Comparing Student Outcomes in Blended and Face-to-Face Courses

Abstract

This article reports on a study of student outcomes in a pair of matched courses, one taught face-to-face and one taught in a blended format, in which students completed most of the work online but met several times face-to-face. Learning objectives, course content, and pedagogical approaches were identical but the mode of instruction was different. The data suggest academic performance was not influenced by the mode of instruction. Additionally, the data show few differences between the sections in terms of attitudinal outcomes, although students in the blended section reported a lower sense of community. However, one notable outcome was the tendency of students to drop out of the online activities after having a face-to-face session. Lessons for the design of blended courses are discussed.

Keywords:

• blended learning
• online learning

Notes

Note: Instruction Mode is coded 1 for students in the blended section and 0 for students in the face-to-face section. The dependent variable for Exam 1, Exam 2, Exam 3, and Final Grade is a percentage score. The dependent variable for the Essay scores ranges from 0–30. Cell entries are unstandardized OLS regression coefficients; T scores are given in parentheses.

*p < .05. ***p < .001.

*First activity after a face-to-face class session.

Though it would have been possible to simply record the face-to-face lectures, this approach was deemed less optimal than stand-alone podcasts. The main reason is that a lecture given to a group of people takes place within a context that may not be translated well as a simple audio recording. Facial and other nonverbal cues, the history of that particular class and reference points that only they understand (such as inside jokes) and other factors may make listeners feel distanced from the lecture. Thus, it was decided that a lecture created specifically for online listeners would be better received and easier for students to engage.

Few of the control variables are statistically significant. This partly reflects the fact that, for most of the variables, the rank-order relationships are weak or nonexistent. But, also, for some variables, particularly the GPA variables, there is considerable collinearity. In fact, many of the GPA coefficients do become statistically significant if only one of the GPA variables is included in the model.

In an important sense, the coefficients must be taken at face value, as they represent the actual differences among the students in the two sections. Given the specific question of this study—whether there are differences between the students in the blended and face-to-face classes—the concern is less that the inferences to a larger population be accurate and more that potentially important differences between the two groups of students are controlled, as they are in the regression models. In this sense, the research design is much like a pseudo-experiment, with statistical control taking the place of random assignment.

Essays A and D were on Exam 1; Essays E and F were on Exam 2; Essays B and C were on Exam 3.

Control variables are not appropriate here, because there is no reason to expect academic aptitude and diligence to be key predictors of student attitudes in the same way they predict academic performance in the course. Moreover, the course assessments were necessarily anonymous and completed by the students in different parts at different times in the semester, so it would be impossible to include external data. For the same reason, it is not possible to create any kinds of composite measures.

Small class sizes may be a necessary aspect in this sort of research. Many courses are capped at relatively small sizes for a variety of reasons. It is important to examine these types of research questions in these kinds of classes as well as in large ones in order to expand the contexts to which our findings are applicable. Pooling over several semesters may be possible, but changes in the content or pedagogy of the course over time make this difficult. Moreover, as pointed out earlier, the most critical concern is that the comparison groups be representative of each other, or made so through statistical control. This concern is far more important than the issue of making the classes representative of some larger population of students. Indeed, even in studies with very large classes, the students are not representative samples of a population.