Optimizing Thin-Slice Observations for Toddlers with Autism

Background: The Thin-Slice is a measure of behavior from using only small observations (less than 5-minutes). When multiple raters observe the same small behavioral sample, the average score is often reliable with other measures of the same construct and valid for predicting future outcomes (Slepian, Bogart & Ambady, 2014). The Thin-Slice has been used to rate a wide variety of personality characteristics and psychological outcomes such as, marital satisfaction, depression, success, and communicative ability. The Thin-Slice has only recently been applied to observations of children with autism (Walton & Ingersoll, 2015). Although a recent application of the Thin-Slice to children with autism demonstrates reliability and validity, it does not provide guidelines for achieving an optimal Thin-Slice score based on observation length and number of raters.

Objectives: The objective of the current study is to extend the current Thin-Slice work in Autism by demonstrating the optimal number of raters and optimal number of observations to determine a stable estimate of communication ability in toddlers with autism by answering the following questions. 1) Are thin-slice observations from a) a 2-minute video and b) a 2-minute audio recording from the home environment a reliable and valid for estimating communicative ability in toddlers with autism? 2) What is the optimal number of raters and 2-minute samples from a) a video and b) audio-only to achieve a stable estimate of communicative ability in toddlers with autism?

Methods: Participants include 40 2-year olds (72% male) with autism. Each child received a diagnostic assessment using the ADOS at the start of the study. Day-long audio recordings, using LENA technology, were also collected at the start of the study. Two-minute slices of assessor-child interactions from the ADOS and 2-minute slices from the LENA recordings will be used to rate the children on 6 features: communication, speech ability, engagement, play, joint attention, and imitation using a 5-point Likert scale by 10 raters. Each two-minute slice will be rated by 20 speech-pathology students, and the scores of each rater will be averaged to obtain a single score for each observation.

Results: Results from this study are ongoing. Final analysis will include generalizability estimates for each additional rater, and a decision study will estimate the optimal number raters and slices necessary to achieve a stable measure of communication ability. Reliability with other baseline communication measures will be established (naturalistic language sample, CSBS, and MCDI) as well as predictive validity to overall ADOS severity score.

Conclusions: Although this tool is highly appealing, it is necessary to provide guidelines for researchers on how to best optimize this tool and provide reliable and valid estimates. Using snap judgements to inform diagnostics in autism may a useful and important tool not only to increase research efficiency, but to potentially provide greater access to diagnosis for more children by simplifying the screening procedure. Discussion will include applications for LENA technology and the Thin-slice methodology for increasing efficiency and accuracy in estimating communication in toddlers with autism.