2026 ASEE Annual Conference & Exposition

In this paper we present a formal model for characterizing the causal impact numerous factors have on college students’ time-to-degree (TTD). This model was constructed using the data associated with a large-scale observational study involving thirty different institutions of higher education in the United States, including more than 750,000 students. The aim of this study was to better characterize the impact of curricular complexity on various student success outcomes, including TTD. The importance of TTD is reflected in the fact that any unnecessary delays in graduation may lead to numerous negative outcomes, including an increase in student debt, a decrease in lifelong earnings, a decrease in institutional ranking, etc. This issue is particularly relevant in engineering programs, which are notoriously difficult to complete in a timely manner.

One experimental approach for determining the causal impact of curricular complexity on TTD involves constructing a randomized control trail (RCT). This approach involves randomly assigning students with various backgrounds characteristics to programs with different curricular complexities at different institutions. The approach is commonly used in medical studies to determine the efficacy of particular treatments. In our case, we can consider curricular complexity as the treatment a student receives, and TTD as the outcome. If we have a sufficiently large sample of students, this RCT will isolate the impact of curricular complexity on TTD relative to any other student background factors. Specifically, the randomization ensures that at each complexity level the biases associated with confounding variables are “averaged out.” Because RCTs eliminate the biases introduced by confounders, they are considered the “gold standard” for estimating the impact of particular treatments on specific outcomes. Borrowing from the terminology of the medical profession, we will use the term average treatment effect (ATE) to designate the average causal effect of a treatment or intervention on an outcome across an entire population of study participants. In the case of a RCT, the ATE is easily estimated by comparing the average outcome of the treated and untreated subpopulations in the study, or by comparing the differences between two treatment levels if there are multiple possible treatments. From this, it is easy to see the ATE represents the average change in outcome an individual in a population would expect to experience if they were exposed to one level of treatment, versus a different level of treatment. Note that the intervention in our study is not binary, i.e., a person receives the intervention or not; rather, the students in our study received many different levels (i.e., doses) of the intervention, and the ATE in this case must be computed at each intervention level. A dose-response function can be constructed from these ATEs to describe how the average outcome changes as the intensity (i.e., dose) of the intervention varies.

Because it is not possible, indeed ethical, to require students to enroll in academic programs that are not of their choosing, an RCT is not practical for studying the impact of curricular complexity on a student’s TTD. Rather, we have a retrospective dataset that includes the academic programs students graduated from over the 2010-2020 time period at the institutions involved in this study, along with various factors related to these students. Because this dataset lacks a controlled intervention, i.e., assignment of students to academic programs, the best we can do is design an observational study that closely approximates an RCT, and in so doing provide insights into the relationships between curricular complexity and student success outcomes without actually manipulating them. The main challenge associated with observational studies involves accounting for the various biases that are introduced due to confounders. For instance, it is well known that students with superior high school preparation, as indicated by high school GPA, are more likely to graduate on time. However, our data also shows that these same students are more likely to choose majors with higher curricular complexity, thereby negatively impacting their TTD. Thus, high school GPA is a confounding variable, and there are numerous other such confounding variables related to TTD, including first generation status, ethnicity, gender, Pell award status, etc.

In this observational study, we started from the set of all students in our dataset who graduated over the past ten years, and we modeled the impact curricular complexity had on the number of years it took these students to graduate. Because this “backwards-looking” approach starts from those who graduated, and then traces their trajectories backwards in time in order to uncover how they achieved their degrees, it serves to isolate factors such as the complexity of the curriculum a student confronted. Specifically, we know each student in this dataset confronted, at the very least, the complexity of the curriculum associated with the degree program they graduated from, i.e., the curriculum they completed, even if they had a change of major along the way. The same can be said for all other students who graduated from the same degree program.

Without controlling for any confounding factors, the data in this study shows that for most students, curricular complexity has little impact on their ability to graduate within six years; however, it has a significant negative relationship with a student’s ability to graduate on time, i.e., in four years. We know, however, as described above, that numerous factors impact a student’s TTD. In order to control for these factors, we utilized a generalized propensity score (GPS) methodology that involved estimating the probability of specific treatment levels conditioned on these confounding variables. These GPS values were then used to construct a causal model that allowed us to control for the effects of the confounding variables. From this we are able to construct a dose-response function that characterizes the causal impact of curricular complexity on TDD. By controlling for factors such as a student’s high school GPA, we found the impact of curricular complexity on TTD is more profoundly negative. Specifically, the effect of curricular complexity on TTD is more pronounced (and statistically significant) when accounting for high school GPA, first generation status, and Pell award status. Because student with higher high school GPAs tend to self select into higher complexity programs, while first generation and Pell award students tend to select away from higher complexity programs, the ability to accurately account for these in a causal model allows us to more accurately asses the true impact of curricular complexity on TTD. Another important aspect of this causal model is that it supports causal inferencing, allowing us to reason about counterfactual events; that is, events that did not actually occur. For instance, the model provided here allows us to ask student success-related questions such as, “what would happen if we were able to reduce the complexity of all undergraduate degree programs in the College of Engineering by 10%?” The usefulness of such a model in guiding student success efforts is easy to see.