Information Economics and Policy 48 (2019) 32–39
Contents lists available at ScienceDirect
Information Economics and Policy journal homepage: www.elsevier.com/locate/iep
Providing MOOCs: A FUN way to enroll students?R Julien Jacqmin∗ HEC Liège, University of Liège, B. 31 Place des Orateurs 3, 4000, Liège, Belgium
a r t i c l e
i n f o
Article history: Received 25 April 2018 Revised 29 August 2018 Accepted 25 October 2018 Available online 30 October 2018 JEL classification: I23 M3 L86
a b s t r a c t This paper aims at assessing whether universities improve their enrollment in traditional taught inperson programs by providing MOOCs (Massive Open Online Courses). Focusing on French universities, we look at the impact on enrollment figures of providing MOOC on France Université Numérique (FUN), a MOOC platform. We find that new students intake of universities offering MOOCs rises over 2% the following academic year, all else being equal. Analyzing data related to the characteristics of students enrolled in online and traditional programs as well to media coverage, we see two intertwined explanations for this positive relationship. One relates to the enhanced information available to students when making their enrollment decision; the other lies in the attention-grabbing - and likely persuasive - character of the event created around MOOCs.
Keywords: MOOCs Enrollment Higher education institutions
1. Introduction In 2011, two Stanford professors, Peter Norvig and Sebastian Thrun created a website offering a free access to their “Introduction to Artificial Intelligence” course. The course consisted of short videos, quizzes and online discussion forums. With over 160,0 0 0 students enrolled worldwide, including 23,0 0 0 of whom received a certificate of completion, they gained unprecedented success. From there, Sebastian Thrun went on to create Udacity, a platform hosting other MOOCs. Following the development and success of two other MOOC platforms, edX and Coursera, The New York Times declared 2012 as the “Year of the MOOCs”. Clayton Christensen, who first coined the concept of “disruptive innovation”, further predicted “‘wholesale bankruptcies’ over the next decade among standard universities” (The Economist, 2012). Nonetheless, it remains questionable that MOOCs or other online educational programs will soon substitute sometimes centuries-old, higher education institutions. Various empirical studies (Figlio et al., 2013; Xu and Jaggars, 2013; Alpert et al., 2016; Bettinger et al., 2017) have questioned the quality equivalence between the online and live mode of delivery. Distance education technologies do have an impact on higher educaR Special thanks to Mathieu Lefebvre, Vincent Starck, Marco Martiniello, Paul Belleflamme, Sandrine Delacroix-Morvan, Isabelle Peere, Sébastien Broos, Tommaso Agasisti and Matthieu Manant, as well as seminars and conference participants in Maastricht, KULeuven, Rouen, Besancon and Paris for valuable comments. ∗ Corresponding author. E-mail address:
[email protected]
https://doi.org/10.1016/j.infoecopol.2018.10.002 0167-6245/© 2018 Elsevier B.V. All rights reserved.
© 2018 Elsevier B.V. All rights reserved.
tion institutions, although of a different nature, yet as complements rather than substitutes for them. First, as argued by Deslauriers et al. (2011) and Belleflamme and Jacqmin (2016), internet technologies facilitate the implementation of new pedagogical approaches e.g. by providing students with (quasi) instantaneous feedback and customizing courses enhancing the quality of the programs. Second, providing online courses can reduce the costs of traditional higher education programs by curbing Baumol’s cost disease (Bowen et al., 2014). For example, basing themselves on data relating to U.S. post-secondary education, Deming et al. (2015) find that institutions providing online courses - and especially for-profit and low-ranked public ones are charging lower tuition fees. Finally, MOOCs can be seen as a means of improving the visibility of higher education institutions (Howarth et al., 2017), i.e. as a sample/stepping stone towards traditional programs. Based on this last argument, this paper studies the effects of online courses provision on subsequent enrollment figures in traditional programs, focusing on French universities, some of which have provided MOOCs for free on the France Université Numérique (FUN) platform. Since its creation by the French government in collaboration with French universities in 2013, FUN has hosted close to 1.2 million different students in over 350 courses. To analyse these figures, we elaborate a panel dataset with data aggregated at the university level. We believe that the French open enrollment system and the absence of editorial line on the platform make for a suitable context for studying the impact of MOOC provision on future enrollments.
J. Jacqmin / Information Economics and Policy 48 (2019) 32–39
Using a fixed-effect identification strategy that controls for year- and university-specific unobserved heterogeneity, we observe that MOOC provision has a positive and significant impact on new students intake. The effect for the universities concerned is roughly equivalent to climbing up 75 ranks in the Shanghai ranking. We also find that non-Parisian universities and universities with a good Shanghai ranking observe a comparatively larger effect on their enrollments. Finally, students enrolled in high school programs preparing them for university are more likely to be influenced in their enrollment decision by the MOOC provision. The study coming closest to the present one is that of Goodman et al. (2019) bearing on the parallel offer at Georgia Institute of Technology of a fully online and a taught in person Master’s programs in computer science, and analysing the impact of the former on enrollment on the latter. While both programs are similar contentwise, the former is less selective and six times less costly to students. What comes out of this study is first, that owing to this disparity, both programs appear to attract non-overlapping pools of candidates, but also that introducing the online program has increased the amount of applications in the traditional one. Thus these authors observe that this higher education institution, by investing the field of online education, did not by any means ”cannibalize” its traditional activities, on the contrary. In addition, considering traditional student characteristics, media coverage and enrollment data on the MOOC platform, brings further evidences of the two underlying mechanisms behind this positive relationship. The first one finds its roots in the informational asymmetries between students and prospective traditional higher education institutions. By providing MOOCs, universities inform students about imperfectly observed characteristics, such as their will to innovate in new forms of pedagogy, their fields of specialization or their scholastic tradition.1 As they hear about or try out the free online samples of the traditional programs available, MOOCs may function as a means of informing students and impacting their enrollment decision. Hence, this channel sees MOOCs as a way to better inform students in their enrollment decision. This explanation comes close to a recent strand of the literature which examines the effect of information provision on students’ enrollment decision. So far though, the majority of papers have considered the provision of information about financial aid policies (Bettinger et al., 2012), personalized counseling (Pistolesi, 2017), university rankings (Luca and Smith, 2013), student satisfaction surveys (Chevalier and Jia, 2016) or about job prospects (Hurwitz and Smith, 2018). The second explanation lies in the fact that attention is a scarce resource. The buzz surrounding MOOCs in the social and traditional media has caught students’ attention and influenced their enrollment decision. Accordingly, MOOCs are a persuasive rather than an informative marketing tool. Similar attention-catching events have been analyzed in the higher education context: e.g. the performance of college sports teams (Pope and Pope, 2014) or negative incidents like campus scandals (Luca et al., 2016). This paper develops as follows: Section 2 describes the French higher education system. Section 3 presents the data; Section 4 develops our estimation strategy; Section 5 presents our results and Section 6 our conclusions. 2. The French higher education system Reviewing the French higher education system, and especially about universities and the MOOC landscape draws attention to two 1 Hence, our work also relates to the literature that has looked at how new digital goods and services have impacted closely related live counterparts in the music, movie, advertising or news sectors (see for example Zentner, 2012; Smith and Telang, 2010 or Cho et al., 2016).
33
significant peculiarities: the open admission (or non selective) system of French universities and the absence of editorial line in FUN, the French MOOC platform. Contrary to the ‘grandes écoles’, universities are not allowed to select students, i.e. enrollment is open to all; and contrary to professional colleges, universities transmit academic rather than applied knowledge and skills. Besides teaching, university professors also do research. In 2014, 57% of the students in higher education were enrolled in a university (Ministère de l’enseignement supérieur, de la recherche et de l’innovation, 2016b). Compared to other OECD countries, tuition fees are very low: per year it is about 184 euros for a bachelor’s program and 256 euros for a master’s program. These fees apply to all universities and have been constant over the past few years. Overall, universities enroll more than 1.6 million students, including 14% of international ones. To this, it is important to add that public funding is allocated to universities largely on the basis of the number of students enrolled. Enrollment at French universities is open at both the bachelor and the master level.2 This is a striking difference compared to the other settings where the determinants of enrollment in higher education have been studied. The only condition of enrollment at a university is to have obtained the baccalauréat at the end of their high school years, a criterion which is fulfilled by about 80% of the students in recent age cohorts. Despite the impossibility to select students, in 2008 the French ministry of higher education decided to create a website called Admission Post-Bac (APB) to regulate and centralize first year student enrollment (see Pistolesi, 2017 for an in-depth discussion). This is due to the fact that some disciplines (law, sports studies and psychology) are locally constrained in terms of capacity. By the spring preceding the start of the academic year, students must indicate the institutions of their choice in the descending order of preference and have until the last day of May to modify this. Based on an (unpublicized) algorithm whereby students stand a better chance to enroll in universities of their own region, they are then assigned to a program and an institution in June/July. In 2016, 86.1% of the applicants were enrolled in the discipline taught at their first university choice (Ministère de l’enseignement supérieur, de la recherche et de l’innovation, 2016a). On October 2013, Geneviève Fioraso, the then minister of higher education, launched France Université Numérique (FUN). A government-supported MOOC platform, FUN has emerged from the collaboration between French universities and uses the open source program Open edX (Dupont-Calbo, 2013). It is important to note that FUN was developed to host courses provided by all French higher education institutions, and - unlike US platforms such as Coursera or edX- without editorial filters from the platform or a privilege granted only to top ranked institutions.3 As of beginning 2018, FUN hosted more than 1.2 million different students and 360 courses. Despite hosting courses mostly in French, 33% of the students are from abroad. Most of the institutions providing courses on FUN are French universities. All the courses are free of access and require no selective procedure. The vast majority of them are taught in French and focus on a variety of topics like computer science, health or economics. While some “grandes écoles” are active on edX and Coursera, the two most well-known MOOC platform at the world scale, this is not the case for French universities. As of the beginning of 2018, only one uni-
2 As of the academic year 2017–2018, the rules have changed and universities are now allowed to select students at the master level, even if in some cases this was unofficially already taking place. Starting from the academic year 2018–2019, universities will be able to set prerequisites to select students at entry. 3 Studying the consequences of platforms with an editorial line would necessitate a more evolved estimation strategy, like two-sided matching methods, as not all institutions may provide the MOOC they want on the platform of their choice.
34
J. Jacqmin / Information Economics and Policy 48 (2019) 32–39
versity, Sorbonnes Université, had courses on edX. Hence, in the timespan considered here, French universities almost exclusively provide MOOCs via the FUN platform. 3. Data To test the relevance of universities’ MOOCs provision to their subsequent student enrollment, we have constructed a database about French universities’ enrollment characteristics and their information about the MOOCs they have provided. Our enrollment data is coming from the French ministry of higher education (Ministère de l’enseignement supérieur, de la recherche et de l’innovation, 2018). For each year t and each of the 72 French universities i, we have the (log of the) number of newly enrolled students (newenrollmentit ). Across these five academic years between 2012 and 2016, the number of newly enrolled students has increased by 17%.4 The information about the MOOCs provided on FUN was scrapped from the FUN website. This information was translated in a database into different explanatory variables. Note that the years of our data on enrollment and MOOC are matched according to the timing of the centralized APB system where students indicate their institutional preferences. Hence, we examine the impact of a MOOC offered before the end of May on enrollment in September, i.e. the beginning of the following academic year, the reason for this being that receiving the “MOOC treatment” after May would otherwise have no impact on the students’ enrollment decision. MOOCdumit is a dummy equal to one in year t if a university i provided a MOOC in year t. Besides, to clarify the impact of MOOCs, we consider the following explanatory variables. MOOCperyearit is the number of MOOCs per year provided by an institution. MOOCrerunit is a dummy equal to one if one of the MOOCs aired that year was rescheduled or had a sequel. This can be seen as a proxy to measure the quality of the MOOC offered. Finally, we also classify the MOOCs according to their fields (law and business, health, humanities and science).5 In addition, to test the mechanisms at work behind our main relationship, we also use proprietary data from FUN about MOOCs enrollment as a measure of student exposure. # of participantsit is the number of person and # of young participantsit is the number of young people (aged between 17 and 23, i.e. the age of prospective students) who enrolled on a MOOC from the university i in year t aired on the FUN platform. In addition, as a proxy of the publicity of the MOOC created in the media, we computed the number of search results on Google, for the year preceding the enrollment process, as done in Luca et al. (2016). This variable is media coverageit . Our search terms were “the title of the MOOC” + “MOOC”. Finally, using the europresse archives, we counted the number of times the MOOC was mentioned in the regional and national press to construct the variable press coverageit . All these variables are then aggregated at the university/year pair. In total, 12 MOOCs from 6 universities were provided before the enrollment decision of the academic year starting in 2014, 27 MOOCs from 18 universities for 2015 and 35 MOOCs from 22 universities for 2016. On average the universities providing MOOCs have attracted 9452 students annually in their online courses, 15% of whom are in the age range we classify as young. 4 In further robustness checks presented in Section 5.3, we analyze data about these students’ characteristics (namely, their academic standard, geographical origin and gender). This figure, however, covers students who passed the baccalauréat a few months before entering university, not those who enrolled later, e.g. on a master’s program. Unfortunately, data at the subject-group level about newly enrolled students is not available. 5 Most if not all the MOOCs provided on FUN state that they require no prerequisite and very few of them are taught in English. Hence, we were not able to test any hypothesis about these characteristics.
In addition, we also control for institution level characteristics to limit the presence of omitted variable bias. As we will use both year- and institution- fixed effects, the challenge is to find variables varying across time and institutions. The first control variable, (the log of) enrollmentlevelit measures the size of the institution with the total number of enrolled students. We lag this variable by one year as it measures the number of students at the time the students decide to enroll and it avoids the issue of “bad control”. We use a proxy of the (perceived) prestige of the institution as reflected in the Shanghai ranking, which features the world’s best 500 higher education institutions. Despite the many issues, related among other things to their microeconomic underpinnings, several papers have shown that they indeed impact students’ choice of institution (see for example Broecke, 2012; Luca and Smith, 2013 or Beine et al., 2014). To facilitate our interpretation, the variable unirankit is 500 minus the institution’s rank, 500 being the lowest one. It is also important to note that this ranking is highly censored, as on average only 16 French universities are ranked annually. For this reason, we work in two steps. First we build a dummy variable equal to 1 if the institution is ranked and to 0 otherwise. Second we use a variable with the ranking information and where the censored institutions are assigned a value of 0. We also control for the attractiveness of the city hosting the institution by using the ranking of the best university cities published annually by the French magazine L’étudiant.6 The ranking is based on various aspects, such as the housing context, the cultural offers, the job market prospects, the quality of the public transport system or the weather conditions. It receives a lot of press coverage and is much discussed on social media. As for the Shanghai ranking, we construct Cityrankit by subtracting the rank from the rank of the number of cities ranked and we work in two steps as not all the university cities are ranked. For these two ranking measures, we do not include the dummy variables. We also control for mergerit : a dummy equal to 1 if the university has merged with another one. In the five-year span of our analysis, 9 institutions went into a merger process. We also control for the demographic context with the number of people between the age 15 and 24 living in the university’s region with the variable demo1524it . Finally, we control for the unemployment level of people aged between 15 and 24 in the same region (unemploymentit ), as pursuing studies is a good safety net against weak labor market conditions. Table 1 shows some descriptive statistics. Fig. 1 shows the evolution of the distribution of newly enrolled students of both universities that have and have not provided MOOCs. Starting from 2014, the academic year after the first MOOCs were provided on FUN, we observe from these boxplots that traditional programs from institutions that have provided MOOCs have tended to see an increase in their enrollment figures. The objective of this paper is to test if this correlation stands to more robust econometric approaches. 4. Empirical strategy Identifying the causal impact of the provision of MOOCs poses a potential problem due to self-selection. Given that the factors that encourage universities to provide MOOCs and those tending to boost student enrollment are similar, there is a risk of inflating the impact of MOOCs, hence leading to spurious results. To handle this selection-based endogeneity bias, we take advantage of the panel structure of our data and explicitly consider
6 This data was gathered from the internet archives of the Bibliothèque Nationale de France.
J. Jacqmin / Information Economics and Policy 48 (2019) 32–39
35
Fig. 1. Newly enrolled students in French universities.
Table 1 Descriptive statistics.
Dependent variables Newenrollment Explanatory variables Moocdummy Moocperyear Moocsrerun # of participants # of young participants Mediacoverage presscoverage Control variables Enrollmentlevel Uniranking Cityranking Merger Demo1524 Unemployment
N
Mean
Std. dev.
Min
Max
360
4345.93
2332.97
683
13058
360 360 360 360 360 360 360
0.13 0.20 0.09 1220.41 188.02 3.56 0.56
0.33 0.62 0.32 3979.70 644.28 12.98 2.26
0 0 0 0 0 0 0
1 5 3 33451 5209.08 162 19
360 360 360 360 360 360
20630.15 56.89 22.70 0.04 12.06 25.77
12233.2 127.18 13.26 0.19 0.86 8.68
2078 0 0 0 9.92 17.10
63447 465 42 1 13.99 57
a series of control variables. The former enables us to control for time- and place-invariant unobservables while the latter controls for time- and place-variant observables that can be both correlated with our dependent and explanatory variables. By including university fixed effects, we control for institutional characteristics that are fixed across time such as location, the types of programs offered or the scholastic tradition of the institution. Some of these institutions are ”small” universities offering either humanities or scientific disciplines, some of which are more likely to be capacity constrained. Using year fixed effects enables us to control for unobserved heterogeneity, such as changes in legislations affecting all institutions or in student preferences and the overall increase in enrollments. For this purpose, we estimate the following equation:
newenrol l mentit = α0 + α1 MOOCdumit + β Xit + γi + f (Tt ) + it
MOOCdumit , a dummy equal to 1 if the university provided MOOCs the previous year. Xit is the vector of university-level and timevarying controls. We include university fixed effects γ i .7 In addition, f(Tt ) is a function of time that includes either year fixed effects, a linear trend or university-specific trends. Our results do not appear to be impacted by how we take into consideration the temporal dimension of our data. Finally it represents the error term and α 0 is a constant. Standard errors are clustered at the university level to control for whithin institution correlations. While this approach does not rule out the presence of reverse causality, we believe that it is limited in our setting. First of all, it is unlikely that French universities choose to set up MOOCs on the basis of their new enrollment figures or even expectations in this respect. Setting-up MOOCs is time-intensive, as some time elapses between the decision to provide them and their effective availability. According to Hollands and Tirthali (2014), once a MOOC project is launched, it takes at least 6 months for the MOOC to be aired online. In addition, while students state their institutional preferences at the end of May, universities only get to know about their enrollment figures in July, with no prior credible informational signal. Second, we can rule out that reverse causality is a problem using our database. With MOOCdummy as a dependent variable and a specification that includes year and university fixed effects, we find no significant effect when our explanatory variable is the number of students enrolled in the previous year or the growth thereof.8 Hence, it does not seem that universities facing an increase in funding (due to larger enrollment numbers in the recent past) use it to provide MOOCs which would have biased our main coefficient. These arguments suggest that reverse causality does not influence our main results. This identification strategy rules out the presence of omitted variable bias from observable factors explicitly accounted for and for unobservable factors fixed across the years or universities considered. The correlation between the provision of MOOCs and unobserved variables, however, is assumed to be constant over time
(1) where newenrollmentit , the number of newly enrolled students in university i in year t, is our dependent variable. We take its log to be able to interpret our results as elasticities. However, this does not change the quality of our results. Our explanatory variable is
7 Using a random effects estimator, the quality of our results remains unchanged. Following the results of the Hausman test, the null hypothesis of no systematic difference between fixed and random effect estimates is rejected. Hence, the fixed effect estimator is more consistent. 8 All these results are available upon request.
36
J. Jacqmin / Information Economics and Policy 48 (2019) 32–39 Table 2 Main results: impact on new enrollments. Dependent variable New enrollments
(1) log
(2) log
(3) log
(4) log
(5) log
(6) level
MOOCdummy
0.053∗ ∗ ∗ (0.014) 0.831∗ ∗ ∗ (0.0297) −0.0 0 01 (0.0 0 02) 0 (0.001) 0.003 (0.027) −0.111∗ ∗ ∗ (0.016) 0.009∗ ∗ ∗ (0.002) 1.206∗ ∗ ∗ (0.358) NO NO NO NO 360 0.42
0.038∗ ∗ ∗ (0.013) 0.586∗ ∗ ∗ (0.0982) 0.0 0 04∗ (0.0 0 02) 0.002 (0.001) −0.016 (0.024) −0.267∗ ∗ ∗ (0.04) 0.005 (0.004) 5.557∗ ∗ ∗ (1.091) YES NO NO NO 360 0.502
0.027∗ ∗ (0.013) 0.363∗ ∗ ∗ (0.1059) 0.0 0 04∗ ∗ ∗ (0.0 0 01) 0.001 (0.001) −0.033 (0.025) −0.017 (0.049) 0.007∗ (0.004) −56.38∗ ∗ ∗ (12.081) YES NO YES NO 360 0.549
0.024∗ (0.014) −0.204 (0.242) 0.0 0 05 (0.0 0 04) 0.001 (0.001) 0.007 (0.026) 0.061 (0.158) 0.005 (0.007) −103.98∗ ∗ ∗ (46.298) YES NO YES YES 360 0.77
0.027∗ ∗ (0.012) 0.411∗ ∗ ∗ (0.114) 0.0 0 03∗ ∗ ∗ (0.0 0 01) 0.002∗ ∗ (0.001) −0.027 (0.023) 0.02 (0.052) 0.006 (0.004) 3.701∗ ∗ ∗ (1.145) YES YES NO NO 360 0.568
324.29∗ ∗ ∗ (86.36) 1660.7∗ ∗ ∗ (608.434) 4.441∗ ∗ ∗ (1.4903) −1.54 (4.593) 33.153 (201.19) −249.26 (463.89) 22.521∗ ∗ (10.79) −9871 (7491.6) YES YES NO NO 360 0.545
Enrollmentlevel Uniranking Cityranking merger Demo1524 Unemployment Constant University effects Year effects Linear trend University-specific trend N R2
Heteroskedasticity-consistent standard errors in parentheses are clustered at the university level. Statistical significance: ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01.
or place. In other words, within a university, we assume that the timing of the provision of MOOCs is exogenous.
5. Results 5.1. Impact on new enrollments Before addressing the reason why providing MOOCs can impact enrollment and what characterizes the students that are impacted, we first study the impact of providing MOOCs on subsequent new enrollments. We also make various robustness checks related to the definition of our explanatory variable. Table 2 shows the impact of providing MOOCs on new enrollment figures. In Regression 1, we run a simple OLS regression merely considering our control variables. We estimate that when a university decides to provide MOOCs, its new student intake increases by 5.3%, all else being equal. In Regression 2, the estimate becomes equal to 3.8% due to the introduction of university-level fixed effects. For Regressions 3–5, we consider the time dimension using 3 different approaches: a linear trend, university-specific linear trends and year fixed effects. The coefficient estimates for the MOOC variable remains stable between 2.4% and 2.7%. We can explain these decreasing estimates by the fact that, over the five-year period, both new student intake figures and the number of universities providing MOOCs have increased. Note that the significance of our results has only been marginally impacted. Hence, taking into consideration the panel structure enables us to control implicitly for time- and place-varying aspects that would otherwise have inflated our estimates. In regression 4, where we consider university-specific trends, we tested the hypothesis according to which the trends are equal and this was rejected. Hence, as it leads to the highest R2 , we use regression 5 where year fixed effects are used as our baseline specification for the rest of the paper. To give a rough idea of the size of the effect observed, we see that providing MOOCs impacts new enrollments in comparable proportion to climbing 75 ranks in the Shanghai ranking. For completeness’s sake, regression 6 shows the baseline specification using levels of new enrollments rather than log estimates. The significance levels of our variable of interest is not impacted and the universities that
have provided MOOCs on the FUN platform increased their enrollment in the subsequent academic year by 324 students on average. Table 3 presents additional results where the definition of our explanatory variable is changed, and regression 5, where we use university and year fixed effects, is our benchmark. In regression 7, we specify the field in which the MOOCs were taught. We classify them into 4 categories: health, science, humanities and law/business. We find that providing MOOCs in a scientific discipline has a positive and significant impact on new enrollments. The same holds for MOOCs in the field of humanities, however in this case the level of significance drops to the 10% threshold. However, we find that providing online courses in law or business has a negative and significant impact on enrollments. There are several potential explanations for this negative sign. First of all, universities providing law or business MOOCs are more likely to be capacity constrained. Traditional law programs have been quite impacted by capacity issues. In addition, in our sample, institutions providing MOOCs on these topics tend to be more based in Paris (42% while only 38% MOOCs provided on the platform are from universities based in Paris), an area that tends to be more congested. However, this argument alone can only explain why the sign of the coefficient is not positive. Not why it is negative, as observed in our data. A second possible explanation lies in the informational asymmetry between the students and the content of what is taught to them in traditional programs. In France, students have to choose their major early on, several months ahead of the start of the academic year, and it is complicated to change from one program to the other during the academic year. Compared with other disciplines, students have not had the opportunity to follow a law course during high school and business/economics tends to be taught very differently at the university level (as it is taught in a much more formalized/math-intensive way). Hence, following a MOOC might have discouraged them to study in these disciplines by decreasing this asymmetry of information while giving them a preview of what law/business programs are really about. A final explanation lies in the fact that only 12 out of our 360 observations have provided MOOCs in these disciplines. Hence, it can be the case that this result is driven in part by some outlier observations.
J. Jacqmin / Information Economics and Policy 48 (2019) 32–39
37
Table 3 Main results: MOOCs characteristics. Dependent variable New enrollments (log)
(7)
Health
−0.022 (0.0346) 0.0315∗ ∗ (0.0155) 0.0219∗ (0.0127) −0.0256∗ ∗ (0.0103)
Science Humanities Law/business
(8)
(9)
(10)
(11)
(12)
0.0313∗ ∗ (0.0126)
0.0105 (0.0153) 0.0 0 01∗ (0.0 0 01) 0.4093∗ ∗ ∗ (0.1138) 0.0 0 03∗ (0.0 0 01) 0.002∗ ∗ (0.0 0 08) 0.0394∗ (0.0226) 0.0352 (0.0517) 0.006 (0.0041) 3.5375∗ ∗ ∗ (1.1358) YES YES 360 0.5739
0.0134∗ ∗ (0.0062)
Moocperyear Moocsrerun
0.0125 (0.0117)
MOOCdummy(t − 1 )
−0.0139 (0.0183)
MOOCdummy Interaction with ranking Enrollmentlevel Uniranking Cityranking Merger Demo1524 Unemployment Constant University effects Year effects N R2
0.4104∗ ∗ ∗ (0.112) 0.0 0 04∗ ∗ ∗ (0.0 0 01) 0.0018∗ ∗ (0.0 0 08) −0.0135 (0.0188) 0.0228 (0.0512) 0.0063 (0.0041) 3.6659∗ ∗ ∗ (1.1318) YES YES 360 0.5738
0.4052∗ ∗ ∗ (0.1136) 0.0 0 04∗ ∗ ∗ (0.0 0 01) 0.0019∗ ∗ (0.0 0 08) −0.0249 (0.0216) 0.0222 (0.0524) 0.0061 (0.0041) 3.7293∗ ∗ ∗ (1.1408) YES YES 360 0.5684
0.4065∗ ∗ ∗ (0.1138) 0.0 0 04∗ ∗ ∗ (0.0 0 01) 0.0017∗ ∗ −(0.0 0 08) −0.0205 (0.0232) 0.0146 (0.0541) 0.0063 (0.0041) 3.806∗ ∗ ∗ (1.1536) YES YES 360 0.5633
0.4039∗ ∗ ∗ (0.1102) 0.0 0 01 (0.0 0 02) 0.0014∗ (0.0 0 08) −0.0245 (0.023) 0.0173 (0.0635) 0.0059 (0.0039) 3.8927∗ ∗ ∗ (1.2513) YES YES 288 0.5283
0.398∗ ∗ ∗ (0.1181) 0.0 0 03∗ ∗ ∗ (0.0 0 01) 0.0014 (0.0 0 08) −0.0169 (0.0192) −0.0091 (0.0518) 0.0062 (0.0042) 4.1838∗ ∗ ∗ (1.1553) YES YES 305 0.6052
Heteroskedasticity-consistent standard errors in parentheses are clustered at the university level. Statistical significance: ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01.
Instead of using a dummy variable, Regression 8 uses the number of MOOCs provided in a year by each university. As expected this approach leads to a positive and significant coefficient. Regression 9 uses a proxy of the quality of the MOOCs provided. moocsrerun is equal to one if one of the MOOC provided that year was re-aired in another course session. We assume that successful MOOCs are more likely to be re-aired. Yet, if re-airing a MOOC is economical in terms of production and organization, it remains as time-consuming in terms of supervision. We find that this does not have an impact on future enrollment in traditional programs provided by the same university. Regression 10 examines whether the advertising effect of MOOCs dies out within the following year. One consequence of this change is that it reduces the size of our sample. We observe that providing MOOCs does not seem to have an impact on new enrollments in the subsequent year. In other words, the impact of MOOCs is only contemporaneous. Finally, we look at whether the impact of MOOCs differs across universities, with respect to their location and their reputation. In regression 11, we disregard the Paris universities, these being likely oversubscribed compared to others. The impact on subsequent new enrollments then increases by 3.1% instead of 2.7% when these universities are included. In regression 12, we look at whether the impact of MOOCs differs across universities. Introducing a cross term with the ranking of the university in the Shanghai, we find that MOOCs benefit relatively more universities that higher better ranked in the ranking. Hence, universities that are more reputable and outside of Paris tend to be more impacted by the provision of MOOCs.
5.2. Explanatory factors While the aggregate nature of our data impedes an accurate observation of the mechanisms at work, we find two possible explanations for the impact of MOOC provision on student enrollment. The first one holds students’ direct influence by MOOCs, whether as active participants to a MOOC or sympathizers to the university’s pedagogical concern. Through the mediation of MOOCs the university reduces the asymmetric information gap separating the universitity from potential students. The second explanation is that students’ attention is indirectly impacted by the buzz surrounding MOOCs. Accordingly, by capturing potential students’ attention in this way the university take advantage of a behavioral bias in their enrollment decision. While disentangling these two intertwined mechanisms may be difficult, Table 4 at least presents some evidence for the coexistence of these two explanations. To test the first of these explanations, we use proprietary data gathered from FUN about the number of participants in the MOOCS provided by a university in a specific year. In regression 13, we use (log of) the number of participants in the course as an explanatory variable. We observe that having MOOCs that double their enrollment of students while lead to an increase in new enrollment in the traditional program by 0.6%. If we restrict our attention to MOOC participants ranging within the age bracket of the wide majority of prospective students (between 17 and 23 years of age), as in Regression 14, we observe that doubling the number of young participants in a MOOC increases subsequent enrollments by 0.8%. One explanation for the slightly higher coefficient
38
J. Jacqmin / Information Economics and Policy 48 (2019) 32–39
Table 4 Explanatory factors. Dependent variable New enrollments (log)
(13)
# of participants
0.003∗ ∗ (0.0013)
(14)
(15)
0.004∗ ∗ (0.0016)
# of young participants
0.0062∗ (0.0031)
Media coverage Press coverage Enrollmentlevel Uniranking Cityranking Merger Demo1524 Unemployment Constant University effects Year effects N R2
(16)
0.4064∗ ∗ ∗ (0.1137) 0.0 0 03∗ ∗ ∗ (0.0 0 01) 0.0019∗ ∗ (0.0 0 08) −0.0267 (0.0215) 0.0189 (0.0521) 0.0059 (0.0041) 3.7656∗ ∗ ∗ (1.1417) YES YES 360 0.5696
0.4065∗ ∗ ∗ (0.1137) 0.0 0 03∗ ∗ ∗ (0.0 0 01) 0.0019∗ ∗ (0-08) −0.0266 (0.0216) 0.0191 (0.0521) 0.0059 (0.0041) 3.7618∗ ∗ ∗ (1.1408) YES YES 360 0.5697
0.4087∗ ∗ ∗ (0.1142) 0.0 0 04∗ ∗ ∗ (0.0 0 01) 0.0018∗ ∗ −(0.0 0 08) −0.0246 (0.0214) 0.0191 (0.0525) 0.006 (0.0041) 3.7405∗ ∗ ∗ (1.1474) YES YES 360 0.5673
0.012∗ ∗ (0.0053) 0.407∗ ∗ ∗ (0.1139) 0.0 0 03∗ ∗ (0.0 0 01) 0.0018∗ ∗ (0.0 0 08) −0.0226 (0.022) 0.0234 (0.0521) 0.0061 (0.0041) 3.708∗ ∗ ∗ (1.1368) YES YES 360 0.5687
Heteroskedasticity-consistent standard errors in parentheses are clustered at the university level. Statistical significance: ∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01.
in this regression as compared to the previous one brings additional evidence for the direct influence of participating in a MOOC might have on the student enrollment decision, as people in this age range are more likely to enroll a university. As a test for the second explanation, we first proxy the attention-generating event provided around the MOOC with the number of search results on Google, in the year before the students’ enrollment decision. Regression 15 shows that media coverage, though marginally significant, expectedly has a positive sign. Finally, in regression 16, we count the number of times the MOOC was mentionned in the (regional and national) press. Once again, we observed a positive and significant coefficient.
5.3. Impact on student characteristics There appears to be no obvious pattern linking student characteristics, MOOC provision and university choice. On the one hand, the more able students could be expected to be more likely to participate in a MOOC and favour institutions developing innovative pedagogical approaches. On the other, the least informed students, such as those residing further away from universities might benefit more from the advertising effect of MOOCs.) To measure the heterogeneous impact of MOOCs on the student population, we use data pertaining to first-time enrollees, i.e. those who obtained their baccalauréat the summer preceding their first enrollment (Ministère de l’enseignement supérieur, de la recherche et de l’innovation, 2018).9 Hence, with this subset of our previous sample, we exclude students doing a gap year, transfer students or master students coming from other institutions. This segment still represents the vast majority of newly enrolled students, close to 85%, but only a minority of foreign ones. It also represents the
9 Note that for this data, the set of universities available goes from 72 to 66 due to the way merging institutions have been considered.
segment of the population that is the least likely to face a selective barrier at the university entrance.10 We can refine our main result by considering the type of baccalauréat obtained, students’ origin and gender. In high school, students choose between 3 tracks leading either to the general, the professional or the technological baccalauréat, as shown in Table 5. The first one targets students preparing to study at a university or a ‘grande école’. The other two respectively targets those prepare to join a professional college or directly enter the job market, although the open nature of the French admission system does not discourage them from applying to a university. The results based on the number of new students enrolled just before entering university first-time enrollees are shown in Regression 17, 18 and 19. Following the statistically not significant results pertaining to students choosing the professional and technological tracks, what appears is that students with the most adequate high school record to enter university are positively impacted by MOOC provision in their choice of university institution. Another dimension relates to student origin. One feature of MOOCs is that they are global public goods and can be a means of attracting students who would be hard to reach via other communication channels. Regressions 20 , 21, 22 and 23 show the number of first-time enrollees coming from the same département where the university is located, from a neighboring département, from another French département or from abroad. All student categories are positively and significantly impacted by MOOC provision but one: students coming from abroad. Note however that only 2% of first-time enrollees coming after obtaining a baccalauréat are from abroad while foreign students make up 14% of the total student population. Hence, this data might not be representative of the whole foreign student population. Finally, we see in regression 24 and 25 that the provision of MOOCs positively impact students enrollment from both genders. However, we have that the impact on female students is bigger than the one on male students. 6. Conclusion This paper demonstrates that providing MOOCs enables universities to increase enrollment in their traditional programs. Using enrollment data from French universities across 5 years helps identify this relationship by considering year- and university-specific unobservables. In a context where students cannot be selected at university entrance and tuition fees are homogeneous, we observe that MOOC provision implies a 2% increase of first-time enrollees in the following year. This increase is equivalent to climbing 75 ranks in the Shanghai ranking. We find additional evidence for MOOCs’ positive impact on the enrollment decision of students following the high school track leading to university. Hence, from the university perspective, we conclude that courses provided on MOOC platforms complement rather substitute traditional university programs. Using data about MOOC enrollment and media coverage, we find some evidences supporting two potential theories behind this conclusion. MOOCs can be seen as a means of marketing universities by improving students’ information about them and capturing their attention at their enrollment decision stage. To get a closer understanding of the functioning of these two explanations, more disaggregated data would be needed, e.g. with data matching students who enrolled in MOOCs against those enrolling at universities. Various courses have recently been hosted on the FUN 10 As taking the log of these new dependent variable decreases the (between and within) variation, we focus on the level of newly enrolled students in each categories. Not doing so leads to less statistically significant results. Hence these results have to be taken with more caution. Results available upon request.
J. Jacqmin / Information Economics and Policy 48 (2019) 32–39
39
Table 5 Student characteristics.
New enrollments moocdummy Control variables University effects Year effects N R2
(17) bacG
(18) bacTEC
(19) bacPRO
(20) dept.
(21) neigh. dept.
(22) other dept.
(23) abroad
(24) female
(25) male
298.08∗ ∗ ∗ (77.52) Yes Yes Yes 330 0.553
1.4 (15.78) Yes Yes Yes 330 0.297
23.44 (18.92) Yes Yes Yes 330 0.44
134.83∗ ∗ ∗ (50.04) Yes Yes Yes 330 0.488
88.45∗ ∗ ∗ (24.58) Yes Yes Yes 330 0.581
70.08∗ ∗ ∗ (24.25) Yes Yes Yes 330 0.32
3.67 (5.46) Yes Yes Yes 330 0.115
169.99∗ ∗ ∗ (39.26) Yes Yes Yes 330 0.437
136.88∗ ∗ ∗ (50.58) Yes Yes Yes 330 0.37
Heteroskedasticity-consistent standard errors in parentheses are clustered at the university level. Statistical significance: ∗∗ p < 0.05, ∗∗∗ p < 0.01.
platform with the view of smoothing the transition between high school and higher education. Hence, from the universities’ point of view, improving our understanding of the impact of MOOCs is key to enhance the complementarity between online and live courses, not only at the enrollment decision stage. Another question that would require more precise data concerns whether providing MOOCs expands the higher education towards new students or steals students from other higher education institutions. This paper considers MOOC provision from a marketing point of view. Future research, though, will benefit also from further exploring MOOCs’ potential to implement new pedagogical approaches or to curb the costs of traditional higher education programs. Supplementary material Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.infoecopol.2018.10. 002. References Alpert, W.T., Couch, K.A., Harmon, O.R., 2016. A randomized assessment of online learning. Am. Econ. Rev. 106 (5), p.378–382. Beine, M., Noel, R., Ragot, L., 2014. Determinants of the international mobility of students. Econ. Educ. Rev. 41, 40–54. Belleflamme, P., Jacqmin, J., 2016. An economic appraisal of mooc platforms: business models and impacts on higher education. CESifo Econ. Stud. 148–169. Bettinger, E., Fox, L., Loed, S., Taylor, E.S., 2017. Virtual classrooms: how online college courses affect student success. Am. Econ. Rev. 107 (9), p.2855–2875. Bettinger, E., Terry Long, B., Oeropoulos, P., Sanbonmatsu, L., 2012. The role of application assistance and information in college decision: results from the h&r block fafsa experiment. Q. J. Econ. 127 (3), 1205–1242. Bowen, W., Chingos, M., Lack, K., Nygren, T., 2014. Interactive learning online at the public universities: evidence from a six-campus randomized trial. J. Policy Anal. Manage. 33 (1), 94–111. Broecke, S., 2012. University rankings: do they matter in the UK? Educ. Econ. 23 (2), 137–161. Chevalier, A., Jia, X., 2016. Subject-specific league tables and students’ application decisions. Manchester School 84 (5), p.600–620.
∗
p < 0.1,
Cho, D., Smith, M.D., Zentner, A., 2016. Internet adoption and the survival of print newspapers: a country-level examination. Inf. Econ. Policy 37, 13–19. Deming, D., Goldin, C., Katz, L., Yuchtman, N., 2015. Can online learning bend the higher education cost curve? Am. Econ. Rev. 105 (5), 496–501. Deslauriers, L., Schelew, E., Wieman, C., 2011. Improved learning in a wide-enrollment physics class. Science 332, 862–864. Dupont-Calbo, J., 2013. Derrière le mooc à la française: Google. Le monde. Figlio, D., Rush, M., Yin, L., 2013. Is it live or is it internet? Experimental estimates of the effects of online instruction on student learning. J. Labor Econ. 31 (4), p.763–784. Goodman, J., Melkers, J., Pallais, A., 2019. Can online delivery increase access to education? J. Labor Econ. 37 (1). Hollands, F.M., Tirthali, D., 2014. MOOCs: Expectations and Reality. Technical Report. Teachers College, Columbia University. Howarth, J., D’Alessandro, S., Johnson, L., White, L., 2017. Moocs to university: a consumer goal and marketing perspective. J. Market. Higher Educ. 27 (1), p.144–158. Hurwitz, M., Smith, J., 2018. Student responsiveness to earnings data in the college scorecard. Econ. Inq. Luca, M., Rooney, P., Smith, J., 2016. The Impact of Campus Scandals on College Applications. Harvard Business School NOM Unit Working Paper 137. Luca, M., Smith, J., 2013. Salience in quality disclosure: evidence from the us news college rankings. J. Econ. Manage. Strategy 22 (1), 58–77. Ministère de l’enseignement supérieurde la recherche et de l’innovation, 2016a. Apb 2016: propositions d’admission et réponse des candidats pour l’année scolaire 2016–2017. Note Flash 17. Ministère de l’enseignement supérieurde la recherche et de l’innovation, 2016b. Etat de l’enseignement Supérieur et de la Recherche en France. Technical Report N. 9. MESNER. Ministère de l’enseignement supérieur, de la recherche et de l’innovation, 2018. data.esr.gouv.fr. Pistolesi, N., 2017. Advising students on their field of study: evidence from french university reform. Labour Econ. 44, 106–121. Pope, D., Pope, J., 2014. Understanding college application decisions: why college sports success matters. J. Sports Econ. 15 (2), 107–131. Smith, M.D., Telang, R., 2010. Piracy or promotion? The impact of broadband internet penetration on dvd sales. Inf. Econ. Policy 22, 289–298. The Economist, 2012. Learning New Lessons. Xu, D., Jaggars, S., 2013. The impact of online learning on students’ course outcomes: evidence from a large community and technical college system. Econ Educ Rev 37, p.46–57. Zentner, A., 2012. Internet adoption and advertising expenditures on traditional media: an empirical analysis using a panel of countries. J. Econ. Manage. Strategy 21 (4), 913–926.