The centerpiece of the study design is the use of random assignment to create a group of teachers exposed to comprehensive teacher induction (treatment) and an equivalent group exposed to the districtís usual set of induction services (control). The study design allows us to measure and compare outcomes for these two groups to estimate the impacts of comprehensive induction relative to the services teachers receive from their districtís prevailing induction program. As discussed below, we used surveys, classroom observations, and school records to measure the background of the study teachers, their receipt of induction services and alternative support services, their attitudes, and their outcomes related to the studyís main research questions: classroom practices, student achievement, and teacher mobility.
We recruited 17 school districts to participate in the study. The districts, which were spread across 13 states, served low-income students, with every district in the study having more than 50 percent of its students qualifying for the federal School Lunch Program. We then assigned each district to one of the two providers of comprehensive induction, either ETS or NTC, based primarily on district preferences. The preference-based method of assigning districts to providers does not allow for and should not be used to make direct comparisons of one provider to the other.
Within each district, a subset of elementary schools participated in the study. The study used an experimental design in which we randomly assigned elementary schools within each of the 17 participating districts to either a treatment group, which received comprehensive teacher induction—from ETS or NTC, depending on the district—or a control group, which took part in the districtís usual teacher induction program. Districts nominated approximately 500 schools across the 17 districts. It turned out that some schools that we targeted for random assignment had no eligible teachers, so the final sample sizes included 418 schools: 100 treatment schools and 103 control schools in the 9 ETS districts and 110 treatment and 105 control schools in the 8 NTC districts.
With each study school, we selected all eligible teachers, defined as beginning teachers who met certain criteria: taught in an elementary grade (K-6); were new to the profession; and were not already receiving induction support from a teacher preparation or certification program. The 418 schools participating in the study contained 1,009 eligible teachers.
Not all of the 1,009 teachers eligible for the study were eligible for all analyses. We limited the collection of classroom practices data to 698 teachers who met certain eligibility requirements such as teaching English/language arts to a self-contained classroom. Because we focused on reading instruction, it was not appropriate or even possible to include teachers such as music, art, or math specialists who were not responsible for teaching reading. We limited the collection of student test score data to teachers meeting another set of eligibility criteria, including teaching a self-contained classroom in a tested grade and subject. This resulted in the collection of reading test scores for 281 teachers and math scores for 261 teachers.
Eligible teachers in a school were either all exposed or all not exposed to treatment, a method known as cluster random assignment. Cluster random assignment was necessary because varying the types of induction services available in the same school building could result in contamination of the control group. Therefore, we assigned all eligible teachers to treatment or control status based on the school where they were expected to teach at the point of random assignment.
We found that random assignment produced groups that were equivalent on a wide variety of teacher and school characteristics. Of the dozens of baseline attributes we examined, we found statistically significant differences between treatment and control groups in one area: teacher assignments.2 The control group contained a higher percentage of special subject teachers (such as art and music) than did the treatment group (7 versus 3 percent) and consequently a lower percentage of teachers who taught just a single grade (79 versus 85 percent) and who said they were responsible for math (85 versus 90 percent) or reading outcomes (83 versus 91 percent). Accounting for such differences did not change the studyís conclusions.