Test-Retest Reliability in Infant Speech Perception Tasks

A long line of research investigates how infants learn the sounds and words in their ambient language over the first year of life, through behavioral tasks involving discrimination and recognition. More recently, individual performance in such tasks has been used to predict later language development. Does this mean that dependent measures in such tasks are reliable and can stably measure speech perception skills over short time spans? Our three laboratories independently tested infants with a given task and retested them within 0-18 days. Together, we can report data from 12 new experiments (total number of paired observations N = 409), ranging from vowel and consonant discrimination to recognition of phrasal units. Results reveal that reliability is extremely variable across experiments. We discuss possible causes and implications of this variability, as well as the main effects revealed by this work. Additionally, we offer suggestions for the field of infant speech perception to improve the reliability of its methodologies through data repositories and crowd sourcing.