Abstract:
Project Talent is a large, nationally representative longitudinal study developed by the American Institutes for Research and conducted from 1960 to 1974. The goals were to assess the interests, abilities, and demographics of 9th–12th graders and to follow their trajectories into adulthood. More than 1,200 junior and senior high schools participated. Replicate weights were not constructed at the time, preventing the estimation of standard errors. Today, Project Talent is being revived to study the physical, cognitive, economic, and social processes of aging. In this paper, the retrospective construction of 104 sets of student-level replicate weights is described. Partitioning analysis was performed to generate variance strata and variance primary sampling units. The student-level replicate weights were constructed using a jackknife procedure. The process included adjustment of the base year weights and calibration of (full sample and replicate weights) to the total number of secondary school students in the U. S. in the spring of 1960. The use of replicate weights is illustrated by estimating standard errors for means of composite cognitive scores constructed from student questionnaires. We also describe construction of mortality- and nonresponse-adjusted weights for the three Project Talent followup data collections.
Keywords:
Project Talent, Weight adjustment, Recursive partitioning, CHAID, Jackknife replicate weights, Survey analysis
tr194.pdf