All together now: Random Forests analysis reveals the joint impact of multiple statistical regularities on...

A large and growing number of recent studies has embraced a statistical learning view of reading, revealing that readers utilize an array of regularities that are available in writing systems as they process printed words and texts. However, previous studies have focused on the impact of one regularity (or an otherwise small number of cues). Therefore, we currently have a limited understanding of (1) whether different regularities each carry unique explanatory power, beyond other (collinear) cues; (2) how do regularities at different levels of the input contribute to reading behavior; and (3) whether regularities vary in their contributions across processing stages. To answer these questions, we employ Random Forests analyses on a large-scale, eye-movement, passage-reading database from English first- and second-language readers, evaluating the relative importance of a large number of regularities on multiple eye-movement dependent variables. First, our findings demonstrate that, each regularity uniquely contributes to the model's performance. Second, we show that both text-level regularities (e.g., predictability) and word-level regularities (including print-speech and print-meaning regularities), contribute to continuous text reading. Third, we document varying contributions of some regularities over time, with later reading measures being more impacted by text-level regularities. These results support and extend statistical learning theories of reading, showing that readers are attuned to a range of regularities in their writing system, which jointly guide naturalistic reading behavior.