Paper 4aNaEzZ3 PDF
Paper 4aNaEzZ3 PDF
Paper 4aNaEzZ3 PDF
December 2018
Abstract: We evaluate two low-cost college support programs designed to target insufficient study time, a
common problem among many undergraduates. We experimentally evaluate the programs across three
distinct colleges, randomly assigning more than 9,000 students to construct a weekly schedule in an online
planning module and to receive weekly study reminders or coach consultation via text message. Despite
high participation and engagement, we estimate precise null effects on student credit accumulation, course
grades, and retention at each site for the full sample and for multiple sub-groups. The results suggest that
students are simply not responsive to low-cost scheduling assistance, encouragement, or reminders for
studying. Possible explanations for this unresponsiveness are discussed.
Acknowledgements: At the University of Toronto, we are indebted to the first-year economics instructors for their
willingness to incorporate an experiment into their courses for a fourth consecutive year. We especially thank James
Lahey, our web developer, for his tireless commitment to designing and perfecting the experiment’s website, as well
as for his help with organizing and extracting the experimental data. Spencer Dean, Chelsea Kowalski, Catherine
Tessier, Chester Madrazo, Ophelia Au, and Erica Rzepecki showed great enthusiasm and professionalism in their role
as coaches. At WGU, we thank Jason Levin, Chelsea Barnett, and Narendra Pandya for their help designing,
programming, and executing this study. Thanks also to seminar participants at the University of Michigan, Michigan
State University, UC San Diego, Notre Dame University, the Institute of Behavior & Inequality, MDRC, the
Association for Public Policy Analysis and Management (APPAM) 2017 annual research conference, the Society of
Labor Economists’ 2018 annual meetings, and the College Board. All remaining errors are our own. The experiments
in this paper were both registered with the AEA RCT Registry. The RCT IDs are AEARCTR-0000972 and
AEARCTR-0000810. Oreopoulos – University of Toronto and NBER, philip.oreopoulos@utoronto.ca; Patterson –
United States Military Academy, richard.patterson@usma.edu; Petronijevic – York University, upetroni@yorku.ca;
Pope – University of Maryland, pope@econ.umd.edu.
1. Introduction
College enrollment has steadily increased in recent decades, as policymakers, popular media, and
parents all emphasize the importance of post-secondary education for labor market success. Yet
proportional increases in completion rates have not followed suit and many students who do
complete their degrees struggle and develop limited skills along the way (Arum and Roksa 2011).
Student effort is a key determinant of academic outcomes, and many students devote an alarmingly
low amount of time to regular studying (Babcock and Marks 2011). Despite a clear positive
association between study time and academic outcomes (Brint and Cantwell 2010; Stinebrickner
and Stinebrickner 2004, 2008), underachieving students in both traditional and online colleges
often manage their time poorly and study very little (Dohetry 2006; Beattie, Laliberté, Michaud-
Leclerc, and Oreopoulos 2017; Beattie, Laliberté, and Oreopoulos 2018). Indeed, any initiative to
improve student outcomes is likely to be far more effective when students are consistently engaged
four-year college, a less-selective four-year college, and an online university – that differ in terms
of student characteristics and how they deliver education, but share the problem of having many
students who manage their time poorly. Recent studies using students from our traditional college
settings (campuses at the University of Toronto) show that struggling students have a high
propensity to procrastinate and study little throughout the academic year (Beattie, Laliberte, and
Oreopoulos 2018). Upon initially experiencing poor performance in first semester, these students
do not increase their planned study time, despite many acknowledging that the biggest challenge
to their academic success is poor time management (Beattie, Laliberte, Michaud-Leclerc, and
Oreopoulos 2017).
1
A lack of available time is unlikely to be the reason for low study times. Figure 1 plots the
distribution of time that students have available and the distribution of time they self-report
studying during a regular week in the fall semester, using information from baseline and follow-
up surveys among students from the two colleges in our traditional college samples. 1 Available
1F
study time is calculated from a baseline survey that elicits students’ self-reported expectations for
weekly hours of work (for pay), commuting time to and from campus each week, time spent
attending lectures each week, and time spent sleeping. Self-reported study time during a regular
week in the fall semester is gathered from a follow-up survey at the end of the semester. The two
distributions are almost non-overlapping, suggesting that many students are nowhere close to
pushing up against their available time constraints when studying. Indeed, the median student has
approximately 93 hours a week available but only chooses to devote 12 hours to studying outside
the classroom. Half of students in our sample therefore self report studying less than 12 hours a
week, while the bottom quarter of students report studying less than 5 hours per week.
At the online college we study, Western Governors University (WGU), students also
appear to study infrequently. Although students have easy access to material online, the average
student logs into to their portal only 2.1 days per week. In addition, 90 percent of students log in
less than 3.7 days per week and 18.5 percent of students log in less than 1 day per week. More
generally, online education is a setting where time management issues are particularly likely to
drive poor performance. Indeed, recent experimental and quasi-experimental evidence finds that
students in online courses perform worse than students in traditional classroom settings (Bettinger
et al., forthcoming; Figlio et al. 2013). One possible reason for these performance gaps is that the
1
We describe the data used to construct this figure in greater detail in Section 4.
2
asynchronous, unstructured nature of online courses makes students particularly prone to issues
To address issues of poor time management and low study times, we design and evaluate
two programs that aim to increase study time by helping students create and follow realistic
undergraduate students at the University of Toronto (UofT) and a sample of over 6,000
undergraduate students at WGU. At UofT, we implement our program at both the more-selective
main campus, St. George (UTSG), and the less-selective suburban commuter campus, the
2,000 UTSG students and 1,500 UTM students. Across all three locations, we randomly assign
incoming students to treatment or control groups. Students in the treatment group are provided
information to motivate the benefits of sufficient study and complete an online planning module
in which they make a calendar describing their planned weekly commitments in the upcoming
year, including the times during the week they plan to study. To keep these plans salient, we also
encouraged students at the UofT campuses to provide their phone numbers and students at WGU
to download the WGU mobile application so that we could send students reminders about their
scheduled study times via text message throughout the academic year. Students in the control
groups at the UofT campuses were given a personality test, while students in the control group at
WGU did not receive the planning module but still completed the standard online student
orientation.
Our planning interventions relate to a broader and growing literature on the application of
behavioral insights to education settings (Lavecchia, Liu, and Oreopoulos 2016; Damgaard and
Nielsen 2018). Recent attempts to help improve academic outcomes focus on prompting students
3
to think about future goals (Clark et al. 2017; Dobronyi et al. 2017), encouraging more healthy
perspectives for dealing with setbacks or anxiety (Yeager et al. 2016; Bettinger et al. 2018), and
low-cost encouragement or advising (Fryer 2016; Castleman and Meyer 2016; Oreopoulos and
Petronijevic 2018). We focus instead on targeting study time, treating it as a crucial necessary
(but not sufficient) condition for academic success. Poor performing students typically studying
less than 10 hours a week are unlikely to benefit from any intervention that does not increase this
variable.
Our planning interventions are designed to improve study time management through three
key channels. First, by providing information about successful students’ study habits through an
online module, they make students aware of how much time is usually required to perform well in
their courses. Second, by requiring that students create a weekly plan that details all their
commitments, the interventions help students better understand the time commitment required for
all their other obligations outside of school. Third, the periodic reminders that students receive
about their planned study times help keep their goals salient throughout the academic year.
Despite our time-management program being well-received and generating a high degree
of student engagement, we find no impact on academic outcomes across all three experimental
sites (the two campuses of UofT and WGU). Specifically, we estimate no treatment effect on
credit accumulation or course grades at UTSG and UTM and no treatment effect on student credit
accumulation or retention at WGU. These results hold even after investigating potentially
2
The experiments at WGU and at UofT were both pre-registered with the AEA RCT Registry. The RCT IDs are
AEARCTR-0000972 and AEARCTR-0000810 at WGU and UofT, respectively. Our analysis of treatments effects in
the full sample and across student subgroups closely follows our pre-registered analysis plans.
4
Considering the mechanisms behind the null effects, we show that the intervention did not
change objective measures of study time (such as frequency of log-ins and web activity) at WGU,
nor did it change the amount of time students at UofT studied during midterm and exam periods.
We do find that the intervention increased self-reported weekly study time by approximately 1.65
hours among UofT students. After a deeper investigation, however, we conclude that treated
students at UofT were likely primed to self-report slightly inflated values for study time during the
average week relative to the control group, implying that treatment may have also been ineffective
Although we cannot definitively identify the barrier(s) that prevented our intervention from
improving outcomes, it is unlikely that a lack of salience around students’ goals can explain our
results because treated students received weekly text messages reminding them of these goals. We
also rule out binding time constraints as the mechanism driving our null effects, showing that a
vast majority of students do have the required slack in their weekly schedules to increase study
time. Furthermore, we estimate similar treatment effects among students with high and low
preferences are unlikely to be driving our results. Instead, the null results are consistent with
students finding effort investments too costly or facing ambiguity about either the returns to
Our results are also consistent with an increasingly common finding in the economics
literature on low-cost, scalable interventions in education – namely, that such interventions are
effective at nudging students toward taking relatively simple, one-time actions but are less
effective at causing improvement in outcomes that require meaningful and sustained changes in
student behavior. For example, text-messaging programs that push helpful information to students
5
have proven effective at causing students to enroll in college once admitted (Castleman and Page
2015) or to renew financial aid (Castleman and Page 2016), but such programs have largely been
unable to affect students’ academic outcomes, such as course grades or overall GPA (Fryer 2016;
Castleman and Meyer 2016; Oreopoulos and Petronijevic 2018). Similarly, encouraging people to
make a concrete plan for action has shown promise in settings with a single action such as voting
(Nickerson and Rogers, 2010) and getting the flu vaccine (Milkman et al., 2011), but has not shown
to be effective in increasing sustained actions, such as attending the gym (Carrera et al., 2018).
We find that helping students create clear schedules and providing them with periodic reminders
has no detectable effect on student grades or credit accumulation, an important finding for higher-
education policymakers considering the efficacy of programs that emphasize to students the
Because our planning module requires students to set study-time goals, our results also
contribute to the recent literature on goal-setting interventions. In a large field experiment among
traditional college students, Clark et al. (2017) find that requiring students to make goals over
course outcomes has a small positive effect on performance but that assigning task-based goals
has a larger and more robust impact on course performance. The authors interpret their results as
evidence that task-specific goals generate positive outcomes by addressing both present-bias and
setting, Dobronyi et al. (2017) find that asking students to make specific, meaningful, and
attainable goals did not have any measurable impact on student outcomes. The authors also found
that providing students with additional growth-mindset training (Dweck, 2006) and text/email
6
reminders did not improve academic outcomes. 3 Our results suggest that helping students set
2F
specific goals for study hours may not be an effective way to improve academic outcomes.4
The remainder of this paper is organized as follows: The next section offers a brief conceptual
framework for thinking about why students encounter challenges with managing their study time
well. Section 3 provides a detailed description of our intervention and its implementation at the
UofT campuses and WGU. Section 4 describes the experimental data from both experiments and
outlines our empirical strategy for estimating the treatment effects, while Section 5 presents the
results. We discus and interpret our results in Section 6 and Section 7 provides concluding remarks.
2. Conceptual Framework
In this section, we outline the mechanisms through which our planning interventions could
improve student outcomes, followed by a brief discussion of the potential obstacles to their
effectiveness.
Many individuals tend to underestimate the time required to complete a task (Kahneman and
Tversky 1979), with more complicated tasks, such as navigating university courses, usually
resulting in greater underestimation (Buehler et al. 1994). Decomposing a task into smaller
segments, however, helps individuals form more accurate estimates about the time required to
3
Dobronyi et al. (2017) base their work on research conducted by Morisano et al. (2010) and Schippers et al. (2015),
which both find large positive effects of goal setting interventions.
4
Our paper also relates to the recent literature on pre-commitment devices in higher education. Himmler et al. (2017)
find that asking students to pre-commit to taking exams at a certain time improved their overall performance in a
graduate business school, while Baker et al. (2016) find that prompting students via email to schedule times to watch
lecture videos in a Massive Open Online Course had no impact on performance. In a similar setting, Patterson (2015)
finds that enabling students to pre-commit to limits on distracting internet time increased completion rates and
improved performance but reminding students about their coursework and allowing students to block distractions
while working had no significant impact on course outcomes.
7
complete it (Buehler et al. 1994; Forsyth and Burt 2008). Accordingly, to help students better
appreciate the time they require and have available for the tasks in their courses, our planning
interventions guide them through unpacking their weekly study schedules into smaller study
We also remind students of their study goals and weekly completion benchmarks via text
message throughout the academic year. The use of follow-up reminders is motivated in part by
economic models of limited memory and inattention (Mullainathan 2002; Ericson 2014; Karlan et
al. 2010), which predict that individuals are susceptible to inattention to their prior plans, thereby
causing delays or even failures in plan completion. Reminders have been shown to successfully
increase plan completion in a variety of domains, including exercise (Calzolari and Nardotto
2012), repayment of loans (Cadena and Schoar 2011), savings accounts deposits (Karlan et al.
Our planning interventions are also designed to help students better manage their time by
increasing ‘implementation intentions,’ a term that refers to the process of identifying when,
where, and how one will fulfil a plan (Gollwitzer 1993). Recent experimental evidence suggests
that fostering implementation intentions can increase desired behavior across many domains,
including exercise, diet, recycling, project completion, and voting (Gollwitzer and Sheeran 2006;
Nickerson and Rogers 2010). By requiring students to define implementation intentions at the
beginning of the academic year, our planning intervention helps them establish clear study goals
As we describe in greater detail below, treated students at the UofT campuses were also
assigned to a senior-undergraduate student coach, whose job was to check-in once a week via text
message to inquire about how students were progressing and offer encouragement. Personal
8
coaching or advising done over the phone or in person has proven effective in improving students’
academic outcomes at both two-year and four-year colleges (Scrivener and Weiss 2013; Bettinger
and Baker 2014; Oreopoulos and Petronijevic 2018). Despite there being less evidence on the
effectiveness of personal coaching that occurs via text message,5 we offered treated students a text-
message coaching program to help them address any individual-specific challenges to following
There are, however, several reasons our planning interventions may be ineffective at increasing
study time and improving students’ academic outcomes. First, students’ tendency to procrastinate
may limit the efficacy of a planning intervention. Specifically, students may exhibit time-
inconsistency and behave more impatiently in the moment than they had previously planned
(Laibson, 1997; O’Donoghue and Rabin, 1999). Solomon and Rothblum (1984) find that more
preferences keep students from following through on their plans, then improving the quality of
students’ plans and reminding them about their plans may not be enough to affect academic
outcomes.
Second, students may be overconfident in their abilities. Twenge et al. (2012) find that a
majority of college students believe they have above average abilities. If students overestimate
their abilities, they may make insufficient study plans and underestimate the penalties they will
face from failing to follow-through on their plans (Dunlosky and Rawson, 2012).
5
Oreopoulos, Petronijevic, Logel, and Beattie (2018) show that while personal coaching via text message did not
improve academic outcomes in a sample of students at UofT, it did significantly and positively impact non-academic
outcomes, such as student mental health and feelings of belonging at the university.
9
Third and finally, the benefits to studying are long-term, uncertain, and highly ambiguous. It
is therefore also possible that students have low expectations on the returns to studying on grades,
or the returns to grades on longer term outcomes. Perhaps obtaining a degree, for example, matters
much more than obtaining a ‘B’ average instead of a ‘C’. Or perhaps these expectations are not
correct. Motivating the benefits to studying or reducing ambiguity, may therefore increase study
3. Description of Intervention
In this section, we describe the implementation of the experiments at both UofT and WGU,
providing greater detail about the planning interventions and the follow-up messages students
received.
We conducted our experiment at UofT throughout the 2017-18 academic year. At both the main
campus, UTSG, and the satellite campus, UTM, we partnered with all first-year economics
instructors to include a ‘warm-up’ exercise at the beginning of the course worth 2 percent of
students’ final grades. The exercise had to be completed within the first two weeks of the fall
semester for students to receive course credit, with the type of exercise each student completed
depending on whether he or she was randomly assigned to the treatment or control group. All
students logged in using their university accounts and completed a brief introductory survey, in
which they provided information about their parental education, their own expected educational
attainment, their work plans, their educational history, and their self-reported tendency to
procrastinate or become distracted. Students assigned to the treatment group were then required to
10
complete an online module that first taught them about the importance of sufficient study time and
then guided them through a planning intervention, while students assigned to the control group
were given a personality test. Below we describe the treatment and control modules in more detail.
All students in the planning intervention at UofT completed a three-part online module. We offer
an overview of the module in this subsection and provide full documentation in Appendix B.
During the first part of the planning module, we presented the college’s recommendation
for weekly study time (at least 4 to 6 hours per course, or at least 20 to 30 hours per week for a full
course load) and information on the importance of sufficient study time for academic performance
and general life satisfaction. We motivated the latter by showing descriptive evidence (gathered
from previous experiments we ran at UofT) about the positive associations between study time
and grades and study time and measures of mental health. In the second part of the module, we
asked students to read testimonials from former students, each of which described a common
challenge faced by university students and how making a schedule and studying regularly can help
students avoid these pitfalls. After reading through the stories, students wrote about how they
could motivate themselves to stick to a regular study routine and identified the study strategies
they thought would be the most helpful for doing so. Students were encouraged to slow down or
write a little more if they tried to continue through the exercise below a minimum time or word-
count restriction.
Having discussed the importance of keeping an organized schedule and studying enough,
the third part of the online module asked students to make their own weekly schedule by building
11
a weekly calendar. We first asked students to populate their calendars with class times, which they
could do by downloading a standard electronic calendar (ICS) file from their university platform
and then uploading the ICS file to our platform. Students then scheduled their anticipated job
schedules along with their regular sleep routines. Once they had accounted for items with little
scheduling flexibility, students were asked to populate their calendars with weekly study times.
The module asked them to reconsider the importance of sufficient and regular study time and
would not allow them to proceed unless the number of scheduled study hours throughout the week
matched their self-imposed target for study hours. As the final step toward completing their
calendars, students scheduled personal time for seeing friends and family and engaging in other
To help students stay on track throughout the academic year, we made their weekly
calendars available to them. If students already had a Gmail account, they simply had to provide
their Gmail address and we then uploaded their calendars directly into their Google calendars. If
students did not have a Gmail account, we gave them the option to create one or to simply
download their calendar as an ICS file and upload it to whichever calendar application they prefer
to use. 6
4F
For the last step of the exercise, all students were encouraged to enroll in a virtual coaching
program called You@UofT. 7 We explained that students would be matched with an experienced,
5F
senior-undergraduate coach whose job would be to check-in once a week via text message to
inquire about how students were doing with their study goals, offer support and encouragement,
6
A total of 1,685 students completed the planning intervention at UofT and 1,424 (84.5 percent) provided a Gmail
address for us to upload their calendars directly into their Google calendars. The remaining students downloaded their
calendar from our platform as ICS file.
7
As in Oreopoulos and Petronijevic (2018), we chose the name to emphasize that the program would help coach
students toward their individual definitions of success.
12
and answer any questions. Across both campuses, 80 percent of students opted-in to the coaching
Our coaches were hired through a research opportunity program, which allows students to
participate in a research project for course credit (rather than pay). Coaches were solicited to apply
for the program through various student service offices and we sought recommendations for keen,
talented senior undergraduates who had prior experience helping new students (as, for example,
residence dons, orientation volunteers, or tutors). Upon joining the team, coaches reported to our
program manager, a graduate student in economics, who communicated best practices and ensured
Once students opted-in to the coaching program, they were assigned to a specific coach
and each coach was assigned a few time slots during the week to be the coach who was on call.
During each on-call time for a given coach, we sent a batch message to all students who were
assigned to that coach to spur productive conversation. If students replied while their coach was
still on call, that coach would continue the conversation. If students replied after their coach’s shift
ended, the coach who was currently on call or the team manager was responsible for closing the
conversation.
The batch messages we sent to students fell into two general categories. The first message
type consisted of a weekly study tip on how to use study time effectively. When sending these
messages, we took advantage of knowing when students planned to study from the calendars they
completed, sending the messages 15 minutes prior to one randomly selected study session. The
second type of message was a weekly check-in from the students’ coaches, which was designed
mostly to offer support and inquire about how well students were managing their time. To help
13
encouragement if the student did not respond to the original check-in message. A list of example
check-in text messages that we sent throughout the academic year is available in Table C1 in
Appendix C.
Student engagement with the text-messaging program was quite high, with 26 to 66 percent
of treated students responding to our messages each week. In terms of cumulative engagement, 80
percent of treated students sent at least one text message back to their coach during the academic
year. We also asked students via text message for feedback on our coaching program, and many
expressed gratitude and appreciation for the study tips and support.8
As in Oreopoulos and Petronijevic (2018), students who were assigned to the control group at both
UTSG and UTM were given a personality test measuring the Big Five personality traits of
The test tended to take about 45 to 60 minutes to complete, and students were emailed a report
describing their scores on each trait upon completion of the exercise. Beattie, Laliberté, and
Oreopoulos (2018) describe the personality test in greater detail in the appendix of their paper and
8
An anonymized list of student response to our feedback request and more detailed information on student engagement
with the text-messaging program are available upon request.
14
In this subsection, we provide an overview of the planning module students completed at WGU.
WGU is a large non-profit online college in the United States. 9 Prior to the beginning of
6F
his or her first semester, each new student participates in an online student orientation. As part of
our experiment, randomly-selected undergraduate students who enrolled between January 2 and
March 1 of 2017 were additionally required to complete a two-part planning module at the end of
the online orientation. 10 The planning module was similar to that which was completed by students
7F
at UofT.
In the first part, we again shared the college's recommendation for weekly study time (1-2
hours per “competency unit”/credit or 3-6 hours per typical course)11 and required students to
complete an interactive weekly planning activity, in which they allocated their time among four
categories (work, study, recreation, family and home) and 21 subcategories. 12 Upon completion 8F
of the planning exercise, the second part of the module asked students to organize the college-
assigned Google calendar associated with their WGU email account. This calendar was pre-
populated with categorical events from each of the four primary activity types and students were
required to organize the calendar to match their planning activity allocation. When students
finished organizing their calendars, they submitted a screenshot of their completed calendar as an
enrollment requirement.
9
See Appendix A for a broad overview of WGU.
10
Students in the control group only completed the regular online orientation.
11
Among students taking 5 courses, this recommendation amounts to 15 to 30 hours per week of total study time,
which is very similar to the recommendation at UofT of 20 to 30 hours.
12
Work- working, commuting, and other work time; study--mentor support, course readings, course writing, group
activity, and other study time; recreation-watching tv, socializing, reading, exercise and sports, browsing the internet,
and other recreation; and family and home-caring for family, preparing and eating meals, cleaning and laundry,
household management, lawn and garden, sleep, and other home and family.
15
With each student having a completed calendar in hand, WGU recreated study events onto
treated students’ calendars each week for the remainder of the semester. Students were able to
modify their study schedules at any time, with study events being visible on students’ Google
calendars, the calendar in the WGU student web portal, and the WGU mobile application. The
81.8 percent of students who installed the WGU mobile application also received mobile
notifications 15 minutes prior to two randomly selected study sessions between 9am and 8pm each
week. Additionally, all treated students received study notifications in the WGU web portal
“notification center.”
To help students unpack their semester schedules, we also populated their calendar with
“completion benchmarks.” WGU students digitally meet with a counselor to set their course
schedules prior to the beginning of each semester. In this meeting, they outline the anticipated
start and end dates for each course. Nearly all courses at WGU have a “Course of Study Guide”
or syllabus that divides the course into 4-8 segments or blocks. We combined students’ anticipated
start and end dates with their course syllabi segments to create evenly spaced intermediate
completion benchmarks for each course in which a student is enrolled. These benchmarks were
populated in students’ WGU Google calendars and automatically adjusted to any changes made
by WGU or the students to the scheduled start or end date in WGU's system. Students could view
these benchmarks in the Google calendar, WGU web portal, and WGU mobile app, and students
with the mobile application received a reminder at 4pm two days before each completion
benchmark, reminding them that they would need to complete their benchmark task in the next
16
two days to stay on track. 13 Examples of the benchmark reminders can be found Table C2 in
9F
Appendix C.
In this section, we describe the data we collected from UTSG, UTM, and WGU, along with our
We begin the description of the data at UofT with Table 1, which reports the total number of
students in the treatment and control groups, as well as the fractions of students sorted to treatment
and control at each campus. Prior to the experiment, we intended to sort one third of students to
both the treatment and control groups at UTSG 14 and to evenly divide students across treatment
10F
and control groups at UTM. Table 1 shows that slightly more than one third of students (35.8
percent) were sorted to the treatment group and slightly less than one third (30.4 percent) were
sorted to the control group at UTSG, while we reached our target fractions at UTM, as the
percentages of students sorted to treatment and control are not statistically different than 50
percent. Across both campuses, we have 3,581 students in our study, with 2,044 coming from
UTSG and 1,537 coming from UTM. The completion rates for the online modules are very high
across both campuses, ranging between 97 and 98 percent. We can match 94 percent of our
13
These completion benchmark notifications were also displayed on all WGU student's web portals.
14
The remaining one third of students was sorted to a different treatment group, which is the subject of a separate,
standalone paper.
17
experimental sample to the university’s administrative data on course grades, leaving us with an
Tables 2 and 3 present balancing tests UTSG and UTM, respectively, showing that the
treatment and control group are balanced along observable characteristics. The lone exception (out
of 30 tests for mean differences) is that students in the treatment group at UTM report being
slightly more likely to often think about their futures. We demonstrate below that our treatment
effect estimates are robust to controlling for this variable and many other covariates.
In terms of the sample characteristics, approximately half our sample at the UofT campuses
is male, the average student is 18 years old, approximately 40 percent of students speak English
as their mother tongue, 50 percent of students are international, and approximately 75 percent are
in their first year of studies. These characteristics are similar across UTSG and UTM. Differences
start to emerge, however, when one considers variables related to academic preparedness. The
average incoming high school grade average at UTSG is 91 percent, while it is 85 percent at UTM,
reflecting the differences in selectivity across the two campuses. The 75th percentile student at
UTM has a high school grade average of 88 percent, which corresponds to the 25th percentile
student at UTSG. Also consistent with differences in selection criteria, many students at both
campuses intend to earn at least an A- grade average and more than a bachelor’s degree, but the
factions are higher at UTSG (74 percent and 48 percent) than at UTM (62 percent and 40 percent).
It is also the case that only 23 percent of the UTSG sample consists of first-generation students,
15
The university’s grades data only include students who are registered at the end of September in the fall semester
of 2017, which is why we are unable to match a small fraction of students who are no longer registered at that time.
The match rate to the grades data is not differential by treatment status.
18
With respect to student time commitments, the average student at UTSG expects to work
approximately 6.4 hours a week for pay and spends approximately 24 minutes commuting to
campus (in one direction). At UTM, students expect to work 8.2 hours for pay and spend 31
minutes commuting to campus. On average, students at UTSG and UTM report spending 13.6 and
11.8 hours per week, respectively, studying outside of class in high school. In subsection 4.3, we
provide descriptive evidence on student study times during the fall semester at UTSG and UTM,
along with the associations between study time and academic performance.
At WGU, our study sample includes 6,065 undergraduate students who enrolled between January
2 and March 1 of 2017. Students were randomly assigned to either the treatment or control group
based on the last two digits of their sequentially assigned student number. Table 4 shows the
balance of observable characteristics across treatment and control, indicating that the groups are
the table, four are statistically different across treatment and control groups. Students in the
treatment group are approximately half a year older, 2 percentage points more likely to work full
time, 2 percentage points more likely to have annual incomes between $45,000 and $65,000, and
2 percentage points more likely to be first-generation students. While there are more statistically
significant differences than one would expect from random assignment, we are able to verify that
the treatment assignment mechanism was followed in over 99.9% cases.16 Furthermore, these
differences are not economically large, and we show below that controlling for these variables
(and many other covariates) does not affect our estimated treatment effects. Finally, our
16
Based on the last two digits of student’s id numbers, only 4/6065 are assigned to a treatment group that does not
correspond to the assignment rule. Our estimates are robust to exclusion of these observations.
19
experimental design also involved randomly assigning graduate students to the planning
treatment.17 While our analysis plan specified that these graduate students be dropped from our
analysis, we show in Appendix D that our sample balances across observable characteristics (1/16
variables differ at the 5% level) when graduate students are included and that our results remain
unchanged.
our study are male and the average student is 35 years old – a marked difference from the UofT
sample, where half of the sample is male, and the average student is only 18 years old. Nearly 80
percent of the sample consists of white students, while Hispanic and black students each comprise
approximately 11 percent of the sample. A large majority (75 percent) of students are employed
full time and many (40 percent) have annual incomes of $65,000 or more.18 Approximately 42
percent are first-generation students whose parents did not complete post-secondary education.
Figure 1 (discussed above in the Introduction) shows that many students at UofT study far less
than the time they have available to do so, with the median student reporting that they studied only
12 hours per week in the fall semester despite having more than 90 hours available. As mentioned,
we construct time available in Figure 1 using the information students provide in our baseline
survey about their expectations for upcoming weekly hours of work (for pay), commuting time to
and from campus each week, time spent attending lectures each week, and time spent sleeping. 19
17
Graduate students assigned to the treatment were not sent benchmark reminders in all courses, but were otherwise
treated identically to undergraduate students.
18
One may be concerned that students who work full time do not have available time to increase study intensity. In
our analysis of heterogeneous treatment effects below, we show that our estimates do not differ across WGU students
by employment status or household income.
19
We acknowledge that there are other demands on students’ time that are not captured by these variables, such as
eating, sports and clubs, self-care, church going, etc. To make sure that we are not drastically overstating the time
20
We gathered information on actual (self-reported) study time during the fall semester by
conducting a follow-up survey with students at the end of the fall semester, asking how many hours
they spend studying outside of class during an average week (which is the reported study time
variable in Figure 1) and how many hours they spend studying during a week in which they are
preparing for midterms or exams.20 Because the follow-up survey did not have grade incentives
attached, the aggregate response rate was only 48 percent, with 47 percent of students responding
at UTSG and 50 percent of students responding at UTM. However, attrition from the follow-up
In Figure 2, we quantify the amount of available time students at UofT are not using toward
studying by subtracting reported study time from available time and plotting the resulting
distribution of remaining time. 21 The vertical lines in the figure represent the 25th, 50th, and 75th
percentiles, respectively, indicating that three-quarters of students expect to forgo at least 65 hours
per week in potential study time, 50 percent of students expect to forgo at least 78 hours, and one-
quarter of students expect to forgo more than 87 hours a week. We note again that these
calculations already account for sleeping time, class time, and self-reported expectations for time
required for working and commuting to and from school each week. 22
Table 5 presents summary statistics for self-reported student study time at UTSG and UTM in
the fall semester during a typical week and during a week spent preparing for midterms or exams.
students have available, we have also done calculations where we conservatively assume that students only have 60
hours per week for being productive in school. Taking 60 hours per week as the total available time and subtracting
time spent working (for pay), commuting, and attending class, the median student still has 41 hours remaining and 90
percent of students have at least 27 hours per week remaining.
20
The sample in Figure 1 is restricted to students in the control groups across both campuses of UofT.
21
We construct this figure by restricting the sample to students in the control group and pooling together students at
UTSG and UTM.
22
Using the more conservative calculation that assumes students only have 60 hours per week for being productive in
school (see footnote 19), the median student expects to forgo 26 hours per week and 75 percent of students expect to
forgo at least 14 hours per week.
21
Across both campuses, the average student in the control group reports having spent only 15.6
hours outside of class studying during average week in the fall semester. 23 During a week of
11F
preparing for midterms or exams, students report studying 24.8 hours, on average – an increase of
nearly 10 hours from a typical week but still only marginally more than the number of hours one
typically spends at a part-time job. In terms of the breakdown across campuses, students at UTSG
study more than those at UTM: the average student at UTSG reports studying 16.8 hours, on
average, during a regular week and 28.1 hours during a week before exams, while the average
student at UTM reports studying 14 hours during a regular week and 20.1 hours before exams.
The survey evidence implies that students at UTSG and UTM study relatively little. Yet
descriptive associations between study time and academic outcomes imply that many students
could likely benefit from increasing their study time. In Figure 3, we pool the control groups
across both campuses and plot descriptive associations between self-reported hours spent studying
during a typical week in the fall semester and the average grade across all courses taken in that
semester, the GPA earned across all courses, and the number of credits earned. All relationships
are positive and statistically significant, implying that an increase in weekly study time of one
standard deviation (13 hours) is associated with an increase in average course grades of 13.5
The relatively small magnitudes of these associations are likely driven by measurement error
in study time attenuating the relationships, as student study time is self-reported retrospectively. 24
23
We focus only on students in the control group in this subsection, deferring an exploration of whether treatment
significantly increased student study time to Section 5.
24
To mitigate the attenuation bias stemming from measurement error in study time, we have instrumented for study
time using the following variables from the baseline survey: self-reported study hours per week in high school,
tendency to regularly “cram” for exams, expected hours per week spent working for pay during the semester, and
22
Comparing our estimates to those in previous work, Brint and Cantwell (2008) also use
Experience Survey to show that a one standard-deviation increase in weekly study time is
associated with an increase in GPA of 10 percent of a standard deviation, an estimate that is very
close to the one we report above. Accounting for measurement error in retrospective self-reports,
Stinebrickner and Stinebrickner (2004) use time-diary data collected at six different times during
the academic year at Berea College to estimate that a one standard-deviation increase in (daily)
study time is associated with a 0.43 standard deviation higher college GPA. 25 Further addressing
the inherent endogeneity of student study time, Stinebrickner and Stinebrickner (2008) use the
same data but instrument for study time with a variable indicating whether a student’s roommate
brought a video or computer game to campus, finding that a one-standard deviation increase
Our data on student study time from WGU do not suffer from problems related to retrospective
self-reporting, as the online delivery of education allows us to gather very accurate information on
student activity. These data indicate that students taking courses online with WGU also appear to
study quite little. Figure 4 shows the distribution of how many days per week they log into WGU’s
online portal. The average student logs into the WGU portal 2.1 days per week. In addition, 90
percent of students log in less than 3.7 days per week and 18.5 percent of students log in less than
expected commuting time to campus. IV estimates indicate that a one standard deviation increase in study time is
associated with an increase in mean grades and GPA in the fall semester of 27 and 28 percent of a standard deviation,
respectively. OLS and IV estimates for credits earned are very similar.
25
Stinebrickner and Stinebrickner (2004) also find evidence of non-linear effects on study time on grades, where the
effect of study time is diminishing. We tested for non-linear effects by adding a quadratic study time term in each of
the specifications in Figure 3 but the quadratic terms were not significant in any specification.
26
To compare the estimates from Stinebrickner and Stinebrickner’s daily study time data with those from our weekly
data, note that the standard deviation of daily study time in their data is 1.62 hours per day (or 11.34 hours per week)
and the standard deviation of GPA is 0.686 points. In our data at UofT, the standard deviation of study time is 13
hours per week and the standard deviation of GPA is approximately 1 point.
23
1 day per week. Although it is possible that students are studying outside of the WGU website, the
log-in data indicate that many students access materials on WGU’s portal quite infrequently. In
Figure 5, we plot the correlation between days logged in per week and credits earned. We find a
strong and statistically significant positive relationship between log in activity and credits earned,
with a one standard deviation increase in days logged in per week (1.3 days) correlating with a
Taken together, the descriptive evidence implies that many students at all three experimental
sites study quite little, with large slack for potentially increasing study intensity. We explore
whether our planning intervention was effective at increasing student study time and academic
Having successfully randomized students across treatment and control groups at UTSG, UTM,
and WGU, we estimate the effects of the planning treatment with a comparison of mean outcomes
in a simple regression framework. The main specification, which we estimate separately at each
site, is given by
𝑦𝑖 = 𝛽0 + 𝛽1 𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡𝑖 + 𝜌′ 𝑋𝑖 + 𝑢𝑖 . (1)
Here, the outcome of student is regressed on an indicator for the student being assigned to the
The main parameter of interest is 𝛽1 , the estimated effect of the planning treatment. This
parameter represents an Intent-to-Treat effect, as students are included in the treatment group
regardless of whether they completed the online exercise, took it seriously, provided their phone
number, responded to a coach, or used their weekly calendar. Given that our completion rates and
24
opt-in rate are quite high, these estimates are likely close to the average treatment effect of
With respect to outcomes, at UofT, our main outcomes of interest are course grades, overall
grade point average (GPA), the number of credits attempted, the number of credits earned, and
persistence into second semester. At WGU, our main outcomes of interest are the number of credits
attempted, the number of credits earned, the number of days until a student completed his or her
first credit, and retention. 28 When the outcome of interest is course grades, we stack all course
grades and run a regression at the student-course level, clustering standard errors at the student
level. The effects on all other outcomes are estimated with regressions at the student level and
5. Results
In this section, we present the estimated effects of the planning treatment on student self-reported
study times (at UofT), online activity (at WGU), and academic outcomes (at both UofT and WGU),
We begin by discussing treatment effects on student self-reported study time from the follow-up
27
Recall that 97 percent of students completed the online exercise at the UofT campuses. In addition, 80 percent of
students who were invited to participate in the text-messaging program provided a phone number. All students who
were assigned to the treatment group at WGU were required to complete the planning module and submit a
screenshot of their study calendar as a condition of enrollment. The enrollment module at WGU does not allow
students in the treatment group to advance until they have completed these steps.
28
We do not include grades as an outcome at WGU because WGU does not give traditional grades in courses. Instead,
all courses at WGU are graded as pass/fail.
25
The average student in the control group at UofT spent 15.6 hours studying outside of class
during a regular week in the fall semester and 24.8 hours studying when exams were approaching.
Table 6 reports estimated treatment effects on both outcomes in the full sample of UofT students
and separately by campus. The estimated average treatment effects are presented, with and without
control variables, respectively, in columns (3) and (4). Treatment effects on study time during a
regular week in the pooled sample range between 1.65 and 1.69 hours and are statistically
significant at the 1 percent level. Students who were assigned to the planning treatment therefore
self-report studying nearly two more hours during a regular study week than non-treated students
and treatment effects are nearly identical across UTSG and UTM, as in indicated in the bottom
two panels of Table 6. The estimates in Table 6 also reveal that the planning treatment does not
affect student study time during exam or midterm periods, on average, as the effects are small and
throughout the distribution of students by plotting separate densities by treatment and control
group for student study time during an average week and for study time during a week with
midterms or exams approaching. The average treatment effect on study time during an average
week (reported above) appears to stem from the planning intervention causing fewer students to
self-report studying less than 15 hours per week and more students to report studying between 15
and 45 hours per week. The patterns for the densities of study time during an exam period are less
clear, as the planning module resulted in more students reporting studying between 17 and 37
hours but fewer students studying above 50 hours. Because of these competing forces, the
26
In addition to study hours at UofT, we test whether the treatment at WGU affected study
times as measured by logins and click data. The main outcomes of interest are the number of days
per week a student logs into WGU’s online portal and the log number of mouse clicks, log number
of mouse moves, and log number of page scrolls. Although these data have limitations because
students could be studying outside of the WGU website, they do contain precise information on
frequency and intensity of student interaction with the online portal. Table 7 shows that for all four
outcomes variables there is no evidence that the intervention affected student study time. Figure
7 underscores this point, showing that the average number of days students log into the WGU
website during each week of the semester do not differ across the treatment and control groups.
Taken together, we find suggestive evidence that treatment caused an increase self-reported
study time during an average week at UofT, while we find little evidence to support treatment
causing an increase in study time during midterm and exam periods. Further, we find no evidence
that treated students at WGU changed their study time in response to the intervention. Despite
estimating a statistically significant effect of the intervention on study time during a regular week
at UofT, we are cautious about interpreting this as a real effect because treated students may have
been primed to inflate their self-reported study time relative to students in the control group. We
revisit and expand on this notion when we discuss our results in Section 6 below, at which point
we reconcile the estimated effects on study time with the effects on academic outcomes.
Table 8 reports treatment effects for several academic outcomes estimated separately at UTSG,
UTM, and WGU. Outcomes at UofT are measured throughout the entire 2017-18 academic year,
while outcomes at WGU are recorded for all students who enrolled between January 2 and March
1 of 2017. We define the ‘retention’ outcome as a binary variable capturing whether a student was
27
enrolled in the winter semester of the 2017-18 academic year at UofT and whether a student was
The planning treatment appears to have no effect on students’ academic outcomes. The
results in Table 8 indicate that treated students do not attempt or earn more credits than students
in the control group and they are not more likely to persist into second semester. These results are
robust across all three experimental sites and to estimating treatment effects with and without other
student-level control variables. 29 At WGU, there is suggestive evidence that treatment may have
12F
actually reduced retention into next semester, with students in the treatment group being 1.5
percentage points less likely to enroll. This is a small effect, however, corresponding to 1.7 percent
In Table 9, we investigate treatment effects on course grades and GPA at the UofT
campuses and the number of days until a student earns his or her first credit at WGU. At UofT,
we show treatment effect estimates on course grades from stacked regressions where the unit of
observation is a student-course and standard errors are clustered at the student level. We also
present estimated treatment effects on courses taken during the fall semester, courses taken during
the winter semester, and all courses taken during the full academic year. 30 When the outcome is
13F
student GPA from the full academic year, we run the regression at the student-level and report
29
At UTSG and UTM, control variables include student age, self-reported study hours per week during high school,
expected paid-work hours per week, tendency to think about future goals, tendency to study at the last minute,
difficulty transitioning to university, commuting time (in minutes) to campus, and indicator variables for first-year
status, international student status, first-generation status, gender, English mother-tongue status, a self-reported desire
to earn more than an undergraduate degree, and a self-reported expectation to earn an A- average grade or greater. At
WGU, control variables include age, sex, race, first generation status, employment status, and income (bins).
30
Courses from the entire academic year include fall semester courses, winter semester courses, and courses that span
both semesters.
28
The planning intervention did not significantly affect student grade outcomes at either
campus of UofT. This result is robust to considering courses from each semester separately and to
including additional control variables. Similarly, at WGU, we find that the planning intervention
did not have any impact on the number of days students needed to complete their first credit. We
provide a more detailed discussion of these estimated null effects in Section 6 below, where we
interpret and reconcile these results with the effects of treatment on study time.
We now present estimated treatment effects on academic outcomes across a variety of student
subgroups.31 Specifically, at both the UofT campuses and WGU, we investigate whether treatment
effects are differential by student gender, age, employment status, and first-generation status. At
the UofT campuses, we also explore potentially different treatment effects across international and
domestic students and first-year and non-first-year students; while at WGU, we also differentiate
In Tables 10 and 11, we report the effects of the planning intervention on all course grades
across student subgroups at UTSG and UTM, respectively. The planning module does not appear
to have caused an improvement in student grades among any subgroup of students at UTSG, as no
treatment effect is economically or statistically significant. At UTM, treatment effects are negative
and marginally statistically significant (at the 10 percent level) for male students and those who
expect to work than 8 hours per week at the start of the academic year. However, given the many
hypotheses being tested in the subgroup analyses across UTSG and UTM (24 hypothesis) and the
31
In our analysis of subgroups (and treatment effects in the full sample above), we closely follow our AEA pre-
registered analysis plans (registration ID AEARCTR-0000972 at WGU and AEARCTR-0000810 at UofT).
29
lack of an overall treatment effect in the main sample, we interpret these negative effects
subgroups at WGU. As in the aggregate analysis, the planning module appears to have no effect
on credit accumulation for any group of students. 32 In particular, we note that there are no
differences in treatment effects across students who are employed full-time, part-time, or
unemployed, suggesting that the absence of a treatment effect in the full sample is not driven by
students who work full-time not having the time available to increase their study effort. Treatment
effects are also similar across students from households with different incomes.
Comparing the estimated treatment effects across all three experimental sites, treatment effects are
similar across older (20 years of age or older at UofT and 30 years of age or older at WGU) and
younger students, suggesting that student maturity (as proxied by age) is not an important factor
in explaining our null treatment effects. It is also the case that treatment effects do not differ by
first-generation status (at both UofT and WGU), international student status (at UofT), or first-
year status (at UofT), indicating that familiarity with institutional features is also unlikely to be an
6. Discussion
We now discuss potential mechanisms underlying the estimated null effects on academic
achievement and the contributions of these findings to the broader literature on student decision-
32
Treatment effects across subgroups on credit accumulation and persistence are similarly small and insignificant at
both WGU and the campuses of UofT. The results are available upon request.
30
We first revisit the idea of whether the intervention affected study time at UofT, arguing that
the small positive effect we estimate above of treatment on study time during a regular week at
UofT likely suffers from self-reporting bias. In particular, treated students were likely primed to
inflate their answers relative to students in the control group. We believe this is likely for three
reasons. First, having gone through an exercise that emphasized the importance of adequate study
time, students in the treatment group may have felt compelled to inflate their reported hours on the
follow-up survey, as a result of feeling like they made a (indirect) promise to study more. Second,
the estimated effects on study time during a week spent preparing for midterms or exams are much
weaker. If the effect on study time found during a regular week is real and treated students
consistently studied more throughout the semester, we would expect to find clear effects on study
time during the exam period as well. Because the calendar intervention did not explicitly discuss
study time during exams or require students to plan their schedules during exam periods, the
weaker effects on exam week study time are consistent with treated students inflating their self-
reports for the input that the intervention did target (i.e., regular study time). Third, although the
samples at UofT and WGU are different, the precisely estimated null effects of the intervention on
objective measures of study time at WGU serve as evidence consistent with the idea that real study
Even if one assumes the intervention did increase student study time, the impact was likely not
large enough to translate into a significant effect on achievement. The descriptive relationships in
Figure 3 (discussed in subsection 4.3) indicate that increasing study time by 1.65 hours (the
estimated effect of the planning intervention) is associated with an 0.22-point increase in average
course grades, a 0.02-point increase in GPA, and an increase in credits earned of 0.01. All three
implied effects are within the confidence interval pertaining to the point estimate for effect of the
31
planning treatment on the relevant outcome. 33 Indeed, it may be the case that marginal increases
in student study times are not enough to generate meaningful improvement in academic outcomes.
What explains the lack of student responsiveness to the intervention and what might an
effective intervention look like? The data we gathered on time commitments from the baseline
survey and on student study time from the follow-up survey (at UofT) clearly show that binding
time constraints are not preventing students from studying more, as our most conservative
calculations indicate that 75 percent of students are forgoing at least 14 hours per week of potential
study time. Further, the fact that the intervention was ineffective despite maintaining weekly
contact with students and reminding them of their study goals also implies that helping students
keep their study goals salient is likely an ineffective way to improve academic outcomes.
We also investigated whether our null results can be explained by student procrastination or
time-inconsistent preferences. That is, students may intend to devote sufficient time to studying
but fail to follow through with their intentions. We used information collected during the baseline
survey at UofT to explore treatment effects in subgroups of students who may have greater or
lower propensities to procrastinate. Specifically, students reported on a five-point scale (i) their
tendency to study at the last minute or cram for exams, (ii) their assessment of their time
management skills, and (iii) whether they are likely to finish what they start. We also recorded the
number of days elapsed between when the online intervention was made available and when each
33
Pooling across both UofT campuses, the estimated treatment effect on the average grade across all fall semester
courses is -0.2 grade points, with a 95-percent confidence interval ranging between -1.14 and 0.75. The estimated
treatment effect on fall semester GPA is -0.001 with a confidence interval ranging between -0.074 to 0.072, while the
estimated treatment effect on the number of credits earned in the fall semester is -0.014 with a confidence interval
ranging between -0.061 to 0.033. As mentioned in Section 4.3, our estimated association between GPA and study time
is likely attenuated by measurement error in study time due to retrospective self-reporting. Using instead the estimated
relationship between GPA and study time in Stinebrickner and Stinebrickner (2008), an increase in weekly study time
of 1.65 hours would be associated with an increase in fall GPA of 0.09 points. This is barely outside of the upper end
of our 95-percent confidence interval but within the associated 99-percent confidence interval (-0.097, 0.095).
32
student started their respective exercise. 34 At each UofT campus, we split the sample by whether
students are above or below the median with respect to each of these four variables and estimated
separate treatment effects in each subgroup. We rarely find any evidence that treatment effects
differ across students with different procrastination tendencies and we never estimate a statistically
significant positive treatment effect among students who have a low propensity to procrastinate.35
The evidence is therefore inconsistent with student procrastination tendencies being a barrier to
treatment effectiveness.
Given that binding time constraints, a lack of goal salience, and student procrastination are
unlikely explanations for our null results, we believe that there are two remaining possibilities for
why students study relatively little and why our intervention was ineffective. First, it may be the
case that students do not want to study more because, despite being more likely to produce better
grades, the high-effort strategy is too costly compared to the low-effort strategy in which students
have more time for leisure. Most students in our sample do eventually receive their degrees (at U
of T) or have a fallback career if they do not graduate (at WGU), which perhaps makes the
perceived benefits of effort relatively small. Second, despite going through the intervention,
students may remain unsure about how additional study time translates into higher grades or about
the benefits of attaining higher grades on post-graduation outcomes. Under either scenario, the
ambiguity around the benefits of increasing study time could prevent students from putting forth
the costly effort. Future research should aim to better understand the role of student motivations,
goals, and perceptions when it comes to effort investments during their time in college and design
34
Using a recent sample of UofT students, Beattie, Laliberte, and Oreopoulos (2018) demonstrate that these measures
of procrastination are strong predictors of students failing to realize their performance expectations.
35
The results are available upon request.
33
7. Conclusion
In this study, we examine whether an intervention focusing on study time can improve student
outcomes in three distinct academic environments: a selective four-year college, a less selective
four-year college, and an online university. Our analysis is motivated by patterns of very low
study times observed among students in our populations and documented by other scholars (e.g.
Babcock and Marks, 2011). Despite recommendations to treat studying like a full-time job,
students at the UofT campuses only report studying 16.8 and 14 hours per week, on average, at
UTSG and UTM, respectively. Further, the median student at UTSG studies only 12 hours per
week while the median student at UTM studies only 10 hours per week. At WGU students only
log on to the course website an average of 2.1 days per week. Although students in each
environment appear to have the ability to increase their studying, we find no impacts of our
planning treatments on student study times or academic outcomes at any of the three academic
environments we study.
The lack of positive effects of our treatments on study time, grades, credits earned, and
retention across the three academic environments we study and across all demographic subgroups
we observe suggest that the planning interventions we test are broadly ineffective at improving
student outcomes. Although we cannot definitively identify the barrier(s) that prevented the
intervention from working, we argue that it is unlikely that binding time constraints, a lack of study
goal saliency, or student procrastination tendencies are driving our null results. Instead, it is
possible that students find effort investments (in the form of increasing study time) too costly or
that they face ambiguity about either the returns to studying on grades or the returns to grades on
post-graduation outcomes. We are currently exploring these possibilities in our ongoing work, in
which we have combined a related time management intervention with unique survey questions
34
designed to further tease out student motivations, goals, and perceptions when it comes to effort
35
References
Arum, Richard, and Josipa Roksa. 2011. “Academically Adrift: Limited Learning on College
Babcock Philip and Mindy Marks. 2011. “The Falling Time Cost of College: Evidence form Half
a Century of Time Use Data.” The Review of Economics and Statistics. 93(2): 468-478.
Baker, Rachel, Brent Evans, and Thomas Dee, 2016, “A Randomized Experiment Testing the
Open, 2 (4)
Beattie, Graham, Jean-William P. Laliberté and Philip Oreopoulos. 2018. "Thrivers and Divers:
2018. " What Sets College Thrivers and Divers Apart? A Contrast in Study Habits,
Attitudes, and Mental Health," NBER Working Paper # 23588. Cambridge, Mass.:
Bettinger, Eric, and Rachel Baker. 2014. “The Effects of Student Coaching: An Evaluation of a
Analysis. 35(1):3-19.
Bettinger, Eric, Sten Ludvigsen, Mari Rege, Ingeborg F. Solli, and David Yeager. 2018.
36
Bettinger, Eric, Lindsay Fox, Susanna Loeb, and Eric S Taylor, “Virtual Classrooms: How
forthcoming.
Brint, Steven and Allison M. Cantwell. 2010. “Undergraduate Time Use and Academic
Buehler, Roger, Dale Griffin, and Michael Ross. 1994. “Exploring the "planning fallacy": Why
people underestimate their task completion times.," Journal of personality and social
psychology, 67 (3),366.
Cadena, Ximena and Antoinette Schoar. 2011. “Remembering to Pay? Reminders vs. Financial
Incentives for Loan Payments”. NBER Working Paper # 17020. Cambridge, Mass.:
Calzolari, Giacomo and Mattia Nardotto. 2012. “Nudging with Information: A Randomized
Field Experiment". CEPR Discussion Paper # DP8671. London, UK. Center for
Carrera, Mariana, Heather Royer, Mark Stehr, Justin Sydnor, and Dmitry Taubinsky. 2018 "The
Making Plans to Exercise." NBER Working Paper # 24959. Cambridge, Mass.: National
Castleman, Benjamin, and Lindsay Page. 2015. “Summer nudging: Can personalized text
messages and peer mentor outreach increase college going among low-income high
37
__________. 2016. “Freshman Year Financial Aid Nudges: An Experiment to Increase FAFSA
Castleman, Benjamin, and Katharine Meyer. 2016. “Can Text Message Nudges Improve
Academic Outcomes in College? Evidence from a West Virginia Initiative.” Center for
Clark, Damon, David Gill, Victoria Prowse, and Mark Rush. 2017. “Using Goals to Motivate
College Students: Theory and Evidence from Field Experiments.” NBER Working Paper
Doherty, William. 2006. “An Analysis of Multiple Factors Affecting Retention in Web-Based
Community College Courses.” The Internet and Higher Education. 9(4): 245-255.
Damgaard, Mette Trier and Helena Skyt Nielsen. 2018. “Nudging in education.” Economics of
Deslauriers, L., Schelew, E. and Wieman, C., 2011. "Improved learning in a large-enrollment
Dobronyi, Christopher R., Philip Oreopoulos, and Uros Petronijevic. 2017. “Goal Setting,
self evaluations undermine students’ learning and retention.” Learning and Instruction,
22(4), pp.271-280.
38
Dunlosky, J., Rawson, K.A., Marsh, E.J., Nathan, M.J. and Willingham, D.T., 2013. “Improving
students’ learning with effective learning techniques: Promising directions from cognitive
and educational psychology.” Psychological Science in the Public Interest, 14(1), pp.4-
58.
Ericson, Keith M Marzilli, 2017. “On the Interaction of Memory and Procrastination:
Larry G Epstein, Yoram Halevy. 2018. “Ambiguous Correlation,” The Review of Economic
Figlio, David, Mark Rush, and Lu Yin, “Is it live or is it internet? Experimental estimates of the
(4), 763-784.
Forsyth, Darryl K and Christopher DB Burt. 2008. “Allocating time to future tasks: The effect of
task segmentation on planning fallacy bias," Memory & cognition, 36 (4), 791-798.
121.
Gollwitzer, Peter M. 1993. “Goal achievement: The role of intentions," European review of
39
Gollwitzer, Peter M, and Paschal Sheeran. 2006. “Implementation intentions and goal
Himmler, Oliver, Robert Jaeckle, and Philipp Weinschenk.. “Soft Commitments, Reminders and
Kahneman, Daniel and Amos Tversky, 1979, “Prospect theory: An analysis of decision under
Karlan, Dean, Margaret McConnell, Sendhil Mullainathan, and Jonathan Zinman. 2016. “Getting
to the Top of Mind: How Reminders Increase Saving." Management Science, 62(2):
3393-3411.
Laibson, David. 1997. “Golden Eggs and Hyperbolic Discounting." The Quarterly Journal of
Lavecchia, Adam, Heidi Liu and Philip Oreopoulos. 2016. “Behavioral Economics of Education:
Milkman, Katherine L., John Beshears, James J. Choi, David Laibson, and Brigitte C. Madrian.
Morisano, Dominique, Jacob B. Hirsh, Jordan B. Peterson, Robert O. Pihl, and Bruce M. Shore,
40
Mullainathan, Sendhil. 2002. “A Memory-Based Model of Bounded Rationality,” The Quarterly
Nickerson, David W and Todd Rogers. 2010. “Do you have a voting plan? Implementation
intentions, voter turnout, and organic plan making," Psychological Science, 21 (2), 194-
199.
O’Donogue, Ted and Matthew Rabin. 1999. "Do It Now or Later" American Economic Review,
89(1): 103-124.
Oreopoulos, Philip and Uros Petronijevic. 2018. "Student Coaching: How Far Can Technology
Oreopoulos, Philip, Uros Petronijevic, Christine Logel, and Graham Beattie. 2018." Improving
Patterson, Richard W. 2018. “Can Behavioral Tools Improve Online Student Outcomes?
Intervention Closes Both the Gender and Ethnic Minority Achievement Gap.” Palgrave
Scrivener, Susan and Michael J. Weiss. 2013. “More Graduates: Two-Year Results from an
41
Solomon, L.J. and Rothblum, E.D., 1984. Academic procrastination: Frequency and cognitive-
Stinebrickner, Ralph and Todd R. Stinebrickner. 2004. “Time-use and college outcomes.”
Stinebrickner, Ralph and Todd R. Stinebrickner. 2008. “The Causal Effect of Studying on
Academic Performance.” The B.E. Journal of Economic Analysis & Policy. 8(1): 1-55.
Twenge, J.M., Campbell, W.K. and Gentile, B., 2012. Generational increases in agentic self-
evaluations among American college students, 1966–2009. Self and Identity, 11(4),
pp.409-427.
Yeager, David, Carissa Romero, Dave Paunesku, Christopher S. Hulleman, Barbara Schneider,
Cintia Hinojosa, Hae Yeon Lee, Joseph O’Brien, Kate Flint, Alice Roberts, and Jill Trott.
2016. “Using Design Thinking to Improve Psychological Interventions: The Case of the
Psychology. 108(3):374–391.
42
Tables
43
Table 2: Balancing Tests at UTSG
Treatment Status
Control Treatment
44
Table 3: Balancing Tests at UTM
Treatment Status
Control Treatment
45
Table 4: Balancing Tests at WGU
Treatment Status
Control Treatment
46
Table 5: Summary Statistics for Study Habits from Follow-Up Survey at UofT
47
Table 6: Treatment Effects on Self-Reported Study Times at UofT
(1) (2) (3) (4)
Sample and Control Mean Treatment Treatment
Dependent Variable [Standard Deviation] Difference Difference
UTSG
Regular Week Study 16.756 1.769** 1.618*
[13.885] [0.870] [0.844]
953 873
UTM
Regular Week Study 13.978 1.505* 1.633**
[11.842] [0.836] [0.796]
766 755
Controls? No Yes
Notes: The dependent variable in each regression and the sample used are indicated in the rows
of column (1). The unit of observation is a student. Control variables include student age, self-
reported study hours per week during high school, expected paid-work hours per week,
tendency to think about future goals, tendency to study at the last minute, difficulty
transitioning to university, commuting time (in minutes) to campus, and indicator variables
for first-year status, international student status, first-generation status, gender, English
mother-tongue status, a self-reported desire to earn more than an undergraduate degree, and a
self-reported expectation to earn an A- average grade or greater. Robust standard errors are
reported in brackets in columns (3) to (4). The number of observations used in each regression
appears below the standard errors. *** indicates significance at the 1 percent level; **
indicates significance at the 5 percent level; and * indicates significance at the 10 percent level.
48
Table 7: Treatment Effect on Click Data at WGU
(1) (2) (3) (4)
Dependent Variable Control Mean Treatment Treatment
[Standard Deviation] Difference Difference
Controls? No Yes
Notes: The dependent variable in each regression is indicated in the rows of column (1). The unit of
observation is a student. Control variables include age, sex, race, first generation status, employment status,
and six income bins: (1) less than 16,000, (2) 25,000-34,999. (3) 35,000-44,999, (5) 45,000-64,999, and (6)
65,000+. Robust standard errors are reported in brackets in columns (3) to (4). The number of observations
used in each regression appears below the standard errors.
49
Table 8: Treatment Effects on Credit Accumulation and Retention
(1) (2) (3) (4)
Dependent Variable and Control Mean Treatment Treatment
Sample [Standard Deviation] Difference Difference
Credits Attempted
UTSG 3.799 0.025 0.026
[1.243] [0.058] [0.056]
1,860 1,860
50
Table 9: Treatment Effects on Grades at UofT and Days to Completion at WGU
(1) (2) (3) (4)
Control Mean Treatment Treatment
Sample and Dependent
Variable [Standard Deviation] Difference Difference
UTSG
Fall Grades (2017-18) 71.020 0.039 0.002
[14.954] [0.622] [0.584]
5,413 5,413
Controls? No Yes
Notes: The dependent variable in each regression and the sample used are indicated in the rows of column (1).
Control variables used in the UofT samples are described in the notes of Table 6. Control variables use in the
WGU sample are described in the notes of Table 7. When the outcome is course grades, standard errors are
clustered at the student level and the unit of observation is a student-course. For other outcomes, robust
standard errors are reported, and the unit of observation is a student. Sample size from the regression appears
below the standard error.
51
Table 10: Treatment Effects on Course Grades by Student Subgroup at UTSG
(1) (2) (3) (4) (5)
Subgroup Observations Control Mean Treatment Treatment
[Standard Deviation] Difference Difference
Male 5,799 70.231 -0.313 -0.384
[16.398] [0.916] [0.862]
Controls? No Yes
Notes: The dependent variable in each regression is course grades. Control variables are described in the
notes of Table 6. The subsample of students used for each regression is indicated by the rows of column
(1). The unit of observation in each regression is a student-course and standard errors are clustered at the
student level.
52
Table 11: Treatment Effects on Course Grades by Student Subgroup at UTM
(1) (2) (3) (4) (5)
Subgroup Observations Control Mean Treatment Treatment
[Standard Deviation] Difference Difference
Male 4,270 65.030 -2.390** -1.902*
[16.353] [1.153] [1.095]
Controls? No Yes
Notes: The dependent variable in each regression is course grades. Control variables are described in the
notes of Table 6. The subsample of students used for each regression is indicated by the rows of column
(1). The unit of observation in each regression is a student-course and standard errors are clustered at the
student level.
53
Table 12: Treatment Effects on Earned Credits at WGU by Student Subgroup
(1) (2) (3) (4) (5)
Subgroup Observations Control Mean Treatment Treatment
[Standard Deviation] Difference Difference
Male 2110 15.130 0.083 0.128
[12.130] [0.525] [0.515]
Female 3955 14.063 0.228 0.200
[9.592] [0.311] [0.303]
Age >= 30 3820 14.644 0.423 0.330
[10.535] [0.351] [0.341]
Age < 30 2245 14.089 -0.329 -0.199
[10.578] [0.433] [0.423]
Black 670 11.172 -1.250∗ -0.788
[10.627] [0.760] [0.691]
White 4781 15.060 0.311 0.259
[10.715] [0.312] [0.307]
Hispanic 651 12.339 0.652 0.469
[7.499] [0.657] [0.677]
Employed full time 4465 13.998 0.202 0.281
[9.740] [0.292] [0.289]
Employed part time 801 15.216 -0.360 -0.348
[9.740] [0.292] [0.289]
Not Employed 580 14.879 -0.037 -0.055
[10.325] [0.769] [0.771]
First generation student 2631 13.679 0.306 0.365
[9.708] [0.388] [0.374]
Household income below 45, 000 1838 13.005 0.322 0.524
[10.875] [0.501] [0.463]
Household income above 45, 000 4227 15.082 0.044 0.001
[10.341] [0.325] 0.325]
Controls? No Yes
Notes: The dependent variable in each regression is earned credits. Control variables are described in the notes of
Table 7. The subsample of students used for each regression is indicated by the rows of column (1). The unit of
observation in each regression is a student and robust standard errors appear in brackets.
54
Figures
.06
.04 Reported Study Time and Available Time
Density
.02
0
55
Time Remaining Each Week After Studying
.03
.02
Density
.01
0
10 20 30 40 50 60 70 80 90 100
Available Time minus Study Time
56
Course Grades vs. Study Time per Week
Fall GPA vs. Study Time per Week
76
3
74
2.8
72
Fall GPA
Coeff = 0.134
(0.026) Coeff = 0.012
(0.002)
2.6
70
2.4
68
66
2.2
0 10 20 30 40 50 60 0 10 20 30 40 50 60
Hours Studying per Week Hours Studying per Week
1.4
Coeff = 0.006
(0.002)
1.3
1.2
0 10 20 30 40 50 60
Hours Studying per Week
Figure 3: Relationships between Fall Semester Study Time and Academic Outcomes at UofT
Notes: This figure presents estimated associations between the number of hours students self-report studying during a
regular week in the fall semester and various academic outcomes in that semester. In panels (a), (b), and (c),
respectively, the outcomes are average grade across all fall semester courses, grade-point average (GPA) across all
fall semester courses, and the number of credits earned during the fall semester. The sample in each panel is restricted
to students in the control group across both campuses of UofT. We construct each panel by first grouping students
into 20 equally-sized (vingtile) bins of the study time distribution and then calculating the mean study time and
outcome within each bin. The blue circles in each panel represent these means, while the red dashed lines represent
the associated linear relationships, estimated on the underlying student-level data.
57
Figure 4: Days Logged in Per Week
Notes: This figure shows the distribution of the average number of days a student logs into WGU’s online portal per
week. The data used is for all WGU students from January 1, 2015 to July 23, 2018. The vertical line represents the
median of the average number of days per week a student logs in.
58
Figure 5: Relationship between Days Logged in and Credits Earned at WGU
Notes: This figure presents estimated association between the days per week students log in to the WGU web portal
and credits accumulated during the semester. The sample is restricted to students in the control group at WGU. We
construct this figure by first grouping students into 20 equally-sized (vingtile) bins in the distribution of the mean
number of days logged in per week and then calculating the mean number of credits earned within each bin. The
plotted circles represent these means, while line represents the associated linear relationship, estimated on the
underlying student-level data.
59
Study Time During Regular Week
.05
.04
.03
Density
.02
.01
0
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95
Hours/Week
Control Treatment
.01
0
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95
Hours/Week
Control Treatment
60
Figure 7: Days Logged in By Week of Semester
Notes: This figure shows the average number of days students log into the WGU website for each week during their
first semester. The solid line is for the treatment group and the dashed line is for the control group. The data used is
for WGU students in the experimental sample.
61
Appendix A: WGU Details
Western Governors University (WGU) is an accredited non-profit online university. 36 It has an 14F
universities, offers degrees in business and vocational areas including (1) Business, Management,
and Marketing, (2) Computer and Information Science, (3) Education, and (4) health professions.
Like many online student populations, students at WGU are older than traditional college students,
have a higher utilization of federal student loans (59%), persist and graduate at lower rates than at
traditional universities, with 74% persisting into second year and 26% graduating within 6 years. 37 15F
WGU also differs in a several important ways from many other online universities. First,
course and term schedules at WGU are completely asynchronous. Students can enroll in any month
and work through courses at their own pace. Students pay approximately $3000 for each six-
month semester and can complete as many courses in that time as they would like. Second, instead
projects or proctored course competency exams. 38 Third, with few exceptions, WGU students are
16F
required to successfully complete at least one college course at another institution before they can
36
Accreditation by Northwest Commission on Colleges and Universities for general programs and CNURED for
nursing programs. Source: https://nces.ed.gov/collegenavigator/?q=Western+Governors\&s=all\&id=433387.
Accessed 8/16/2017.
37
Source for information on degree programs, federal loan utilization, retention, and graduation rates comes from the
Integrated Postsecondary Education Data System (IPEDS):
https://nces.ed.gov/collegenavigator/?q=Western+Governors\&s=all\&id=433387. Accessed 8/16/2017
38
Students who repeatedly fail course competency exams not only fail a course but are also required to leave the
university.
62