Using Complexity Metrics to Assess Silent Reading Fluency: A Cross-Sectional Study Comparing Oral and Silent Reading

Abstract

Reading typically undergoes a qualitative shift around Grade 4, becoming more fluent and silent, but there is no established measure for fluency in children’s silent reading. The present study presents a measure of self-paced reading in children, examining the use of complexity measures for time-series analyses recently established with adults. Cross-sectional groups of adults and children in Grades 2, 4, and 6 read the same passage of a Grade 2-level story, either silently or aloud. The dynamical structure of fluency in reading times was explored with fractal and recurrence quantification analysis. Results revealed that more fluent reading (with increasing age) was marked by greater structure and stability and that oral reading compared with silent reading showed less fractal structure, indicating silent reading as a more flexibly stable, adaptive coordinated behavior. The complexity metrics show promise for an alternate way to characterize reading fluency.

ACKNOWLEDGMENTS

We thank the students who participated in the research and the school principals and teachers who assisted in this work; Brenda Miller, Jacquelyn Rowedder, Kevin, Hardy, Melanie Hogue, Rachel Pappenfus, Herman Daniels, and anonomous reviewers for their input. Deep gratitude is also felt for the vision provided by our late colleague, Guy Van Orden, who remains an inspiration.

Notes

¹All data sets were entered into the analysis with parameters of delay = 1, dimension = 5, and Euclidean norm.

²The interevent time series of word-reading times was also converted into a continuous time series of words-per-minute (WPM), analogous to BPM calculated from RR intervals in heart rate. We used an “oversampled” WPM-series of maximal sensitivity of the analysis with a 200-ms window (Wallot, Fusaroli, Tylén, & Jegindø, 2013). DFA was performed on this data as before, and the effect of oral versus silent reading remained stable, F(1, 66) = 17.77, p < .001. There were no other significant effects.