Grant Closed

Attrition Benchmarks Across Students and School Contexts: Evidence from Student Mobility in National Longitudinal Survey Data

NCER

Program: Statistical and Research Methodology in Education

Program topic(s): Early Career

Award amount: $199,980

Principal investigator: Jordan Rickles

Awardee:

American Institutes for Research (AIR)

Year: 2015

Award period: 3 years (07/01/2015 - 06/30/2018)

Project type:

Methodological Innovation

Award number: R305D150026

Purpose

This research provided empirical benchmarks for the amount of attrition that can arise from natural rates of student mobility. Student-level, school-based longitudinal evaluations of education practices and interventions often encounter participant attrition when students move away from, or drop out of, study schools. In some cases, this mobility-induced attrition simply decreases sample size, weakening the study's power to detect a treatment effect. Attrition can also introduce bias into an otherwise well-designed study, thereby posing a threat to internal and external validity.

Project Activities

In order to construct a series of mobility-based attrition benchmarks for different student populations and school settings, the researchers used four nationally representative longitudinal surveys. The research team reviewed attrition levels reported in recent randomized controlled trials in order to provide context for the benchmarks derived from the national datasets. The team conducted a simulation study to investigate the quality of the results from typical approaches for estimating treatment effects when attrition is an issue (e.g., weighting and multiple imputation). Researchers disseminated the results of the study via peer-review conference presentations and journal articles. The research team also produced an on-line guide that applied researchers can use to gauge, potential expected attrition rates during the planning phase of a study and analytical approaches to deal with realized attrition once data collection is complete.