Co-enrollment density predicts engineering students’ persistence and graduation: College networks and logistic regression analysis

Highlights

• College retention is a concern for educational institutions and researchers.

• Low retention of engineering students may cause workforce shortages.

• Low retention of engineering students may cause a loss of competitiveness.

• Co-enrollment density is a novel metric estimated using enrollment records.

• Co-enrollment density may predict retention in 4-year programs in engineering.

Abstract

College retention is a concern for educational institutions and researchers. This concern is particularly acute in engineering for reasons including workforce shortages, economic competitiveness, social justice, and socioeconomic equity. This study presents the evaluation of co-enrollment density (CeD) for engineering students at eight medium and large American public universities over 24 years. CeD is a novel metric estimated using enrollment records that may predict retention in 4-year bachelor of science programs in engineering. Graduation and persistence were fitted to CeD with logistic regression. Students in denser co-enrollment clusters—high CeD—tend to graduate more than their classmates in less dense neighborhoods—low CeD. The regression models predict graduation with odds ratio intervals 95 % CIs [3.24, 4.81] and area under the receiver operating curve [0.76, 0.80]. CeD is more sensitive to students who do not persist, particularly after the first year, so CeD’s cut-off points may be indicators for dropouts' risk.

Introduction

There is evidence of declining interest in engineering and STEM careers (Becker, 2010; Belser et al., 2018; Johnson, 2013; Johnson & Jones, 2006; Sithole et al., 2017). It raises concerns over a workforce shortage and a concomitant negative effect on economic competitiveness (Committee on Prospering in the Global Economy of the 21st Century: An Agenda for American Science and Technology, 2007). Approximately 50 % of students who enroll in engineering graduate from those programs at some point (Aljohani, 2016a; Braxton et al., 2013). A small fraction (<20 %) graduate in four years, and less than half within six years (Johnson, 2013; Zhang et al., 2004). Retention is the rate at which students who start college graduate at the same Institution (Berger et al., 2012; Titus, 2004). The graduation rate for bachelor's degree students was 63.2 % at public 4-year institutions from 2003 to 2009 (Johnson, 2013). Student departure impacts educational success markers, yet institutions have not been able to mitigate it (Aljohani, 2016a; Braxton et al., 2013; Tinto, 2006). This study is motivated by the need to get metrics to forecast students at risk for dropout—early and efficiently (Sithole et al., 2017).

The terms related to students leaving college are not standardized. Table 1 shows the concepts used as a framework for this work, and Fig. 1 shows their primary relationships. The definitions are not meant to oversimplify the complexity of the phenomenon (Rintala et al., 2012) but provide a conceptual framework for this work. Persistence and retention are the student and institution's perspectives of continued enrollment; similarly, dropout and attrition are perspectives on students leaving college. Attrition does not imply that the student does not ever graduate, just that he/she dropout from a particular institution.

Retention is a critical issue in tertiary education; therefore, it has been studied from many different perspectives; however, the diversity of the educational settings where the phenomenon occurs, and its campus-based nature, prevents generalizable results (Aljohani, 2016b; Barbera et al., 2020). The development of multi-institutional datasets (Ginder, 2013), such as the one used in this research (Ohland & Long, 2016), has facilitated identifying general trends typical of multiple institutions and programs, and institution-specific findings.

The engineering curriculum is diverse (Corlu et al., 2018). The roots for western engineering education, as known today, may be traced to early European technical schools. Continental Europe approached engineering as a public service involving knowledge of advanced mathematics and science; the École Nationale des Ponts et Chaussés in France is an example of such schools. In contrast, Anglo-American engineers were trained on the job. England’s early engineering schools represented such a model that evolved after World War I when industries demanded engineers with higher scientific knowledge levels (Corlu et al., 2018). However, even today, there are many skills and knowledge to train engineers (Lucena et al., 2008; Passow & Passow, 2017). The diversity of the discipline’s curriculum offers an additional layer of complexity to the inquiry on persistence and graduation in engineering, which is part of the emerging field of research in engineering education (Borrego & Bernhard, 2011). Jesiek et al. (2009) observed a sense of ambiguity about the identity and status of engineering education research.

Problem/Project–Based–Learning (PBL) and Conceive–Design–Implement–Operate (CDIO) are initiatives that propose to reform the engineering curriculum and may impact engineering education at the colleges involved in them (Bennedsen et al., 2019; Edström & Kolmos, 2012, 2014; Malmqvist et al., 2015). Regional projects like the ATTRACT (Enhance The Attractiveness of Studies in Science and Technology) may improve STEM degrees in Europe. ATTRACT’s preliminary results were reported for the ATTRACT’s work package-8 “Students Retention” that ran from January 2010 to October 2012 (Rintala et al., 2012). The actions to improve retention were divided into three strategies: Structure of studies, progression rules, and human support. Universities in partnership with the project reported that the levels of student's preparation, commitment, and motivation, along with study skills, had the highest impact on attrition (Rintala et al., 2021). The American Board of Engineering and Technology's outcomes-focused criteria for engineering program accreditation are examples of a quality-driven approach for designing and implementing the engineering curriculum (Akera et al., 2019; Dobryakova & Froumin, 2010). There is evidence that engineering program accreditation improves retention (Al Busaidi, 2020). The evolving nature of engineering education makes the study of retention a challenge, but it should not be an excuse to stop the efforts on its research and improvement.

According to Aljohani (2016a), “The larger body of student retention studies were designed and conducted in the American higher education contexts.” Nevertheless, there are notable empirical studies and strategies to improve retention in Australia, Great Britain, and Europe. Australian higher education institutions and government have developed projects to improve retention focused in the first year. They have approached the retention issue based on educational experiences' quality (Hodges et al., 2013; Krause & Armitage, 2014; Willcoxson et al., 2011). The University-Experience Survey, the Course-Experience-Questionnaire, and the First-Year-Questionnaire are examples of instruments to understand how the quality of education impacts its outcomes. Hodges et al. (2013) reported that programs providing free educational access to government-targeted equity groups had higher attrition rates than conventional undergraduate degree programs, suggesting that the lack of financial interest leads to less accountability and engagement. The argument was in line with student engagement theories (Tight, 2020). Jones (2008) reviewed ten fundamental studies on retention in the British educational system. The attrition factors were classified in individuals' characteristics, such as educational goals, preparation for college education, abilities, institutions' teaching quality, and fit and satisfaction with the institution.

There are more than four decades of literature on retention (Aljohani, 2016b; Burke, 2019; Melguizo, 2011; Tinto, 2006). The study of retention/persistence has evolved from a psychological approach (student's attributes) to a sociological perspective focusing on the student/college relationship (Astin, 1999; Lin, 2020; Tinto, 1997, 2006). Despite the large and more recent body of research on retention, the Longitudinal Model of Student Departure (Tinto, 1975, 1993) still has paradigmatic status. While Tinto's theory has enlightened the subject's inquiry, it presents limitations: It applies to traditional, non-minority, homogeneous student-bodies enrolled in 4-year programs, and it does not consider external factors. Further, Tinto did not provide empirical evidence on the relation of integration and persistence or even an instrument to measure integration (Braxton et al., 2000, 2013; Melguizo, 2011; Seidman, 2005). This paradigm has made the longitudinal path analysis expected to inquire on retention (Chapin, 2019; Cominole et al., 2007; Lee & Ferrare, 2019; Ohland & Long, 2016; Wang & Wickersham, 2014; Zhang et al., 2004). Tinto (1997) recognized some of the limitations of his theory, proposed a different perspective based on learning communities, and remarked:

"…we would be well served by supplementing our use of path analysis to study the process of persistence with network analysis and social mapping of student interaction patterns. These will better illuminate the complexity of student involvements and the linkages that arise over time between the classroom and out-of-the-class experiences."—the emphasis is ours.

The study of college student networks is still an emerging field, with few examples exploring the classroom/course relations, and even fewer relating it with persistence (Biancani & McFarland, 2013; Forsman et al., 2014, 2015). Yet, the study of college student networks may assist path analysis to uncover the complexity of students’ interaction patterns and their impact on persistence. Network analysis typically requires the acquisition of relational data, which is usually obtained with surveys. Thomas (2000) published a study of records and surveys for 322 first-year students, applied a network analysis, and fitted the results to persistence constructs. The author found that GPA, intent, institution commitment, and goal commitment explained 26 % of the persistence variance; each of these factors had a more significant predictive relationship than the two measures of integration combined. Grunspan et al. (2014) published a classroom network analysis; the authors related the network's positions with success on exams. The data collection was survey-based, and the construction of relational data applied formal network analysis.

Gardner et al. (2018) and collaborators probed that a CE network may predict grades; this work was the closest to ours. Israel et al. (2020) reported the analysis of a student–course network at the University of Michigan. They found that students of the same majors clustered, but transfer students showed low centrality. These results were expected; however, the authors did not relate the network's parameters with academic outcomes. Their assumption that sharing a lecture hall implies social contact was unwarranted. Some studies use the term co-enrollment to refer to dual-enrollment, which is defined earlier and is conceptually different (Crisp, 2013; Wang & McCready, 2013; Wang & Wickersham, 2014). CE does not provide the type of relational data required for orthodox social network analysis, which may explained why there are so few examples of such an application. Instead of assuming that CE implies social contacts, we propose that CE may be a proxy for academic effort levels and that CeD may reveal clusters of students with similar commitment with the goal of graduation. The Data and Methods section shows the network analysis that explores this idea (Kang, 2019; Niu, 2020).