When there is a sense of unease whether college students are learning and developing well, it is rarely the case that there is a clear digital situation that pinpoints the “difference that makes a difference.” A digital situation is when there is a “correct” or an “incorrect” answer (similar to the on/off switch on a laptop). Certain conventional statistical procedures may lend credence, if interpreted incorrectly, to the notion that the “difference” may be found in an analysis that compares the scores of two subgroups of students; for example (i.e., yes/no p ≤ .05?). No one single finding is persuasive, which is another form of a digital situation. Rather, the purpose of data analytics is to identify patterns and anomalies in students’ learning and development over time. Students’ learning and development are incremental processes that are far too complex for “digital” analyses. It is for this reason that colleges and universities would benefit from applying artificial intelligence to analyze data using an “analogue” strategy (imagine a high-tech control panel for lighting and sound, with a full array of levers and dials that can create an almost infinite variety of options between nothing at all and everything, full blast, at once). Even considering whether different subgroups of students have scores that are more similar than different is an analogue situation due to the awareness that there isn’t one correct answer for all the students on campus. To explain why students’ learning and development are complex processes that defy conventional statistical procedures, it is worthwhile to widen the frame for a moment.
Instead of focusing on “ages and stages” of development, the developmental sciences (developmental psychology, cognitive science, neuroscience) investigate students’ “trajectories.” Shifts in developmental trajectories are intentional veers from that which immutable demographic characteristics and prior learning alone might have predicted. A developmental trajectory is the lifepath of the student that has been influenced by the past, present, and image of the future. Students are placed on a developmental trajectory and will continue along this lifepath unless the trajectory is disrupted. For example, students who are placed on a developmental trajectory, including one that leads to success in college, may reach the conclusion that they were set on this lifepath by others and decide to reorient themselves towards a different trajectory that leads to dropping out of college. They set in motion an intentional veer from their anticipated lifepaths.
Understanding the conditions that impede or facilitate students’ developmental trajectories requires studies that follow individual students from orientation through graduation or subsequent enrollment in other institutions of higher education. A “snapshot” of students’ learning and development does not provide sufficient evidence to inform policy decisions. Merging all the “bits and pieces” of data into long-term cohort datasets ensures that the results of point-in-time evaluations are not held in isolation from other evidence. In contrast to immutable student demographic characteristics, we have found that malleable characteristics among students (such as academic habits of mind, sense of belonging, and future orientation) predict student success.
Our fifteen longitudinal datasets combine data from different computer information systems and performance-based assessments of students’ learning and development (the first study was launched in 2007). Each longitudinal dataset contains over 1.9 million individual data points. The understanding of developmental trajectories presented here emerged from predictive models built using machine-learning techniques and AI cognitive analytics to discern patterns and anomalies in data collected on student success as part of these longitudinal, cohort studies. In addition, SPSS Statistics was used to conduct linear and binary logistic regression analyses, and AMOS to conduct structural equation modeling. The aim of using different analytic approaches was to confirm the research findings. The more that the same findings emerge from different analytic platforms and techniques, the more confidence in the findings.
Similar findings were observed from the machine-learning techniques, AI cognitive analytics, and conventional statistics. In a 2017 paper “Predicting students’ graduation outcomes through support vector machines,” we describe our use of machine-learning techniques to predict students’ graduation. Up to about 100 features, including a set of factors to measure both students’ learning and their development, were employed to construct the predictive model. The findings confirmed what had been learn from AI cognitive analytics: Students’ incoming profiles do not define their destiny. Learning and developmental experiences that they have after enrollment are far more important in predicting retention, academic achievement, and graduation. Applying an AI strategy provided the most useful information. Students’ developmental trajectories are complex, and the AI was able to handle the complexity.
Collecting data on students’ learning and development promotes “analogue” thinking because the developmental sciences consider all the experiences that students have on campus within the same framework, as well as their experiences over time. An AI strategy is useful when analyzing all the “bits and pieces” of data about students’ learning and development.