Researchers seeking to make valid conclusions about engagement for young children with autism spectrum disorder (ASD) must first determine the reliability of estimates obtained across the conditions sampled. Working from that premise, we conducted a secondary data analysis of shared book readings between caregivers and their children with ASD, examining the contribution of measurement error on estimates of four states of child engagement. Caregivers read two different book types, each three times, with their children. With book type and occasion as measurement facets, we conducted a generalizability study and a series of decision studies. With the interaction of "Persons × Book Type × Occasion" contributing the most measurement error for four engagement variables, we only found stable estimates for unengaged behaviors. For the active engagement, visual engagement, and disruptive variables, 4, 5, and more than 10 book types, respectively, were necessary to obtain stable estimates across two occasions.