Longitudinal Rasch Analysis of Imitation in Infants at Risk for Autism

Background: Imitation is significantly impaired in children with autism compared to other groups, and may be an early central feature of the disorder.  However, few studies have examined the development of imitation prior to formal diagnosis, and studies are often hampered by the lack of a standardized developmental measure of imitation.

Objectives: We aimed to test whether individual differences in imitation ability during the second year of life would predict ASD diagnoses at 3 years of age.  A second objective was to use Rasch analysis to evaluate the psychometrics of a standardized imitation battery, and to explore the Rasch model as a multi-level longitudinal logistic regression model to test developmental trajectories in imitation over time as a function of clinical outcomes.

Methods: One hundred forty-six infant siblings of children formally diagnosed with autism or ASD and 87 infant siblings of typically developing children were seen at 12, 18, 24 and 36 months of age.  An imitation battery based on the Uzgiris and Hunt (1975) scales of imitation was given at 12, 18, and 24 months, and clinical outcomes were measured at 36 months of age.  Outcome classifications consisted of autism/ASD (n=20), typically developing (n=158), and other clinical concerns (n=55).  The imitation scale, consisting of ten gestural imitation items, was administered and scored on a 3-point scale – pass, partial-pass, fail.  Item scores were analyzed using a partial-credit Rasch model to estimate item difficulties, differential item functioning across time, and response category thresholds.  A longitudinal multilevel Rasch model was fit to dichotomized scores to evaluate group differences in change over time.

Results: Rasch analysis of item data suggested that all 10 items showed adequate fit using mean-square and infit statisistics..  Analysis of item data across and within the 3 ages revealed good item separation and item reliability and acceptable person separation and reliability.  Examination of category thresholds revealed that items were best represented as dichotomous data.  Analysis of differential item functioning between ages suggested good scale stability of item difficulty estimates.  A multi-level longitudinal logistic regression with item scores modeled at level-1 confirmed Rasch item scaling and further showed main effects of outcome group and age, as well as an outcome group by age interaction (Wald X² = 6.48, df = 2, p < .05), with typical infants showing the greatest growth over time compared to other groups. Group comparisons at each time point revealed that although typical subjects were significantly better than both the ASD and other clinical groups by 18 and 24 months, imitation performance did not differ between the ASD and other clinical groups.

Conclusions: Findings illustrate the use of Rasch analysis to evaluate an imitation test battery, and to model growth over time as a function of outcome.  Early deficits in imitation differentiate infants with autism from typically developing infants by 18 months, but not from those with other clinical concerns.