|
Abstract:
|
RAND Corporation's American Teacher Panel (ATP) is a large survey panel of over 20,000 teachers nationwide that has been built using probability sampling with dramatic oversampling in certain states. However, due to low rates of response (e.g., 30% recruitment rate, 65% response rate among the panel), rigorous statistical methods must be used to account for non-response bias. Calibration is prudent for matching auxiliary panel characteristics to known population totals. Exact population totals are known for several school-level school level characteristics via the CCD; however, populations totals for teacher-level characteristic (e.g., gender, terminal degree, experience) are unknown but have been estimated using government surveys that have a smaller sample size but better response rate than the ATP. The problem of calibrating to estimated totals has been considered in detail by Dever (2008); here, we apply some such methods to the ATP (e.g., we use a random perturbation of population totals for jackknife replication groups). Furthermore, we use simulations to examine the necessity and efficacy of such techniques.
|
