Study
Using a matched, randomized-control design, Low and colleagues (2015) evaluated the elementary version of Second Step (2011 edition). A total of 61 schools across districts in Arizona (1 district) and Washington State (5 districts) participated in the study. Both urban and rural settings were represented. After accounting for attrition, 319 teachers (223 from Washington, 96 from Arizona) participated and 6,558 students (4,232 from Washington, 2,326 from Arizona) participated. The 61 schools were randomly assigned to either the early-start treatment group (n=31) or the delayed-start control group (n=30). Schools were paired and matched based on percentage of nonwhite students and participation in free and reduced lunch. No significant differences were found between treatment and control groups.
Students in the study were in kindergarten through second grade. The sample in Arizona was 40.1 percent white, 0.3 percent Asian, 5.9 percent African American, 47.1 percent Latino/a, 6.3 percent Native American, 0.3 percent Native Hawaiian or Pacific Islander, and 10.1 percent unknown. The sample in Washington was 45.8 percent white, 18.2 percent Asian, 8.1 percent African American, 14.7 percent Latino/a, 1.6 percent Native American, and 1.7 percent Native Hawaiian or Pacific Islander; in addition, 9.9 percent reported more than one race, and 20.4 percent were unknown.
Fall 2012 and spring 2013 data was collected for baseline and follow-up observations, respectively. Student behavior was measured using the Devereux Student Strengths Assessment (DESSA) via online surveys completed by teachers. The Strengths Difficulties Questionnaire (SDQ) was also completed by teachers to measure student behavior.
A behavioral observation system was implemented and conducted by trained graduate students who coded their observations of students as either on-task, off-task, or disruptive. This system was developed based on the Behavioral Observation for Students in Schools (BOSS). Students were observed for 2 minutes total across 10-second intervals. Data was then analyzed using a mixed-model, time x condition analysis.