First Years Inventory: Examining Measurement Invariance across Age, Sex and Prematurity Status

Background: The heterogeneous nature of infants at risk for autism indicates a need to better understand variation in early risk markers, yet there are few studies testing measurement invariance on parent-report or self-report autism screeners. The evaluation of item invariance across individual characteristics is important to assess the impact of diagnostic bias. Recent research has indicated item functioning differences by age for the M-CHAT (Sturner et al., 2017) and by sex for the AQ-10 on adults (Murray et al., 2017), which suggest the need to further assess how these factors influence the utility of autism screeners for very young populations.

Objectives: To examine whether items on the First Years Inventory (FYI) version 3.1, a parent questionnaire designed to assess behaviors in 12-month-olds that suggest risk for a later diagnosis of autism, function differently with infants of different age, sex, and prematurity status.

Methods: A community sample of 6,395 parents who had a child 8 to 16 months of age completed the FYI consisting of 69 items, which can be classified into two domains: Social Communication (SC) or Sensory Regulatory Functioning (SR). Differential Item Functioning (DIF) as an indicator of measurement invariance was assessed using a hybrid ordinal regression-item response theory (IRT) method across age (8-10, 11-13, 14-16 months), sex (male or female) and prematurity status (<36 or ≥36 gestational weeks) groups, separately for each domain. The cumulative impact of DIF on domain scores was evaluated if DIF was detected, where ΔR² exceeded the thresholds generated through Monte Carlo simulations of DIF-free samples.

Results: 30 out of 36 SC items and 18 out of 26 SR items were detected as having DIF across age groups, 15 among them showed large DIF (ΔR²≥.07), while only 2 SR items showed moderate DIF (.035≤ΔR²<.07). 12 SC items and 5 SR items were flagged as having negligible DIF (ΔR²<.035) by sex. Regarding prematurity status, 3 SC items were flagged as having negligible DIF, while none of the SR items were flagged. The impact of cumulative individual-level DIF on SC domain scores (differences between scores accounted for and not accounted for DIF) varied by the level of measured trait (theta), while the mediation effect of trait was not significant for SR domain scores (Figures).

Conclusions: Results suggest significant differences in the responses of parents with infants of different ages, particularly for behaviors related to social communication. The SC domain scores were overestimated if DIF by age was not accounted for in the younger group (8-10 months). Furthermore, the level of overestimation increased as children showed less actual risks in social communication. In contrast, there was an underestimation of SC scores for the older groups (11-13 and 14-16 months), which became more significant as children showed more actual risks in social communication. On the other hand, the item functioning was found to be invariant by sex and prematurity status for both domains. The findings revealed the potential utility of the FYI with very young infants, especially in consideration of the developmental effects on social communication.