ABSTRACT
Although generalizability theory has been used increasingly in recent years to investigate the dependability of behavioral estimates, many of these studies have relied on use of general education populations as opposed to those students who are most likely to be referred for assessment due to problematic classroom behavior (e.g., inattention, disruption). The current study investigated the degree to which differences exist in terms of the magnitude of both variance component estimates and dependability coefficients between students nominated by their teachers for Tier 2 interventions due to classroom behavior problems and a general classroom sample (i.e., including both nominated and non-nominated students). The academic engagement levels of 16 (8 nominated, 8 non-nominated) middle school students were measured by 4 trained observers using momentary time-sampling procedures. A series of G and D studies were then conducted to determine whether the 2 groups were similar in terms of the (a) distribution of rating variance and (b) number of observations needed to achieve an adequate level of dependability. Results suggested that the behavior of students in the teacher-nominated group fluctuated more across time and that roughly twice as many observations would therefore be required to yield similar levels of dependability compared with the combined group. These findings highlight the importance of constructing samples of students that are comparable to those students with whom the measurement method is likely to be applied when conducting psychometric investigations of behavioral assessment tools.