predictors. In this analysis, AoA remained a significant predictor of performance in both of the latency analyses, but the inclusion of frequency trajectory suggested the presence of a suppressor variable. Specifically, while the signs of beta did not change for AoA whether frequency trajectory was included or not, the signs of beta for frequency trajectory in both of the lexical-decision analyses change depending on the presence or absence of AoA in the equation. Furthermore, the beta signs corresponding to frequency trajectory when AoA is included in the equation make little theoretical sense. When AoA is included in the equation, frequency trajectory is positively related to response latencies (i.e., words that have higher frequencies early in development are associated with slower response latencies than are words that have lower frequencies early in development) and negatively related to decision accuracy (i.e., words that have higher frequencies early in development are associated with lower accuracies than are words that have lower frequencies early in development).