<p>Reliability of IDELA and ECDI 2030.</p>

Many directly assessed and caregiver-reported measures of early childhood development (ECD) are used to monitor outcomes and evaluate interventions in low- and middle-income countries (LMICs). Many assessments of ECD for children 0–3 years old have robust psychometric properties, but relatively little is known about their ability to predict later developmental and learning outcomes. To address questions about predictive and concurrent validity, we used longitudinal data on three widely used measures of ECD from 487 children in Tanzania. This study examined the predictive validity of the Caregiver-Reported Early Development Instrument (CREDI) at 12 months of age with the directly assessed International Early Learning and Development Assessment (IDELA) and the caregiver-reported Early Childhood Development Index 2030 (ECDI 2030) administered at 3.5-6 years of age. We also examined the concurrent validity of the IDELA and ECDI2030 administered on the same day. Consistent with prior literature examining the predictive validity of ECD assessments in very young children, we find statistically significant but practically small (~.1 to.2) correlations between age-adjusted CREDI scores measured in infancy and both IDELA and ECDI2030 scores measured at preschool age. Contrary to expectations, CREDI scores were more predictive for children who took IDELA when older and for children whose mothers did not complete primary education. IDELA overall and domain scores had statistically significant medium-sized (~.4 to.5) correlations with ECDI2030 scores that were slightly larger than relations found in prior research. Our results demonstrate limited predictive power from a caregiver-reported assessment administered in infancy, highlighting how variable ECD is for young children. At preschool age, a directly assessed measure of ECD had better reliability than a shorter caregiver-reported measure, but scores on both measures correlated strongly, suggesting that caregiver-reported measures are capturing similar constructs to directly assessed measures of ECD when implemented at the same point in time.