In light of these findings, we can draw several conclusions. There is a significant difference between large class and small class performance. And on average, larger UBC classes tend to produce lower scores. This is significant because it means we can confirm Becker and Powers’ findings that there is a negative correlation between class size and academic performance. Secondly, we can also confirm Bandiera, Larcinese, and Rasul’s findings that there are significant differences between large and small class sizes in terms of average marks.
Although we have found a good correlation between class size and grades, this by no means suggests a causal relationship between the two variables. Two possible confounding variables that could explain for such observation are year levels and faculties. We studied undergraduate courses as a whole, from 100 to 400-level courses. However, upper division classes tend to be smaller in sizes given the broad selection of courses, and one could argue that it is easier to achieve better grades because their curriculum differs from 100- or 200-level classes. In addition, although we found that the overall university trend of large class to lower average mark was sustained throughout all faculties, the differences in faculty course types and evaluations meant that the slope of the trend varied among faculties. These distinctions are especially evident when comparing small faculties like Music to larger ones like Engineering. The nature of Music classes (more engagement between instructor and students) dictates few available seats and more objective form of evaluation. Further research can be done to attempt to draw more specific conclusions by controlling the sampling method in which courses are selected in order to control for confounding variables.
Our sampling method is unique in that it does not involve surveys, so there is little room for measurement bias. However, our study was intended to assess the class size-average mark effect in only the broadest terms, without accounting for confounding variables. We did not attempt to control for differences within faculties, difficulty of upper versus lower courses, the marking discretion of professors, or any other possible factors that could influence the average mark. This was, of course, due to the fact that we only hoped to confirm or deny the existence of this effect from observation, as opposed to a detailed analysis. Future sampling methods could strive to either segregate faculties or year levels and study each separately or attempt to normalize them to a weighted standard before assessing the university as a whole.
We conducted stratified sampling, which allowed us to represent the entire university by sampling equal representation from most of the faculties at UBC in related groups. However we had trouble defining our population of interest and placing them into the appropriate strata. Some general courses, like LFS 250, encompass many disciplines and doesn’t belong in any of the strata outlined in this study. As we cannot justify placing courses like such in one discipline over the other, we had to exclude from our population. Moreover, our ways of setting strata may be questionable. Some people may disagree with placing music and visual arts\, or math and engineering in the same strata. But if we were to give each discipline its own strata, we wouldn’t have enough samples to distribute. One way to resolve this dilemma may be expanding our years of interest from 2009-2013 to the past 10, or 15 years.
Furthermore, whether or not the average marks alone could reflect a student body’s academic performance could be debated. One additional variable we should have considered is withdrawal rate. For example, let’s say a class had a certain number of students to begin with and the final average ended up being 85%. The students seemed to have performed well, right? But what if a significant portion of the class dropped out before the class was over? In our study, we failed to account for the withdrawal rates, which was not considered in calculating class averages. There are a number of ways to make up for this if we are to repeat the study again. We could recalculate the average mark appeared on PAIR report by factoring in students who withdrew, giving them a mark of 0%. Another method which is even better is to perform an additional z-test combining the percentage of students who withdrew and those who failed.
Overall, the fact that we found a statistically significant relationship between class size and average mark in both the university and individual faculties before controlling for confounding variables is promising, as it establishes a baseline trend for future researchers to compare more precise results to. We are confident that future studies in this field will generate useful data to allow universities to continue to improve their standards of education.