Thursday, Dec. 1, 2016
In the US, students dropping out of high school is an important issue, as failing to graduate is linked to multiple problematic life outcomes, such as unemployment, lower living standards, health hazards, and shorter lives. Dropping out is especially severe in large urban schools where up to one fifth to half of the students leave secondary school before they formally complete their education. Indeed, a persistent question is to what extent students who drop out of school can be identified early in high school in an effort to help provide resources and assistance to help students succeed. In the past, researchers treated dropouts as one single group, and did not analyze covariates’ (e.g., behavior problems and school size) different influence on subgroups of potential dropout students. However, recent literature suggests that there may be multiple subgroups of students who drop out of school, a “dropout typology” theory. In a recent research study, Alex J. Bowers and Ryan Sprott (2012) detailed four types of students based on their academic performance in high school over time, with two subgroups found to be most likely to drop out.
Their research explored the data from the Education Longitudinal Study of 2002 (ELS:2002), a dataset of students sampled from US high schools by the U.S. National Center for Education Statistics (NCES) from 2002 to 2012. These samples represent students nationwide and findings are generalizable to the population of US high school students. The data involve a multitude of variables, including self-reported aspirations, dropout status, and GPA for allcourses taken in the 9th-12th grades. The two researchers focused their study on public schools that practice the semester or quarter system.
To analyze the data, the authors used Growth Mixture Modeling (GMM), a form of hierarchical linear modeling (HLM) similar to structural equation modeling. This model has three advantages. First, this model makes it possible for researchers to identify homogeneous subgroups within a heterogeneous dataset and detect relationships between selected variables at different levels of the model. Second, this model is helpful to produce insights on education data. For example, Janosz, et al. (2008) utilized GMM to identify dropout typologies of French Canadian students, and Muthen (2004) identified three different mathematics growth trajectories linked to dropping out or not. Third, with this model, researchers can guarantee that some factors in the model will not influence the outcome, so that researchers can study how the factors they care about influence the outcome. For example, if researchers intend to study the effect of behavior problems instead of school size on dropout, they may select students with behavior problems from schools of similar size and compare their dropout risk.
With GMM, the authors found four subgroups of students: Mid-Decreasing, Low-Increasing, Mid-Achieving, and High-Achieving. The results are illustrated in the figure. Based on student patterns of non-cumulative GPA (y-axis) during the first and second semesters of grade 9 and the first semester of grade 10 (x-axis), four latent class growth trajectories (shown in the four figures) are identified from the growth mixture model. The non-cumulative GPAs were calculated for each student as the grade point average from all recorded grades at each time point. In each part of the figure (see Figure 1), individual student GPA growth or decline patterns are plotted (about 5,000 students) and the latent class trajectory model overall average trajectory is plotted as a bold line.
As indicated, not all four groups of students had the same trajectory of non-cumulative GPA. Two groups stood out: the Mid-Decreasing and Low-Increasing trajectories that were 10.8% and 13.8% of the sample (about a quarter of the sample together), but surprisingly, account for over 90% of all of the dropouts. The authors identified these two subgroups as most likely to drop out. We can see that the non-cumulative GPA of students in these two trajectories started generally low and was still low at the third time point. By contrast, the non-cumulative GPA of students in the other two growth trajectories started generally high and continued to be high across all the time points. By using data from only three semesters of high school, the researchers identified the different growth trajectories of non-cumulative GPA for students in different trajectory subgroups and used this powerful strategy as a means to identify the vast majority of students who dropped out of school.
These findings can help inform educator and policymakers’ dropout prevention efforts. First, superintendents and principals should examine students’ longitudinal non-cumulative GPA, which indicates dropout risk in an early stage. Second, school leaders should consider whether it is appropriate to treat all potential dropouts as one group. Instead, the results of this study suggests that leaders may be able to identify early intervention points by examining subgroups of longitudinal student data. Third, once identified, schools may wish to provide additional but different specific resources to students in different subgroups. For example for the Mid-Decreasing students, schools could encourage participation in social aspects of schooling, such as extracurricular activities, so that students can increase engagement with schooling. For the Low-Increasing subgroup, schools could design remedial training programs to handle students’ history of academic difficulty and reduce frustrations. Schools may also consider offering psychological counseling to students with behavior problems so that they can turn their attention to their studies.
Bowers, A. J., & Sprott, R. (2012). Examining the multiple trajectories associated with dropping out of high school: A growth mixture model analysis. The Journal of Educational Research, 105(3), 176-195. http://dx.doi.org/10.1080/00220671.2011.552075.
Open Access Version: http://hdl.handle.net/10022/AC:P:21782
Janosz, M., Archambault, I., Morizot, J., & Pagani, L. S. (2008). School engagement trajectories and their differential predictive relations to dropout. Journal of social Issues, 64(1), 21-40. DOI: 10.1111/j.1540-4560.2008.00546.x. http://onlinelibrary.wiley.com/doi/10.1111/j.1540-4560.2008.00546.x/full
Muthen, B. O. (2004). Latent variable analysis: Growth mixture modeling and related techniques for longitudinal data. In D. Kaplan (Ed.), The Sage handbook of quantitative methodology for the social sciences (pp. 345-370). Thousand Oaks, CA: Sage Publications.