Code in Place: Online Section Leading for Scalable

Human-Centered Learning
Christopher Piech, Ali Malik, Kylie Jue, Mehran Sahami
Computer Science Department, Stanford University
{piech,malikali,kyliej,sahami}@cs.stanford.edu

ABSTRACT 1 INTRODUCTION
Could it be the case that the number of people who want to teach Demand for educational opportunities in computer science remains
computer science, and have the potential, is roughly proportional to strong, especially given the large number of unfilled positions in the
the number of people who want to learn? During the time of COVID- IT sector [17]. Over the past decade, Massive Open Online Courses
19 we offered a free CS1 class to people around the world. Well- (MOOCs) have seen a rapid rise as a potential means for virtually
aware of the high drop-out rates reported in many massive open- unlimited educational opportunity [9] especially for those who
access online courses (MOOCs), we augmented our course with a traditionally don’t have access [4]. Although MOOCs have attracted
scalable, human-centered solution: section leading. Section leaders millions of students across the globe [18], studies have revealed
teach small, weekly interactive learning sessions. We hypothesize these courses suffer from signficant issues such as extremely low
that the personalized attention adds a sense of responsibility for completion rates [16, 25]. For example, a comprehensive study of
both student and teacher which drives learning. We recruited over over 4 million course participants in MOOCs on the edX platform
900 volunteer section leaders and more than 10,000 students in reported a 5.5% certification rate overall, despite more than half of
the class. To our knowledge this is the largest group of section students indicating an intention to earn certificates in the courses
leaders in a single CS1 course offering and the most small group they enrolled in [2].
interactions. The completion rate in our class was more than 10 In the context of the COVID-19 pandemic, the need for educa-
times that usually reported for similar MOOCs. Additionally, 99% tional opportunities in computing took on new urgency. With many
of the volunteer section leaders taught through the entire span of workers finding themselves furloughed or unemployed, jobs in IT,
the course, showing the potential for large scale volunteer-driven which can often be done remotely, offered new potential employ-
education, and the benefit that teachers themselves derive. We also ment opportunities. Moreover, many people who were sheltering in
discovered the potential for replication of this model, as 34% of place might have had more time to dedicate towards learning. This
students in a representative-sample survey indicated they would confluence of factors motivated the need for a broadly available on-
serve as section leaders for a future offering of the course. This level line introductory programming class that could more successfully
of participation would be more than sufficient to field additional retain students compared to standard MOOCs.
offerings of the course sustainably. We believe this is an intriguing To address this problem, we created “Code in Place” (a play on
case study of a model for significantly scaling human-centric CS the phrase “shelter in place”), an introductory online programming
education for all. class in Python based on the CS1 course at Stanford University.
Notably, we went beyond the traditional MOOC model by centering
CCS CONCEPTS the courses around the idea of section leaders. These were skilled
• Social and professional topics → CS1. volunteers who had experience with the course material, who would
meet weekly with small groups of roughly 10 students each to
KEYWORDS provide more personalized instruction and motivate students to
continue in the course. The section leaders also participated in an
CS1, online learning, teaching at scale online discussion forum to answer students’ questions, provided
ACM Reference Format: guidance and feedback on assignments, and helped foster a positive
Christopher Piech, Ali Malik, Kylie Jue, Mehran Sahami. 2021. Code in community for all participants in the course.
Place: Online Section Leading for Scalable Human-Centered Learning. In The focus of the class was on creating a human-centered learn-
Proceedings of the 52nd ACM Technical Symposium on Computer Science ing model built around a community that stresses learning for all,
Education (SIGCSE ’21), March 13–20, 2021, Virtual Event, USA. ACM, New the importance of kindness, and peer support.
York, NY, USA, 7 pages. https://doi.org/10.1145/3408877.3432562
1.1 Contributions
Permission to make digital or hard copies of all or part of this work for personal or
classroom use is granted without fee provided that copies are not made or distributed Our work explores the potential of harnessing the power of section
for profit or commercial advantage and that copies bear this notice and the full citation leaders and human-centric education in a scalable online educa-
on the first page. Copyrights for components of this work owned by others than the tional setting. We hypothesize that the use of section leaders will
author(s) must be honored. Abstracting with credit is permitted. To copy otherwise, or
republish, to post on servers or to redistribute to lists, requires prior specific permission lead to higher completion rates in an online course, and more nov-
and/or a fee. Request permissions from [email protected]. elly, that it is possible to create a model for section leading that
SIGCSE ’21, March 17–20, 2021, Toronto, Canada scales to a course supporting thousands of students.
© 2021 Copyright held by the owner/author(s). Publication rights licensed to ACM.
ACM ISBN 978-1-4503-8062-1/21/03. . . $15.00 There are many challenges in creating such a human-centered
https://doi.org/10.1145/3408877.3432562 learning experience, most notably recruiting and training large
numbers of section leaders, and maintaining a healthy online com- Table 1: Stated reasons for applying to section lead
munity. The primary contributions presented in this work are:
Reason for wanting to volunteer Section Lead Percent of applicants
(1) Showing that a section-leading model can be scaled far be-
Give back through community service 86%
yond existing practice online.
Improve my own teaching ability 72%
(2) Teaching (to the best of our knowledge) CS1 with the largest Be part of an experiment in online education 66%
group of section leaders and most small group interactions. I just love teaching programming 66%
(3) Assessing how well this approach worked in its first iteration. Be part of a community of section leaders 57%
(4) Positing a model for how this approach might be replicated
sustainably in the future
(5) Providing open source materials for recruiting and training Compared to peer feedback models, section leaders can provide
section leaders as well as lesson plans for weekly teaching more skilled guidance on course material and, like any good teacher,
to make this approach more replicable by others. can also provide encouragement to students to continue in a course
The structure of this paper is as follows. Section 2 presents an through challenges. Such human interaction can create a greater
overview of the structure of the course, including details on the pro- sense of accountability for students as they know that someone is
cess for recruiting, selecting, and training section leaders to allow paying attention to their struggles.
the course to scale, as well as student selection, material covered, While the use of section leaders has grown in the traditional
and community development. Section 3 assesses the impacts of this in-person learning setting, they have not seen much use in large
course in terms of student and staff engagement, and provides a MOOCs. Generally, the use of section leaders to teach live, small
description of how this course model may be sustained in future section meetings and provide feedback on student assignments is
iterations. Section 4 discusses the results and considerations for not seen as a scalable model for supporting thousands of students
future offerings of such courses, as well highlights important limi- enrolling in free MOOCs.
tations. We hope that the insights in this paper allow for others to
replicate our positive learning experience. 2 THE CODE IN PLACE COURSE
The intentions behind Code in Place were to (i) create a joyful
1.2 Related Work learning experience for students, and (ii) provide an opportunity for
Our work continues on a long history of contributions torwards community service for those who wanted to share their knowledge
high-quality open-access education for all. of programming during the time of COVID-19. We aimed to build
Human-Centered Learning in MOOCs. Prior work has tried a course which was a meaningful learning experience, while also
to address the problems of low course completion rates and person- fostering a positive and generous community.
alized feedback in MOOCs by incorporating a more human-centered
learning approach. Meet-ups have been suggested as a way to boost 2.1 Section Leaders at Scale
the human element of MOOCs [6]. However, meet-ups are usually An important aspect of Code in Place was that it welcomed indi-
not considered an integral part of a course and, in many cases, are viduals from all walks of life who wanted to partake in teaching
only attended by a small fraction of students in a class. Recent programming as volunteer section leaders. Their commitment was
research has suggested that social contexts in MOOCs matter in a responsibility to teach a group of roughly ten students once a
subtle ways [3]. Some educational programming tools have also week for an hour and also to contribute to the online community
tried to leverage the human element. For example, the PythonTutor of learners.
platform allows users to help one another when they are stuck [7]. Hiring volunteer section leaders. Perhaps the most formida-
An extension of this work, CodeOpticon, even supports one-to- ble challenge in our effort was to hire sufficiently many experienced
many code help [8]. While useful, such tools primarily focus on volunteer section leaders who would allow the course to scale to
code-level rather than class-level support. thousands of students. Based on prior experience, we wanted to
With regard to providing feedback on students’ work, peer as- maintain a 1:10 ratio of section leaders to students. To recruit sec-
sessments [13] (and auto-graders [11]) have also been employed. tion leaders, we created an online application which required the
While such methods can be effective, they cannot provide person- submission of three components: (1) a recorded 5-minute teach-
alized guidance and support on course material in the way that a ing video, (2) a python code debugging exercise, and (3) a short
human well-versed in the subject can. written application including demographic information. Section
Section Leader Model. The use of section leaders—skilled un- leading was clearly advertised as a volunteer (unpaid) experience,
dergraduates who have taken the course before—to scale introduc- requiring 5 weeks of participation. The opportunity was advertised
tory programming classes has a long history. Originally motivated broadly through word of mouth, posts on a variety of mailing lists
by a desire to contain costs while scaling class size [21], the section and discussion forums, and was substantially aided by an article in
leader model has since been recognized to provide other benefits, Scientific American [15].
such as better fostering a learning community and providing a peer For the teaching video component of the application, applicants
mentoring model [22]. As a result, the use of section leaders has were asked to prepare a 5-minute teaching sample on a given Python
slowly seen more widespread adoption in higher education [5]. problem that covered variables, arithmetic, and user input. This
Section leading is a form of “near peer mentoring" which has many teaching demonstration allowed the applicant to motivate the prob-
well known benefits for both learner and teacher [14, 23]. lem and to identify and teach the concepts they thought would be
(a) (c) Non Binary (d)
Group Number of people 2.8%
Volunteer Section Leaders (SL) 908
Students 10,428 Women
Gender 49.7%
Men
Teaching Leaders (TL) 50 47.5%
Instructors 2

(b) 2500

Number of Students
Age
2000
1500
1000
Section: Teaching Group:
1 Section Leader per 1 Teaching Leader per 500 Num Students
10 Students 20 Section Leaders 0 1 5 10+
18-21 22-25 26-30 31-35 36-40 40+

Figure 1: (a) Course size. (b) Example “Section" and “Teaching Group". (c) There were more female students in the course than
male students and age was well distributed. (d) Students came from around the world.

most challenging for students, while also using the structure of our 1b). Specifically, section leaders were organized into small-groups
detailed lesson plan as a foundation. lead by a highly experienced teaching leader.
The debugging exercise presented applicants with a sample stu- Section leaders were asked to attend three training sessions in
dent solution to a small programming problem. The student solution total: a 30-minute welcome hosted by the course instructors, a
had three errors (of low, medium, and high complexity), and the 60-minute workshop to prepare for their first section, and a 60-
applicants had to identify the bugs and write feedback to the hypo- minute workshop after their first section to reflect on how to create
thetical student. Great care was taken in developing the prompts for an inclusive section culture. They were also welcomed to attend
the teaching demo and the debugging problem, resulting in several “Section Leader Learning Week”, a series of optional 45-60 minute
iterations of testing and refinement before release. workshops, for additional professional development opportunities
In the space of only four days, we received 1,123 applications. after their teaching responsibilities had concluded.
Each section leader applicant was manually reviewed by at least Training focused on good teaching practices as well as instilling
one of 33 different hand-picked evaluators from the Code in Place the importance of engaged problem-solving in sections. We im-
team over a 12 hour time window1 . mersed the section leaders in active learning techniques during the
Section leaders applied from 97 countries, were fluent in 64 differ- small group workshops by having them critique their small group
ent languages, and were located in 21 of the 24 time zones (excluding leader’s example lesson and to collaboratively discuss strategies
three around the international date line). 8% of the section lead- for approaching their first sections. In the final, required training
ers were current computer science teachers, 39% were university workshop, we used scenario-based training to have section leaders
students, and 49% worked in industry as programmers. Applicants tackle potential situations they might encounter when teaching
represented a diverse spread of institutions, including MIT, Stanford, globally diverse students.
and Harvard (US), Oxford (UK), Koç University (Turkey), as well as To cater to to diverse teaching backgrounds and emphasize the
a variety of industry and academic institutions. Applicants were importance of an inclusive section culture, we chose to ground
surveyed about their reasons for applying to section lead (Table the training curriculum in collaborative discussions that built off
1). The most common response was a desire to give back through section leaders’ existing experiences. All section leaders had the
community service (86% of applicants) followed by a desired to ability to communicate and collaborate with one another via a
improve teaching skill (72%); applicants saw this as an opportunity discussion forum that only included teaching staff, and they used
to both contribute and learn. the platform to share teaching tips and ask one another questions
From the over 1,100 applications received, more than 80% of about the material for each week. To further elevate the diversity
applicants were determined to be above the required standard and of expertise among the section leaders, we also provided section
were offered a position. In total, 904 section leaders participated. leaders with the opportunity of volunteering to teach workshops
Training section leaders. While some section leaders had to their peers during “Section Leader Learning Week.”
prior teaching experience, many did not. As a result we developed
a succinct, community-focused training program for all our volun-
teers. Since human-centered learning was one of our central tenets,
we had to ensure that, on top of solidifying course content, sections 2.2 Student Selection
fostered a strong sense of community and belonging. In order to Application. To manage the student-to-section leader ratio, we
align section leaders to these values, we modeled section leader had students apply to take part in the course. Applicants had to fill
training directly after the structure of sections themselves (see Fig. out a short form, read about 5 pages of a textbook, and complete 3
short challenges based on these readings using an online IDE. These
1 The
challenges required no other prerequisite knowledge. The intention
application and evaluation periods were condensed due to various constraints at
the time. Such short application/evaluation periods are not intrinsic requirements nor of the reading and challenge questions were to give students a sense
recommendations for future iterations. of the content and time commitment of the course.
 The overall time to complete the application was approximately had their own discussion forum where they could ask teaching
1 hour and over 20,000 students completed a full application. When questions and discuss section issues. Students and section leader
evaluating applications, we looked to see if students (1) used the names were partially anonymized to preserve privacy and email
concepts in the reading and (2) copied and pasted the solutions. By addresses were never required to be shared.
requiring an application, we made sure that the students passed a Assignments. Students were given three large assignments
minimum bar of engagement. Due to certification requirements for which were based on the first three assignments of our university’s
interacting with minors, we limited the course to adults. CS1 course. Students could either complete their assignments in a
Students. Code in place accepted 10,428 students from around traditional offline IDE (PyCharm) or through an online interpreter
the world. Of those, 49.7% were women and 47.5% were men. We provided by EdStem. The assignments covered (1) an introduction
also had a wide distribution of ages: 15% were under 21, 9.7% were to programming using Karel The Robot (using Python syntax),
above 40 and the rest were in between. See Figure 1 for more details. (2) Python console programs which built up to a program named
KhansoleAcademy, which quizzes the user with randomly generated
arithmetic questions, and (3) an image processing assignment where
2.3 Course Components students implement image filters and a forest fire detector using
The course ran for five weeks from mid-April to late-May 2020. It pixel-level analysis of images. In each assignment, students were
was marketed simply as an experience to learn. Notably, we ex- encouraged to go beyond the minimal requirements and add their
plicitly stated that we would not provide certificates of completion. own creative extensions.
This latter point was actually a logistical requirement for our home Optional Diagnostic and Final Project. Beyond the three
institution. However, it also allowed us to set expectations that this required assignments, students were also given the option to take
class was about the value of learning programming, not a mech- a diagnostic assessment (i.e., test) mid-way through the course
anism to certification. The course included many components to to get a better sense of how well they were understanding the
keep students engaged. material. This assessment was auto-graded. Additionally, students
Course website. The central hub for the course course was who wanted a greater challenge at the end of the course were given
a website where students logged in and were presented with a guidelines for developing open-ended programming projects, with
personalized home page. Their page had a link to their section, as a number of suggestions provided from a standard list. Since the
well as course content such as lectures, worked examples, an online diagnostic and final project were purely optional, we don’t include
textbook [19], and relevant handouts. results based on engagement with those options in this paper.
Recorded lectures. To provide a cadence to the class, course lec-
ture videos were released three times a week (on Mondays, Wednes- 2.4 Course Context: COVID-19
days and Fridays). For each release, approximately 50-60 minutes
This course was built in response to an exceptional moment in
of lecture content was made available. This content was broken
history, when shelter in place mandates went into effect around
into 10-minute chunks to make it more digestible online. The con-
the world in response to the COVID-19 pandemic. As a result, this
tent was largely recordings of the live (online) CS1 class we were
was a time of turmoil for many people. In an early course survey,
teaching at our university with some slight edits. Overall there
students revealed the impact of this moment on their lives. 87%
was over 14 hours of content split between 74 videos. The con-
of students reported a significant life event taking place during
tent covered the basics of python including variables, control flow,
the course. Of those, 17% reported a change in employment and
data structures such as lists and dictionaries, as well as images and
16% reported a change in living situation. The recognition of such
graphics (including animation).
upheaval drove the tone for our class. From the first lecture we
Weekly sections. Once a week students would meet online in
set course values: Humanity, Intellectual Joy, Social Connection,
a group of ten with their designated section leader for a 40-minute
Gratitude, and "Everyone is Welcome." We kept these values visible
section. The section leader would lead an interactive learning expe-
on our course webpage. We wanted to instill in both our staff and
rience where the students would practice the concepts covered in
our students that we are a collective team working for one another.
class that week. Section leaders were provided with detailed lesson
While we presently turn to measuring more quantifiable impacts
plans both to lower their time commitment for preparation as well
of the course, we do note this unique context of the course creates
as to ensure more uniformity and high quality among sections.
a possible confound for the replicability of our results. As an expe-
Section leaders were also encouraged to share their reflections on
rience report, however, we primarily aim to highlight the results of
how their section went with one another to further develop their
the current class offering. Nevertheless, we do believe that many
own teaching skills. Students were assigned to sections first based
of the results reported here would in large part be applicable in
on time-zone preferences and second based on age similarity to
different course contexts as well. While future offerings may or
their section leader (to foster a "near peer" teaching model).
may not occur during the COVID-19 pandemic, the impact of this
Discussion Forum. A key component of the course was a series
moment will be felt far into the future.
of richly-featured discussion forums hosted by EdStem.org. The
course had a primary forum for all students and section leaders,
where students could ask questions that would be answered by the 3 IMPACT
staff and other students, and where staff would post announcements. Throughout the course we kept track of student engagement in
Each student was also in a smaller discussion forum with their assignments, lectures, and on the discussion forum. In addition,
section (10 other students and their section leader). Section leaders we offered an end-of-course survey to all students and section
100% Group Net Promoter (raw) (d) Course lectures
Having section leaders
Students +90.3 (9.7)
Course assignments
Helpful
80% Section Leaders (SL) +70.1 (9.2) Ed discussion forum components
SL: First Timer +76.3 (9.4) Live help sessions (students)
SL: in University +75.4 (9.5) 0% 20% 40% 60% 80% 100%
Percent of Participants

60%
(b)
Length Complete Num Num
Volunteers Teaching Course
40% Component Total Engagement (hours) Rate Students Complete
Students Watching Lectures
Video Watched 60,200 hours CIP Section Leaders 20h 99% 904 902
Students Posting in Discussion
20% Discussion Posts 94,600 posts CIP Students 60h 56% 10K 6K
Students Completing Assignments
Section Attendance 4,520 hours CS101 30h 27% 46K 12K
Week Raised for charity $15,000 Harvard 50x 150h 2% 2M* 40K**
0%
1 2 3 4 5
(a) (c) (e)

Figure 2: (a) Engagement over time (b) Would you recommend this experience? Where +70 is “exceptional" (c) Σ engagement (d)
percent of students who considered different course components helpful (e) comparison of completion rates to other courses
[10, 12]. CIP is Code in Place. *Harvard 50X enrollment is over an 8 year period. **This estimate is a projection.

leaders. We also chose a random subset of 100 section leaders and Section Leaders. Perhaps the most surprising result was the
250 students (including those who did not complete the course) to incredibly high completion rate among the volunteer section lead-
directly target with extra motivation and follow-up messaging. The ers. Out of 904 original section leaders only two dropped out: one
response from this random sample was high, with 89% of section because their life became unexpectedly busy and one because they
leaders and 92% of students responding. The high, but incomplete fell ill with COVID-19. This translates to a 99.8% completion rate
response leaves some room for sample bias. among the volunteer section leaders over the entirety of the course.
To make-up for section leaders who couldn’t complete the course,
other volunteer section leaders stepped in to teach multiple sec-
3.1 Student and Teacher Opinions tions. In preparation for more potential dropouts among section
Both students and section leaders had a high opinion of the experi- leaders we limited all section leaders to only teach one section (10
ence. The overall evaluation score students gave the courses was students) at the start.
measured at 𝜇 = 4.9, 𝜎 = 0.34 on a 5-point scale where 4 is “good" Students. Engagement in the course was surprisingly high. The
and 5 is “excellent". In addition we asked respondents whether they students (over 10,000 of them) watched over 60,200 hours of lecture
would recommend the experience to a prospective student or a (6.02 hours / student) and made 94,600 posts to the discussion
prospective section leader with a similar background to themselves, forum (an average of over 9 posts per student). For comparison,
on a scale of 0 to 10—the standard Net Promoter scale [24]. Students there were under 3,000 posts to Harvard CS50x’s forum during
raw recommendation value averaged 9.7, while section leaders aver- the same time period. We note that about one third of posts were
aged 9.2. This corresponded to a Net Promoter Score (NPS) of +90.3 from students own section discussion forums. Over 56% of the
for students and +70.1 for section leaders. Generally an NPS of +50 original class submitted all three assignments (our definition of
is deemed excellent, and anything over +70 is exceptional [1]. When completing the course). Given that approximately 80% of students
we break down the NPS between different subsets of section leaders were watching videos and posting in the discussion forum it seems
we find that first time teachers and university students enjoyed the that there were students who either were auditing or not submitting
experience the most (NPS of 76.3 and 75.4 respectively), section their work. Most of the students who didn’t complete the course
leaders from “industry" were similarly high (NPS = 72.6) and the didn’t submit the first assignment. In contrast, 75% of students who
NPS was lowest for professional computer science teachers (NPS = showed real engagement with the course by submitting the first
64.1). Students were asked which course components were helpful: assignment went on to also submit the last assignment. Figure 2 (a,
the highest rated components in order were lectures (93% of stu- c, e) provides more details on course engagement.
dents), section leaders (81%), assignments (79%), and the discussion Historical completion rates reported in similar MOOCs with a
forum (72%). This high rating of section leaders was commensurate comparable number of hours of instruction are generally low. Har-
with student responses when asked to rate the “overall quality of vard’s CS50x reports a 2% completion rate [10] while Stanford’s
my section leader", yielding 𝜇 = 4.5, 𝜎 = 0.8 where 4 is “good" and CS101 is 27% [12]. We note that comparisons of completion rates
5 is “excellent". See Figure 2(b). need to be interpreted cautiously. These other courses allowed for
unlimited enrollment, had much higher numbers of initial students,
and presumably had many students who were sampling. Further-
3.2 Completion and Engagement more, the total length of a course impacts completion rates. Figure
From the start, we were interested in making sure that both students 2(e) contains a full comparison of these courses with Code in Place.
and section leaders completed the Code in Place experience – this Of course, all these courses are contributing to the same goal of
was especially important to us since we needed to maintain an increasing exposure to programming.
appropriate ratio of section leaders to students.
3.3 Qualitative Impact numerous countries, all were able to bring inspiration and provide
In our experience teaching the class, we encountered many qualita- education. We believe there is opportunity to broaden participation
tive instances of impact and engagement in the course. Students not only among learners, but teachers as well.
self-reported that after Code in Place they were able to find employ- Section leading as a learning experience. Hands-on teach-
ment as software developers, become teachers, and find meaning ing exposure is an extremely valuable experience, both for improv-
during a difficult time. ing teaching ability and to solidify understanding of the course
On the discussion forum, we found students engaged in an ac- concepts. As such, Code in Place was as much a course for teachers
tive and uplifting community centered around learning. Examples as it was a course for students. By nurturing both students and
include the formation of meme threads to share computer science teachers simultaneously, we were able to create a powerful sym-
jokes, baked cookies in the shape of Karel the robot, and the emer- biotic relationship. A generally hard problem in scaling education
gence of self-organised volunteer student groups that contributed is how to train teachers – especially, how do we give more peo-
by translating and transcripting lectures for other students. The ple their first teaching experience. For many, their first teaching
forum also had an active student answering community that re- experience occurred as one of the 900 section leaders of Code in
sponded to other peers’ conceptual questions. Place.
Students expressed an interest in creating t-shirts as a way to Sense of responsibility drives completion rates. Why was
celebrate participation in the class. We decided to turn this into section leader completion so high? We believe that teaching instills
another venue for expressing our shared humanity. To that end, a notable sense of responsibility that holds people accountable. In-
we ran a crowd-sourced project for students to design and select terestingly, we note that section leaders, who had high completion
a class t-shirt, whose purchase would also result in a charitable rates, did not recommend the class to the exceptional level that stu-
contribution to an organization the students would select. Through dents did, although the latter had more dropout. To us this suggests
a vote, students chose to donate the t-shirt proceeds to https://www. that completing the commitment for section leaders was less about
buildon.org/. Over $15,000 was raised for the organization through “enjoyment" than it was about some function of productive “social
the sale of over 1,000 t-shirts. Perhaps even more heartwarming was pressure". For section leaders, dropping out would mean letting
a drive organized by some of the students in the class to purchase down 10 people for whom they were responsible. In the summer
t-shirts for other class participants who would have liked a shirt 2020 iteration of “CS Bridge" [20] (a CS1 course offered interna-
but were unable to afford one (especially those living in countries tionally), many of the ideas and curricula from Code in Place were
with low per capita income). This effort led to hundreds of t-shirts repeated (and many of the instructors were the same), but students
being purchased for others as a way for more fortunate students in were sourced from high schools and their parents were included in
the class to give thanks and support others. their admission. Section leaders offered office hours in addition to
As the course drew to a close, students expressed interest in section, and sections were held 12 times instead of 5. In this course,
follow-up resources or a next Code in Place. We created a master completion increased to 86% for a comparable set of material.
thread of different continuation resources and found this post to
be extremely popular amongst students. Students still post daily to 4.1 Limitations
the discussion forum, even months after completion of the course. Maintaining section size. For section leaders, any dropout in
students is demoralizing. As such, a 56% completion rate leaves
3.4 A Model for Sustainable Section Leading many sections feeling diminished by the end of the course. One
In this endeavour we set out to see if there was an opportunity to solution would be to start with more than 10 students per section,
jointly offer a teaching and learning experience. This initial iteration anticipating that some students will drop out. However, we believe
was limited by the number of section leaders who applied (50,000 the biggest opportunity to improve is to raise student retention.
students who applied vs. 1,300 section leaders), but dropout mostly Scaling class size. The choice to host a human-centric class was
affected only students. Is there an opportunity to find a balance a difficult one to make as it necessitated that we limit enrollment (in
between section leaders and students? First a few observations: (1) this case to roughly 10,000 students). Of the 50,000 applicants, the
Given that so few section leaders dropped out, there is less need second set of 10,000 looked almost identical to the 10,000 we chose
to have "emergency on call" section leaders which could allow for to teach. Making this cuttoff was difficult. In the future we would
section leaders to teach multiple sections. (2) A large number of like to offer an open-source companion set of materials for students
students both completed the course and said that they would like to who were not admitted to the class. In the first offering, there was
teach if we were to offer it again. In the post course survey, 34% of not enough time to provide such companion materials. Increasing
the respondents either agreed (16%) or strongly agreed (18%) that the number of section leaders by extending the application window
they would like to section lead in future iterations. This suggests to more than 4 days could also help address this issue.
the possibility of perpetually offering a class like this by recruiting
section leaders from previous iterations of the course. 4.2 Conclusion
Code in Place was an uplifting community service project. Through
4 DISCUSSION the experience we showed a proof of concept that many people
Every type of good teacher. In Code in Place we confirmed that who are traditionally not given the chance to teach are capable of
a truly broad set of people make for excellent instructors. With par- doing so. This demonstrates a model for scalable human-centric
ticipation from young university students to retirees, representing CS1 education, part of our mission of CS4All.
4.3 Acknowledgements [10] Andrew Ho, Justin Reich, Sergiy Nesterko, Daniel Seaton, Tommy Mullaney, Jim
Waldo, and Isaac Chuang. 2014. HarvardX and MITx: The first year of open
Code in place was special because of the incredible team that put on online courses, fall 2012-summer 2013. Ho, AD, Reich, J., Nesterko, S., Seaton, DT,
the project. It was a volunteer project, run by a large group of kind Mullaney, T., Waldo, J., & Chuang, I.(2014). HarvardX and MITx: The first year of
open online courses (HarvardX and MITx Working Paper No. 1) (2014).
individuals working towards a common cause. In addition to the [11] Mike Joy, Nathan Griffiths, and Russell Boyatt. 2005. The Boss Online Submission
section leaders, dozens of people volunteered their time to make and Assessment System. J. Educ. Resour. Comput. 5, 3 (sep 2005), 2–es. https:
the course work. We deeply appreciate all of their work, and are //doi.org/10.1145/1163405.1163407
[12] René F Kizilcec, Chris Piech, and Emily Schneider. 2013. Deconstructing disen-
thankful for the chance to get to work with such as special team. gagement: analyzing learner subpopulations in massive open online courses. In
We would like to thank: Lisa Einstein, Ana Saavedra, Yosefa Gilon, Proceedings of the third international conference on learning analytics and knowl-
Laura Mapstone, Brahm Capoor, Zach Birnholz, Jennifer Widom, edge. 170–179.
[13] Chinmay Kulkarni, Koh Pang Wei, Huy Le, Daniel Chia, Kathryn Papadopoulos,
John Mitchell, Kate Rydberg, Scott Maxwell and the EdStem team, Justin Cheng, Daphne Koller, and Scott R. Klemmer. 2013. Peer and Self Assess-
Richard Freling and the CodePost team, Isaac Pohl-Zaretsky, Jason ment in Massive Online Classes. ACM Trans. Comput.-Hum. Interact. 20, 6, Article
33 (Dec. 2013), 31 pages. https://doi.org/10.1145/2505057
Ford, and Nathan Dalal. This is a short list of the fantastic team: for [14] Jean Lave. 1996. Teaching, as learning, in practice. Mind, culture, and activity 3,
more details see the code in place website. 3 (1996), 149–164.
[15] Einstein Lisa. [n.d.]. Help Profs Teach Stanford’s Popular, and Now Free, Online
Intro to Coding Course. ([n. d.]).
REFERENCES [16] David J. Malan. 2016. This was CS50x. https://medium.com/@cs50/this-was-
[1] Swetha Amaresan. [n.d.]. What Is a Good Net Promoter Score? https://blog. cs50x-82be0995862b
hubspot.com/service/what-is-a-good-net-promoter-score [17] U.S. Bureau of Labor Statistics. 2017. Employment Projections. https://www.bls.
[2] Isaac Chuang and Andrew Ho. 2016. HarvardX and MITx: Four Years of Open gov/emp/tables.htm
Online Courses – Fall 2012-Summer 2016. SSRN (2016). http://dx.doi.org/10. [18] Eren Orbey. 2020. How Harvard’s Star Computer-Science Professor Built a
2139/ssrn.2889436 Distance-Learning Empire. The New Yorker (July 21 2020).
[3] Dan Davis, Ioana Jivet, René F Kizilcec, Guanliang Chen, Claudia Hauff, and [19] Chris Piech and Sami Abu-El-Haija. 2020. Human Languages in Source Code:
Geert-Jan Houben. 2017. Follow the successful crowd: raising MOOC completion Auto-Translation for Localized Instruction. In Proceedings of the Seventh ACM
rates through social comparison at scale. In Proceedings of the seventh international Conference on Learning@ Scale. 167–174.
learning analytics & knowledge conference. 454–463. [20] Chris Piech, Lisa Yan, Lisa Einstein, Ana Saavedra, Baris Bozkurt, Eliska Sestakova,
[4] Tawanna Dillahunt, Zengguang Wang, and Stephanie D Teasley. 2014. Democra- Ondrej Guth, and Nick McKeown. 2020. Co-Teaching Computer Science Across
tizing higher education: Exploring MOOC use among those who cannot afford Borders: Human-Centric Learning at Scale. In Proceedings of the Seventh ACM
a formal education. International Review of Research in Open and Distributed Conference on Learning@ Scale. 103–113.
Learning 15, 5 (2014), 177–196. [21] Stuart Reges, John Mcgrory, and Jeff Smith. 1988. The effective use of under-
[5] Jeffrey Forbes, David J. Malan, Heather Pon-Barry, Stuart Reges, and Mehran graduates to staff large introductory CS courses. ACM Sigcse Bulletin 20, 22–25.
Sahami. 2017. Scaling Introductory Courses Using Undergraduate Teaching Assis- https://doi.org/10.1145/52965.52971
tants. In Proceedings of the 2017 ACM SIGCSE Technical Symposium on Computer [22] Eric Roberts, John Lilly, and Bryan Rollins. 1995. Using undergraduates as
Science Education (SIGCSE ’17). 657–658. https://doi.org/10.1145/3017680.3017694 teaching assistants in introductory programming courses: An update on the
[6] Michael E. Goldberg. 2015. MOOCs and meetups together make for better Stanford experience. SIGCSE ’95, 48–52. https://doi.org/10.1145/199688.199716
learning. https://theconversation.com/moocs-and-meetups-together-make-for- [23] Laura S Tenenbaum, Margery K Anderson, Marti Jett, and Debra L Yourick. 2014.
better-learning-35891 An innovative near-peer mentoring model for undergraduate and secondary
[7] Philip J Guo. 2013. Online python tutor: embeddable web-based program visual- students: STEM focus. Innovative Higher Education 39, 5 (2014), 375–385.
ization for cs education. In Proceeding of the 44th ACM technical symposium on [24] Jenny Van Doorn, Katherine N Lemon, Vikas Mittal, Stephan Nass, Doreén
Computer science education. 579–584. Pick, Peter Pirner, and Peter C Verhoef. 2010. Customer engagement behavior:
[8] Philip J Guo. 2015. Codeopticon: Real-time, one-to-many human tutoring for Theoretical foundations and research directions. Journal of service research 13, 3
computer programming. In Proceedings of the 28th Annual ACM Symposium on (2010), 253–266.
User Interface Software & Technology. 599–608. [25] Erin L Woodhead, Preston Brown, Susan Snycerski, Sean Laraway, Nicholas
[9] John Hennessy. 2012. Online Education: The Coming Tsunami? CRA Snowbird Bathurst, Greg Feist, and Ronald F Rogers. 2017. An Examination of the Outcomes
Conference presentation. http://archive2.cra.org/uploads/documents/resources/ of a Brief and Innovative Partnership: SJSU and Udacity. Innovative Higher
snowbird2012_slides/hennesy.pdf Education 42, 5-6 (2017), 463–476.