Individual differences in infants' information processing: reliability, stability, and prediction.
A group of 46 full-term and 54 high-risk preterm (less than 1,500 grams birthweight) infants were tested at 6, 7, and/or 8 months of age (corrected age for preterms) on a battery of problems assessing visual recognition memory and tactual-visual cross-modal transfer. At all 3 ages, scores obtained on aggregates of 6-11 problems in the battery significantly predicted 3-year Stanford-Binet IQ: correlations ranged from r = .37 to r = .63, and clustered between r = .50 and r = .60. When aggregates from 2 or 3 ages were used as predictors, multiple correlations were as high as R = .60 and R = .70. Cutoffs for predicting children at risk for mental retardation (IQ less than 70) or cognitive delay (IQ less than 85) showed reasonable sensitivity and specificity, although low scores were poor at detecting IQs less than 70. The internal consistency of composites, indexed by alpha coefficients, was unexpectedly low, primarily because the problems shared little variance. However, stability coefficients between assessments as much as 1 and 2 months apart were moderate in magnitude, ranging from r = .30 to r = .50. Considering the high degree of predictive validity, the stability figures appear to be better estimates of reliability for these measures than are indices of internal consistency. The relations reported here were similar for both full-terms and preterms.