New Directions for Studying Doctoral Attrition: Employing Agent-Based Modeling to Investigate Attrition Decisions in Engineers

• In Event: Attrition (Optional) and Relationships (Required): Educational Relationships and Decision Making to Increase Engineering Graduate Student Persistence

Abstract

Although doctoral attrition rates in STEM disciplines are lower than in other fields, they are still considerable. In engineering, according to recent reports, the ten-year doctoral completion rates for men and women in engineering are only 65% and 56%, respectively, with numbers much lower for students from marginalized groups. These numbers are substantial given that engineering academic norms tend to contradict oft-cited reasons for doctoral attrition in other disciplines: In addition to being a fast-completing discipline, with the average time-to-PhD being about 5 years, over 80% of graduate engineering students are funded, and students often work in research groups rather than in isolation. There are also many high paying jobs in engineering industry that employ PhD-holders, and most engineering PhDs do not desire academic careers. Engineering education scholars have begun investigating the decision-making process, including when and how engineering students begin to consider departure. However, it is difficult to follow enough students collect statistically significant data to understand generalizable attrition decision-making patterns; infeasible to be able to control for certain factors in a doctoral student’s life; and unethical to place a graduate student research participant within suboptimal situations for the sake of research. As a creative solution, in this paper, we demonstrate an extension of our deep experience studying engineering attrition using qualitative methods into the computational realm using agent-based modeling.
Agent-based modeling is a machine learning technique by which theory is used to inform the decision-making process of a digital “agent.” Given a set of conditions, the “agent” makes decisions according to theory. This procedure has been used to study design thinking in engineering students and practitioners, in military defense strategy, and to teach computers to play strategy-based games such as chess. In this paper, we present the methodological framework for transforming theory and qualitative narrative inquiry from five years and over 100 interviews with current and former engineering doctoral students across the United States into a machine learning decision-making framework called Attrition Laboratory Using Modeling (ALUM). While we anticipate in future work that ALUM will be able to be used as a research platform to test multiple configurations of plausible attrition-decision factors and effects on persistence, this paper focuses on the methodological development of the model and provides proof-of-concept results for the ALUM platform.
There are two impactful contributions this paper presents. First, this paper is a call to explore emergent technologies to harness what researchers do know about attrition from traditional methods, extending into new, cross-cutting, and multidisciplinary domains. Second, within this conversation, though, is a careful emphasis on the disciplinary-specific aspects of attrition: For example, given the unique facets of engineering, while funding, time-to-completion, and future career goals are parts of the “mental calculus” of attrition, they likely are weighted differently in engineering than in other disciplines, even science, technology, or mathematics, based on the landscape of the research enterprise and career trajectories. By extending past traditional qualitative and quantitative research traditions, we can continue to disentangle the complexities of doctoral attrition.

Authors

• Catherine Berdanier, Pennsylvania State University

• Chris McComb, Carnegie Mellon University