Online Course Taking Behavior for On-Campus College Students Free

Online course-taking by students studying in a traditional face-to-face setting is growing rapidly. In the fall of 2016, 17%of higher education students took some but not all of their courses at a distance (Seaman, Allen, & Seaman, 2018). Students who are primarily face-to-face but supplement their programs of study with online courses may selectively enroll in online courses for a number of reasons. For example, online courses can help students navigate timing challenges associated with work or caregiving, enact preferences for online study as a meaningful portion of classwork, complete courses for which no face-to-face enrollment slots are open, and enroll in summer coursework while living far from campus.

Previous research has provided us with a general understanding of which students choose to supplement their coursework with online courses. The next step for researchers is to account for the tremendous heterogeneity that exists due to students being able to blend online and face-to-face courses in a variety of ways. In this article, we examine two specific aspects of this variation: the frequency versus the incidence of online course-taking and academic year versus summer term online coursetaking.

We examine whether the factors that are related to the probability of a student taking any online courses at all (i.e., the incidence) differ from the factors that are related to the share of courses that are online for students who take at least one online courses (i.e., the frequency). For example, technological considerations may be an important driver of incidence as an unease with or lack of access to technology may lead students to forego all online enrollment opportunities. But once a student decides to enroll in at least one online course and faces frequency-related decisions, technological considerations may decline in importance while time-based considerations grow in influence.

We also examine patterns of online course taking during the academic year and during the summer, because the nature of online enrollment can vary meaningfully between those different terms. An online course during the academic year can help a student combine a slate of face-to-face and online courses during a busy semester. In contrast, summer online courses are more likely to be the only courses taken by that student during the summer term, typically because students are not on campus during the summer months. Given these very different contexts, we might expect the student characteristics that predict online course enrollment to differ between the academic year terms and the summer term.

Previous research has not examined whether online course-taking patterns vary by academic term or between incidence and frequency, because past work has employed national datasets that provide very limited information about how students combine online and face-to-face coursework. We instead analyze a rich administrative dataset containing over 140,000 face-to-face students, who enrolled in over five million credits of oncampus and online courses between the 2005– 2006 and 2013–2014 school years. The context for the study is the Pennsylvania State University (hereafter PSU), a large, public institution that has an established online presence. Furthermore, the institution has multiple campuses, which allows for some examination of variation across contexts.

While seeking to understand variation in online course-taking, we focus our attention on three types of student characteristics: those pertaining to time or geography, historical underrepresentation, and academic preparation. To preview our results, we find that all three types of student characteristics are related to online course enrollment in general. We also find meaningful variation by academic term. The results for variables pertaining to time and geography vary between academic year and summer terms and the differences between historically underrepresented students and other students is larger in the summer term than during the academic year. In contrast, we do not find substantial differences between the factors related to the incidence of online course-taking and the factors related to frequency.

A rapidly growing literature is examining the causal effect of online course-taking on student outcomes. A portion of this literature finds a range of negative consequences of face-to-face students taking online classes. For example, studies find that face-to-face students are more likely to fail or withdraw from an online class than a similar on-campus class (Hart, Friedmann, & Hill, 2016; Huntington-Klein, Cowan, & Goldhaber, 2016; Jaggars & Xu, 2010; Johnson & Mejia, 2014; Kaupp, 2012; Xu & Jaggars, 2013). Studies that examine performance differences among face-to-face students in online classes suggest that students in on-campus classes earn higher grades than their peers in online courses (Bettinger, Fox, Loeb, & Taylor, 2014; Hart et al., 2016; Xu & Jaggars, 2014). In addition, face-to-face students who took a prerequisite course online and subsequently enrolled in the next course in the sequence, earned course grades that were on an average 1/12th of a grade point lower than their peers who took the first course on campus (Krieg & Henson, 2016).

The work cited above suggests that patterns of online course-taking negatively impact student outcomes; however, it is far from conclusive as other work finds positive or neutral consequences of face-to-face students taking online classes. For face-to-face students, taking a course online has equivalent outcomes to taking the course on campus in terms of rates of passing the course and on-time graduation (Bell & Federman, 2013; Bowen, Chingos, Lack, & Nygren, 2014). Other work has demonstrated that taking an online course is associated with retention between first and second year for face-to-face students (Fike & Fike, 2008). Shea and Bidjerano (2016) find that community college students who take online classes are more likely to graduate on time than their peers who do not take online classes because when face-to-face courses fill up, students who can take them online graduate more quickly, rather than waiting for an extra semester or two for an on-campus course to have room for them (Shea & Bidjerano, 2016).

A much smaller literature has examined the question underlying this article: which face-to-face students enroll in online courses? This work has primarily been conducted using data from national surveys such as the National Postsecondary Student Aid Study (NPSAS), which has advantages and disadvantages to the institutional dataset used in this study. The National Postsecondary Student Aid Study possesses a national representative sample and extensive background information on the sampled students, but it does not capture detailed information on the student’s online course-taking behavior.

In general, previous research has found that online enrollment is related to student characteristics pertaining to time or geography, historical underrepresentation, and academic preparation. Student characteristics pertaining to time or geography relate to the central feature of online education that helps explain its growth: it eliminates the need for students to commit to being in a specific location at a specific time on a weekly basis throughout a semester. The removal of restrictions pertaining to geography and time will be most helpful to students who have extensive work or caregiving responsibilities, because these students may be unable to fulfill these responsibilities if they had to travel to and attend a large number of face-to-face classes. Past work has found that older students, female students, married students, students with dependent(s), and students currently employed are more likely to enroll in online courses (Jaggars, 2014a; Ortagus, 2017).

Previous work has also found that historically underrepresented students are less likely to enroll in online courses (Jaggars, 2014a; Ortagus 2017; Wladis et al., 2015). Although technological innovations have the theoretical promise of leveling the educational playing field, this potential is not always realized and innovations can sometimes perpetuate patterns of inequality, rather than ameliorate them. Paul Attewell (2001) theorized this phenomenon as the two digital divides: “one of access and one of usage” (p. 253). In order for digital innovations to play an equalizing rather than stratifying role, students of color need similar levels of access to online learning, in addition to similar patterns of usage of online learning. In Attewell’s theorizing, usage described how students interacted with computers (whether they used them to perform rote tasks or complex and creative tasks). Our work makes a contribution to understandings of these digital divides by investigating variation among underrepresented students in their rates of summer online course-taking versus academic year enrollments, in addition to differences in incidence and frequency of online course-taking.

Past work has focused less on whether or not a student’s academic preparation is associated with online course enrollment, but some suggestive evidence does exist. Wladis et al. (2015) found that students with weaker academic preparation prior to college enrolled in online courses at higher rates. From interviews with students, Jaggars (2014b) found that students were more likely to take a course online if they believed the subject area would be easy for them. If they projected the course as difficult, they were more likely to take it in a face-to-face setting.

Past work has also considered student characteristics that are not directly examined within this study. Harringtion and Loffredo (2010) demonstrated that introversion is associated with a preference for online classes and extroversion was associated with a preference for face-to-face classes. While online course takers noted convenience and interest in technology as reasons for preferring online courses, face-to-face course takers explicitly stated preference for that modality being about in-person connection and learning through listening (Harrington & Loffredo, 2010). A study about which face-to-face students persist in enrolling in additional online courses beyond their first experience, found that face-to-face students who enrolled in additional online courses reported greater confidence in their online learning ability and more enjoyment of their recent online learning course (Artino, 2010).

This study uses student-level data from PSU. Founded in 1855, PSU is a public, state-related research-intensive university, with a large campus located at University Park and 19 commonwealth campuses and 5 special-mission campuses located across the state of Pennsylvania. In addition to these 25 residential campuses, PSU launched its World Campus in 1998 to offer online programs. In the fall of 2015, PSU had about 47,000 students enrolled at University Park, 38,000 enrolled across 19 Commonwealth and 5 special mission campuses, and another 12,000 at the World Campus, reaching a total enrollment of 97,000 students.

Although the idea and practice of distance and online education is not new for colleges and universities in the United States, PSU is one of the first major accredited universities to offer online degree programs in the nation. Currently, PSU provides more than 125 degree and certificate programs and has been consistently ranked among the top institutions offering online baccalaureate degree programs by U.S. News and World Report. In this study, we are interested in learning the online course taking behavior of face-to-face students at PSU. Although the vast majority of online courses at PSU are offered through the World Campus to online-only students, face-to-face students across all campuses can take advantage of these online offerings. Other than World Campus, different Penn State campuses can also offer their courses online to their face-to-face students. In this analysis, we do not make a distinction between online courses offered by the World Campus and those offered through other PSU campuses, because the difference is mainly in how these courses are managed rather than how these courses are delivered.

PSU maintains a data warehouse that provides detailed student-level administrative information, including demographic background, transcripts, and academic progress. The data used for this analysis include all undergraduate students who have enrolled and taken courses for at least one credit from AY 2005–2006 to 2013–2014. We start our data collection from AY 2005–2006 because online courses accounted for a very small proportion (about 1%) of undergraduate enrollment prior to that year and have increased tremendously since then. Our sample includes all undergraduate students who enrolled at PSU during any academic year regardless of the number of years they have been PSU students. Because the composition of students, regarding years of study, is relatively stable across years, we can examine variations of online course-taking patterns both within and across years. Other strategies such as organizing students by entering cohorts have proved to be less desirable because it is difficult to isolate the effects of years of study and the increasing supply of online courses over time.

Data files were extracted from various components of the data warehouse. The first set of files provides student characteristics pertaining to three areas: time and geography, historical underrepresentation, and academic preparation. We group variables associated with time and geography together, because previous work has demonstrated that they are associated with the likelihood of enrolling in online courses (Ortagus, 2016). Variables that indicate state residence and citizenship status give us leverage on understanding students who are contending with geographical barriers to accessing education. Nontraditional-aged students, especially women, may have work and care-taking responsibilities that make online course-taking an attractive educational option. The time and geography group of variables include age, gender, state residence, and citizenship. Historical underrepresentation is captured by student self-reported race and ethnicity variables. Academic preparation is measured by high school grade point average (GPA), as well as SAT verbal and math scores. In addition to these key predictors for online course taking, student-level data files also included college enrollment information (e.g., admission year and semester, first admitted campus, academic departments, and credits obtained each semester). A small proportion of students (about 0.1%) without demographic information available are removed from our analysis. In addition, some students have changed their campus affiliation over the course of their undergraduate study. Because the number of online courses offered varies across campuses and by level of study, changes in campus affiliation may give rise to complicated interactions between campuses and levels of study. To keep our analyses focused, we removed students who changed their campus affiliation during the time period for which we had data. Finally, about a quarter of students in the dataset did not have complete data on precollege academic performance. Models are estimated with and without these variables (i.e., high school GPA and SAT scores) to check for the robustness of our findings.

Based on this student-level information, a series of dummy variables are created to indicate whether a student is female, Black, Hispanic, Asian, out-of-state resident, permanent resident, and noncitizen. Table 1 lists individual-level variables extracted and created from these files, broken down by University Park and Commonwealth Campuses. Not listed in this table are also a series of dummy variables indicating a student’s entering year, level of study, and academic department. These variables are used in our empirical models to account for variations of online course taking by year, levels of study, and academic departments. Because the online course offerings vary across levels of study, we use 30 credits per year to compute levels of study to account for the fact that academic progress may vary across students (e.g., full-time vs. part-time students). For example, a student who has taken 40 credits is considered a sophomore, regardless of the actual length of time she has spent at PSU.

The second set of files contain student transcript data, which provide detailed data on each course a student has taken since attending PSU including year and semester, campus location, course number, course section, number of credits, and whether it was offered online or not. These files can uniquely identify an individual course section via the following information: year and semester, campus location, course number, and course section. Two additional variables were created for each student. The first is incidence of online course taking (i.e., whether a student has taken any online courses at all during his/her time at PSU). The second is frequency of online course taking (i.e., the proportion of courses that were taken online for those students who have taken at least one online course).

Table 2 presents the share of course enrollments generated by online courses, weighted by the number of credits for each course. Several observations can be made. First, there is variation in the online share of course enrollment by campuses. Across years and semester, the online share of enrollment has consistently been higher at University Park than at the Commonwealth Campuses. This may suggest, from the supply-side point of view, that the University Park campus has been taking a leadership role in online education. It may also suggest, from the demand-side point of view, that students at University Park have a higher preference for online courses than students at Commonwealth Campuses. Second, summer semesters have a much higher online share of course enrollments than regular semesters, which is consistent with our prediction that geographic considerations become more important when students return home for the summer. Finally, there has been a steady increase in the online course-taking over time, which is true for both University Park and Commonwealth Campuses. Taking University Park as an example, during the fall semester of AY 2005–06, only 2.39%of total course enrollments are generated by online courses, while this share increased to about 6% by the fall semester of AY 2013–14. This upward trend holds for all semesters.

We estimate empirical models at both the course and student level. At the course level, we are interested in understanding the differences in student characteristics between online and face-to-face courses. Logistic regression is used because the dependent variable is either 1 (i.e., online course) or 0 (i.e., face-to-face course). For the convenience of interpretation, we report odds ratios from logistic models. At the student level, we are interested in understanding factors that are related to the probability of a student taking any online course at all (i.e., the incidence) and the proportion of online courses for those students who took online courses (i.e., the frequency). Logistic regression is used for the incidence model because the dependent variable is binary. For the frequency model, because the dependent variable is a proportion, which is a continuous variable between 0 and 1, and skewed to zero, we estimate a fractional logistic model (Papke & Wooldridge, 1996). Fractional logistic regression allows for predictions within the [0, 1] range and captures particular nonlinear relationships especially when the dependent variable is near 0 or 1. Because both incidence and frequency are calculated based on students’ entire academic history at PSU, only those students who joined the university since AY 2005–06 are included for this set of analyses.

Table 3 reports results from a series of courselevel models that estimate the differences in demographic and academic variables between online and face-to-face courses. Models are estimated separately for University Park and Commonwealth Campuses, with three models estimated within each set. The “demographic” model includes students’ demographic variables, while the “academic” model has an additional set of academic variables. Both models control for year and level of study fixed effects that account for variations in online course offerings across years and levels of study. Finally, the “full” model adds department fixed effects that account for variations in online course offerings across departments. The pseudo R-squares across these three models indicate that demographic and academic variables in general have low predictive power in explaining differences between online and face-to-face courses; however, substantial variations exist across departments, especially at University Park. Because these models are estimated at course level, the estimated effects represent the differences in the characteristics of students who are taking online versus face-to-face courses. In other words, they do not differentiate, for example, between one student taking multiple online courses and multiple students taking one online course each.

Results in Table 3 are generally consistent—albeit with minor differences—across the three models and between the University Park campus and the Commonwealth Campuses. Online courses are more likely to be taken by older students than by younger students, and the estimated age effect is slightly larger at Commonwealth Campuses than at University Park. Compared with face-to-face courses, online courses are more likely to be taken by female students than by male students, especially at Commonwealth Campuses. At University Park, the odds of female students filling online enrollment slots are 2%higher than their male counterparts; however, this gender difference is 26%at Commonwealth Campuses. In terms of differences among race/ethnicity groups, online courses are less likely to be taken by minority students, whether they are Black, Hispanic, Asian, or other races. For example, the odds of Black and Hispanic students filling online enrollment slots are 10–20%lower than their White counterparts. The lower odds ratio for Asian students is wiped out and reversed at University Park when we move to the full model where department fixed effects are considered. This finding suggests that the distribution of Asian students at University Park is concentrated in fields of study that offer few online courses.

Results also reveal differences by residence and citizenship. Online courses are more likely to be taken by out-of-state students when compared with Pennsylvania resident students. For example, the odds of out-of-state students filling online enrollment slots are 20% higher than their in-state counterparts at University Park and 5%higher at Commonwealth Campuses. In addition, online courses are less likely to be taken by noncitizen students with permanent resident status. Finally, in the full model where field differences are considered, the odds of noncitizens filling online enrollment slots are higher than citizens at University Park but not at Commonwealth Campuses.

Academic variables exhibit different patterns between University Park and Commonwealth Campuses. In the full model, while online courses at University Park are more likely to be taken by students with higher high school GPA, the opposite relationship is present for Commonwealth Campus students. Higher SAT math scores are positively related to taking online courses at University Park, but this relationship does not seem to hold for Commonwealth Campus students. SAT verbal scores, however, are associated with lower probability of taking online courses, both at University Park and Commonwealth Campuses. Finally, we might expect that online courses are more likely to be taken by students who took more credits in a semester; however, this relationship only holds at Commonwealth Campuses but not for students at University Park.

The inconsistent pattern among the three academic variables (i.e., high school GPA, SAT math, and SAT verbal) is intriguing. To investigate whether this may be due to high correlations among these academic variables, we run a set of models with only one of these three academic variables in each model. Results are reported in the Appendix (top portion for University Park, bottom portion for Commonwealth Campuses). At University Park, this additional exercise yields basically similar results as in the baseline model. In fact, the correlations among these three academic variables are within the range between 0.4 and 0.6. These results seem to suggest that the decision to take online courses is partially based on different cognitive skills, at least for those enrolled at University Park. At the Commonwealth Campuses, the relationship between test scores and online course-taking did not differ by subject area.

Table 4 reports separate analyses for fall/spring and summer semesters. As highlighted earlier in the paper, we might expect academic year course-taking patterns to differ from those occurring during the summer term. For example, at University Park, although online courses are more likely to be taken by older students in general, this age difference is larger during summer semesters than fall/spring semesters. At Commonwealth Campuses, online courses are more likely to be taken by older students during fall/spring semesters but not during summer semesters. While the odds of female students filling online enrollment slots are higher than their male counterparts at University Park, as Table 3 shows, this gender difference is mainly due to the difference during summer semesters. At Commonwealth Campuses, the higher levels of online coursetaking by females occur in both fall/spring and summer semesters. A consistent pattern emerges among minority students: Online courses are much less likely to be taken by minority students than their White counterparts, especially during summer semesters. As expected, the odds of out-of-state students filling online enrollment slots are higher than resident students during summer semesters than during fall/spring semesters.

At University Park the positive relationship between high school GPA and online course taking mainly occurs during fall/spring semester but not during summer semester, while at Commonwealth Campuses, the negative relationship between high school GPA and online course taking is mainly due to summer semesters but not fall/spring semesters. The two SAT scores have shown consistent patterns between semesters and across campuses. In other words, SAT math scores are positively related to online course-taking, while SAT verbal scores are negatively related to online course taking during both fall/spring and summer semesters and both at University Park and Commonwealth Campuses. Finally, the number of credits and online course-taking are negatively related.

Course-level analyses do not differentiate between incidence (i.e., some students are more likely to take at least one online courses) and frequency (i.e., some students take more online courses). To further investigate these differences, we conduct student-level analysis that use incidence and frequency as two separate outcome variables. Results in Table 5 suggest that at University Park although older students are less likely to take any online courses than younger students, those who did take online courses did not take a smaller percentage of their courses online. At Commonwealth campuses, however, older students are both more likely to take any online courses and to take more online courses than younger students. While results for Commonwealth Campuses in Table 3 suggest that female students are much more likely to take online courses, this gender difference is mainly due to higher incidence. For example, at University Park, for those students who took any online courses at all, female students actually took a smaller share of their courses online than their male counterparts. Minority students are both less likely to take any online courses at all and to take fewer online courses among those who took online courses. Compared with Pennsylvania resident students, out-of-state students are both more likely to take any online courses and to take more online courses among those who took online courses at all. Students who are not U.S. citizens but with permanent resident status are left behind in both incidence and frequency of online course taking. Finally, noncitizens have higher incidence and frequency of taking online courses at University Park but not at Commonwealth Campuses. With few exceptions, incidence and frequency tend to go hand in hand.

As far as academic variables are concerned, high school GPAs are negatively related to both the incidence and frequency of taking online courses at both University Park and Commonwealth Campuses. At University Park, students with higher SAT math scores are more likely to take any online courses and take more online courses; this relationship is weaker for Commonwealth Campus students. High SAT verbal scores, however, are associated with lower incidence and frequency of online course taking for University Park students. Finally, as expected, students who have taken more credits are more likely to take any online courses; however, they are not taking a higher proportion of their courses online.

We find that online course-taking is related to student characteristics pertaining to time and geography, historical underrepresentation, and academic preparation in a number of compelling ways, and we turn now to a discussion of the implications of our findings. In regards to the first category, time and geography, we confirm previous findings in the literature that female students and older students are more likely to take online courses. To the extent these students have greater work or caregiving responsibilities, online coursework may help them reduce the opportunities costs associated with advancing their education.

Our results for gender, however, suggest that the differences between male and female students contain considerable complexity. The observed differences by gender were much larger for the Commonwealth Campuses than for the University Park campus. Furthermore, the gender gap was driven by summer coursework for University Park students and academic year coursework for Commonwealth Campus students. The results for Commonwealth Campuses better align with explanations based on time constraints where female students use online courses to help juggle academic work and caregiving responsibilities during the academic term. In contrast, the results for the University Park campus suggest that the gender gap in online enrollment is driven by female students’ willingness to advance their education during the summer, even though they are residing far from campus.

Previous research has found that face-to-face students from underrepresented racial and ethnic groups were less likely to enroll in online courses than White students (Jaggars, 2014a; Ortagus, 2017; Wladis et al., 2015). We extended our understanding of these differences by finding that racial/ethnic differences were greater during the summer semester; Black and Hispanic face-to-face students were much less likely than their White peers to enroll in online courses in the summer semester. We can only hypothesize the possible mechanisms behind this finding. To the extent that these students may be from disadvantaged backgrounds, these differences in summer online enrollment may be related to the competing demands of summertime work and caregiving responsibilities. These differences may also be attributed to differential access to high speed broadband (Skinner, 2017) or to social capital networks that would supply knowledge about summer course enrollment options (Cox, 2017).

Previous research has found general differences by academic background, but differences by test scores areas have not previously been revealed. We find that students at University Park with strong SAT scores in math are more likely to enroll in online courses, while students with strong verbal SAT scores are less likely to enroll in online courses. Future work can further investigate the relationship between online course-taking and test scores and seek to understand the forces underlying the observed relationships.

Beyond enriching our understanding of the relationship between student characteristics and online course-taking, our work contains general insights into online education. We find that meaningful heterogeneity is present in regards to differences between academic year and summer terms, which has important implications for practice, as the benefits of and barriers to online education vary by term. Administrators may want to consider organizational practices and policies that ameliorate barriers to summer online course-taking, as the summer term creates distinct challenges for certain student population. Potential interventions might offer students both technological and financial support.

Acknowledgment:The authors would like to thank Shi Pu, Yahya Shamekhi, and Mark Umbricht for excellent research assistance.