The influence of observation length on the dependability of data.

Although direct observation is one of the most frequently used assessment methods by school psychologists, studies have shown that the number of observations needed to obtain a dependable estimate of student behavior may be impractical. Because direct observation may be used to inform important decisions about students, it is crucial that data be reliable. Preliminary research has suggested that dependability may be improved by extending the length of individual observations. The purpose of the current study was, therefore, to examine how changes in observational duration affect the dependability of student engagement data. Twenty seventh grade students were each observed for 30-min across 2 days during math instruction. Generalizability theory was then used to calculate reliability-like coefficients for the purposes of intraindividual decision making. Across days, acceptable levels of dependability for progress monitoring (i.e., .70) were achieved through two 30-min observations, three 15-min observations, or four to five 10-min observations. Acceptable levels of dependability for higher stakes decisions (i.e., .80) required over an hour of cumulative observation time. Within a given day, a 15 minute observation was found to be adequate for making low-stakes decisions whereas an hour long observation was necessary for high-stakes decision making. Limitations of the current study and implications for research and practice are discussed.