Comparing the Accuracy of Coding Methods for A Low-Incidence Behavior

Background: Event coding provides an accurate and reliable method of assessing the occurrence of behaviors (Thompson, Symons, & Felce, 2000).  Event coding, however, can be impractical in cases where observers must code multiple variables, or the behavior occurs at a low frequency (Thompson, et al., 2000).  Time sampling measures, such as interval coding and momentary time sampling, provide an approximation of the occurrence of behaviors assessed through event coding.  While these time sampling measures are useful in providing a sampling of behavioral patterns, they may misrepresent these patterns in low-incidence behaviors (Hartmann & Wood, 1990; Odom & Ogawa, 1992; Sackett, 1978), such as the social behaviors of preschool-aged children with Autism Spectrum Disorder (ASD).  

Objectives: The purpose of this study is to compare the data from three observational methodologies (event coding, interval coding, and momentary time sampling) when used to assess the social behaviors of children with ASD towards adults and peers.  

Methods: Participants included 100 preschoolers with ASD.  Trained research staff collected a 30-minute, videotaped observational sample during center time.  Social behavior  (defined as any gestural/motor or vocal/verbal behavior) was coded towards adults and peers using three methods: momentary time sampling (coded social behaviors that occurred at the end of each 10-second intervals), interval coding (coded either yes or no for the occurrence of social behaviors within a 10-second interval), and event coding (tallied the number of social behaviors during each 10-second interval). Two trained research assistants obtained 80% IOA across the coding methods, and 20% of videos will be coded by both coders for IOA.  For each method, the total number of social behaviors towards adults and peers were converted to the rate of social behaviors per minute.    

Results: Participants directed social behavior towards adult at a higher rate per minute (0.32 for momentary; 1.49 for interval, and 2.01 for event) than to peers (0.14, 0.35, and 0.46, respectively).  Pearson’s correlation was used to determine the association between each coding method.  For social behaviors towards adults, interval and event were highly correlated (r = 0.97), followed by event and momentary (r = 0.694) and interval and momentary (r = 0.612).  For social behaviors directed towards peers, each method was highly correlated with the strongest correlation between event and interval (r = 0.990), followed by event and momentary (r = 0.941), and finally momentary and interval (r = 0.925). Data is based on 34 videos.  The analysis will be completed by March 2011.  

Conclusions:  Preliminary results indicate a strong correlation between momentary time sampling, interval coding, and event coding for social behavior directed towards peers (Ratner, 2009).  The correlation for social behavior directed towards adults is strong between momentary time sampling and interval coding, but only moderate between event and momentary time sampling, and interval and momentary time sampling.  While there were differences among the coding methods in the rate of social behaviors per minute, momentary time sampling and partial interval coding provided an accurate approximation of the frequency of low-incidence social behaviors.