Approaches to Longitudinal Data Analyses

Abstract

The majority of longitudinal studies featured in this volume will feature cohorts of learners within programs designed to provide a systematic and standardized program of instruction over durations that range from multiple months to as many as 20 years. While each cohort of learners can be considered independently, as long as there are at least three measures of learning outcomes, it is possible to combine cohorts in order to examine sources of variation between cohorts. The combination of within-cohort repeated measures and between-cohort variation provides the opportunity to test hypotheses focused on the absolute trajectory of growth variation among learners, as well as the homogeneity of learner cohorts in aggregate. This can be done with conventional repeated measures analysis of variance. When cohorts are combined, and have common beginning and ending instructional cycles, a time series regression is used. In many multi-year studies, both incidental and planned interventions occur, affording the opportunity to examine their influence on the subsequent portion of the repeated measures series. When the focus is on program processes, recursive path models and cross-lagged panels with latent variables are used to examine the influence of anterior variables on proficiency outcomes. Path analyses across languages with the same sets of anterior or exogenous variables such as aptitude, motivation, achievement, and prior learning provide a basis for examining the commonality and uniqueness of different languages within a program.