Profile Analysis via Principal Component Analysis for Longitudinal and Cross-Sectional Profiles in the Early Childhood Longitudinal Study (ECLS-K)

In education, a primary means of evaluating student performance remains the analysis of standardized test scores. When such tests are taken repeatedly over several school years, the data may be utilized to evaluate student growth over time as well as skill acquisition within each test administration. Such longitudinal and cross-sectional analyses may be useful for a variety of stakeholders, including policy makers, administrators, teachers, and parents. Profile analysis, or the detection of patterns in subtest scores (Stanton & Reynolds, 2000), provides a means to represent samples of students with a small number of summative, or core profiles. These profiles provide information regarding the samples’ levels of achievement at each time point and the pattern of change demonstrated over multiple time points. In addition, individual student profiles can be determined and then compared with other members of specific subgroups, such as between and within gender, ethnicity, and socioeconomic status, thus providing valuable information regarding individual student’s progress as compared to larger subgroups. In this study, the use of Principal Component Analysis (PCA) is evaluated as a method for extracting longitudinal and cross-sectional summative profiles from six assessments given as part of the Early Childhood Longitudinal Study (ECLS-K). For both the full sample and subgroup analyses (gender, ethnicity, SES), there were one level- and two pattern-components retained consistently for both the longitudinal and most of the cross-sectional analyses. The level component explained approximately 50% of the total variance in the data, while the pattern components combined explained approximately 80% of the within-person (pattern) variance. In addition, a regression analysis utilized the cross-sectional pattern-component scores as predictors of future assessment scores and to explore relationships with teacher academic rating scales and student self-descriptive questionnaires. Overall, little of the variance in future scores, teacher ratings, or student questionnaires was explained by the pattern components. Though the PCA successfully extracted meaningful longitudinal and cross-sectional profiles, the usefulness by educational stakeholders is questionable, due to the prevalence of “flat” profiles and the potential difficulties in correctly interpreting the profiles