A collection of post-mortems
Published on: August 19, 2018
We’ve just completed the 2018 Summer Institutes in Computational Social Science. The purpose of the Summer Institutes are to bring together graduate students, postdoctoral researchers, and beginning faculty interested in computational social science. The Summer Institutes are for both social scientists (broadly conceived) and data scientists (broadly conceived). In addition to the site at Duke University, which was organized by Chris Bail and Matt Salganik, there were also seven partner locations run by alumni of the 2017 Summer Institute:
The purpose of this post-mortem blog post is to describe a) what we did, b) what we think worked well, and c) what we will do differently next time. We hope that this document will be useful to other people organizing similar Summer Institutes, as well as people who are organizing partner locations for the 2019 Summer Institutes in Computational Social Science.
This post includes post-mortem reports from all of our locations in order to facilitate comparisons. As you will see, different sites did things differently, and think that this kind of customization was an important part of how we were successful. This post also includes a report from our traveling TA who visited five different sites.
You can read all reports or jump directly to the one that interests you the most: Duke University, Hunter College, New York University, Northwestern University, University of Cape Town, University of Colorado, University of Helsinki, University of Washington, and the traveling TA.
We’ve divided this post into five main sections: 1) outreach and application process; 2) pre-arrival; 3) first week; 4) second week; 5) post-departure.
Having a great pool of participants requires having a great pool of applicants. We took the following steps to spread the word about the Summer Institute:
As far as we can tell, this outreach worked pretty well, but next year, we’d like to reach out more directly to groups that include folks that are under-represented in our applicant pool. We discussed this issue with the participants of the 2018 Summer Institute at Duke who are in the process of creating a directory of experts in computational social science that will make special effort to identify those from under-represented groups who we might approach to help us advertise the 2019 Summer Institutes.
In 2017 we received applications over email, and in 2018 we switched to a system based on Google Forms. This was a huge improvement because it helped the process run more smoothly and it reduced the amount of administrative work for us. The form used as the Duke site in 2018 is available at this link.
During the application process, we received questions over email. The most common question from 2017 was about eligibility so we made that more clear in the 2018 call for applications and the number of eligibility-related questions decreased. The most common question in 2018 was about the possibility of submitting a co-authored writing sample. So, we’ve made that more clear in the 2019 call for applications. We also had a number of questions about whether the application was received and/or whether we have made decisions. In 2019, we’d like to have a system that confirms that applications had been received and one that ensures that all applicants entered their email address correctly. In terms of timing, we found that most applicants arrive right before the deadline.
In order to select applicants, both organizers read all the files and then independently produced a rank list of 30 applicants. The organizers then exchanged lists and had a phone call one week later. During the phone call, the organizers quickly discussed the candidates who were in both of the top 30 lists and the candidates who not in either top 30 list. Most of the time, however, was spent discussing applicants that were on the list of one organizer but not the other. When deciding which of these candidates to admit, we took into consideration not just each person individually, but also the composition of the participant population as a whole. This process is time consuming, but we are not planning any changes.
Finally, looking ahead to 2019, we’ve adjusted the language in the call for applications to encourage more applicants from a wider range of backgrounds.
After admissions, which ended in mid-March, we sent admitted participants materials to help them get ready for their arrival. This year we encouraged participants with less background in coding to complete some modules on DataCamp. We were able to make all the DataCamp materials available for free to all participants through the DataCamp for the Classroom program. We found that some participants did take advantage of DataCamp. One thing we can do better next year is to get the DataCamp subscription for participants sooner. Also, in our communication with participants, next year we plan to separate some of the intellectual pre-arrival activities, such DataCamp, from the logistical pre-arrival activities, such as information about housing.
When sending students access to DataCamp, we also created a Slack for participants to access the TAs during office hours. Next year, this office hours Slack should be called “Participants, SICSS 2019” and it should be used by all participants at all locations before, during and after the Summer Institute. We found that this single Slack helped encourage some cross-location interaction during the Summer Institute. As with this year, we also plan to have a separate Slack for all organizers in all locations. In general, however, we believe it is important to minimize the overall number of Slack workspaces associated with the Summer Institutes in order to minimize confusion about where conversations should take place (both among organizers and participants).
Another important pre-arrival activity we initiated this year was to have students submit photos and bios for the website. We did not do this in 2017, but it was suggested by several of our participants during the 2017 event. According to participants of the 2018 Summer Institute at Duke, this helped create community during the first few days. We will do this again in 2019, and we will encourage all partner locations to also post participant bios before everyone arrives (this year, most of the partner sites waited to add bios and pictures until the first few days of their events).
The teaching model that we use requires all participants to bring a laptop. This has not been a problem yet, but it is important to make this clear to participants before they arrive. Further, efforts should be made to create accommodations for participants that don’t have laptops. In our 2019 email to admitted students, we will note that students who do not have access to a laptop can request our help in arranging to borrow one during the Institute.
In terms of “behind the scenes” logistics, as much as possible should be done before participants arrive. This year we ordered t-shirts for people at all the locations. This was great for community building, but it was logistically difficult because we did not have access to the roster of participants at each institute and because shipping costs to several of our institutes were prohibitively expensive. In 2019, we plan to have all the gifts for participants at each partner location before the Summer Institute begins. We think that mugs might be a good choice because there are no issues with sizes.
As with 2017, in 2018 we used a professional technician for livestreaming and audio-visual services at Duke University. This was a bit expensive, but we think it was worth it given that so many people were watching lectures from the Duke event at partner locations. One improvement that we made in 2018, was having a dedicated laptop used for all presentations. This meant that all the organizers and TAs had access to their own laptops during the entire event. When necessary, slides or other materials were sent via Slack to this laptop.
Finally, in terms of physical space, the room that we used in 2018 worked well, but it lacked two things: a clock and whiteboards for participant work. In both 2017 and 2018, our room had a flat floor with configurable furniture. We recommend this kind of room for partner locations. A bowl classroom with fixed, individual seats is not ideal for group work.
In terms of dividing up the work among organizers, most of the pre-arrival work was modular and fell into one of these categories: Slack, DataCamp, getting participant bios on the website, A/V, housing, food, and visiting speakers.
First week: classes
Most days of the first week of the Summer Institute followed the same general pattern: lectures in the morning, lunch, group activity, feedback, visiting speaker, and then dinner. Many participants—and organizers—felt a bit overwhelmed with everything that was happening; someone described the Summer Institute as drinking from a firehose. But, we think that 2018 was smoother than 2017, and we think that 2019 could be even smoother.
The biggest question by far about this first week this year was how it would work with the partner locations. Fortunately, we think it worked very well.
We posted all slides and annotated code associated with each morning lecture to the website before teaching so that folks at the partner locations could follow along on the slides if they wished. We also had a seperate Slack channel to take questions from partner locations, and we had a TA who was responsible for monitoring this slack channel and asking these questions. This process worked well. We also posted all the materials needed for the group activity to the website, and we provided background information to the location organizers through our organizer Slack. Next year, it would be helpful if we could post the teaching materials and group exercises earlier so that they organizers of partner sites could adapt them to their local audiences as necessary or otherwise prepare for the exercises with their TAs.
There are a four main logistical complexities added by working with partner sites, but these are fairly minor. First, we needed to ensure that we had high quality A/V for the livestream. Second, we needed to ensure that the lecture videos are posted to YouTube very quickly for the partner locations in different timezones. For example, the partner locations in Cape Town, Helsinki, and Seattle all watched some of the videos on YouTube. Third, we needed to have a TA monitoring Slack to ask questions that come from participants at the partner sites. Fourth, we needed to be sure to stay on schedule. We also sought to avoid last minute changes to the schedule, though we did dedicate the first 15 minutes of each morning to discuss local logistics with participants (often these discussions involved issues related to accommodations or food).
After the morning lecture and lunch, students worked on an activity that was designed to practice the material covered in the lecture. Each day we randomly put participants into 5 different groups of about 6 people each. Participants enjoyed having a chance to work in mixed groups, and this provided folks a chance to work together some before the group research projects in week 2 (more on that below). For next year, we might slightly increase the number of groups, which would slightly decrease the size of each group. Also, we might do a bit more work to ensure that the groups are balanced.
About 30 minutes before the visiting speaker, we brought everyone back together for a debriefing about the activity. This debriefing was helpful for the participants to see common themes that emerged across multiple groups. The activities generally worked well, but we rarely had enough time to finish them. This was frustrating to participants, but it is hard to design open-ended activities that can be finished on a specific schedule. Next year, we will try to provide more context so that participants don’t expect to finish. Also, we can further shorten the lectures and start more of the activities before lunch.
After debriefing the group activities, participants did a quick keep-start-stop survey through Google Forms. This means that participants describe the activities they want us to keep, the new activities they want us to start, and the activities they are not enjoying. Then, the next morning (during the 15 minutes dedicated to logistics) we gave feedback on the feedback, reporting what was said, what we changed, and what we decided not to change. We think this feedback and feedback-on-the-feedback process was very helpful in the first few days. We will use it again next year, and we would strongly recommend that partner locations use it as well.
Most days, starting at 4pm a visiting speaker would present his or her research. During this time, participants at the partner locations could ask questions through Slack (again with one TA monitoring and speaking up). In addition to questions during the presentation, we also organized two other ways that the visiting speakers and participants could interact. First, during the afternoon before the presentation, participants could sign up to meet with the speaker for 30 minutes. We generally had more sign-ups than time spots, so participants would often meet with speakers in small groups. Second, we invited the speaker to join us for dinner afterwards. We found that both of these activities worked well. In terms of logistics, this year we had one TA who was in charge of the afternoon meetings: scheduling them and keeping them running smoothly. Have one person totally focused on facilitating these meetings was a big improvement over last year, when the meeting were a bit chaotic. One of the visiting speakers recommended that we consider having one-on-one meetings after the visiting speaker’s lectures so that the content of the talks could be discussed- though this would be nice, it introduces a number of logistical challenges related to travel of the visiting faculty members.
There are four other things that happened during the first week that we think helped build community, all of which we will do in the future. First, we had a very informal opening dinner that briefly introduced the Institute at Duke and prepared participants to introduce themselves to each other in a more formal manner during the following (first) morning. Second, during the week, we had participants write name cards that they put on their desk and we had them wear name tags. Third, participants from all the sites had access to a common Slack, and we found that during lectures they would share link and ask and answer questions there. We observed students from multiple partner institutes helping each other on the Slack channel on numerous occasions. Fourth, we started a system for discussion tables at meals. There were often topics that came up during the lecture or group activity debrief that participants want to continue talking about, so we decided that any participant could conveen a discussion table by writing the topic and their name on a white board near the door. We found that this process worked well logisticall and created more participant-driven learning, which is something we are trying to do more. Here’s a list of discussion tables that took place:
Second week: group projects
During the second week, participants worked on group research projects. The goal was that some of these projects would eventually turn into published research in computational social science. On Monday morning Chris led an intellectual speed dating process. This process involves first crowdosurcing a list of research interests, then having students place 1s or 0s next to the items that are of interest to them. We then run a cluster analysis to identify maximally similar and dissimilar groups. Students meet in each of these two types of groups for 30 minutes in order to come up with possible research topics. These topics are then added to a google spreadsheet and participants were asked in early Monday afternoon to write their name next to the project they would like to join. This process was smoother in 2018 than in 2017 in part because we did a better job of telling students about the group projects during the first week.
In order to create some moments of shared activity during the second, we also had flash talks by participants. These generally took place at lunch time and were 15 minutes long: 10 minutes to present and 5 minutes for Q & A. These flash talks were a mix of presentations about research and useful tools (e.g., open source software). A full list of the flash talks is provided on the Summer Institute schedule. The flash talks were a success and we plan to continue them in future years.
In order to support the group research projects, organizers and TAs would circulate between the groups offering advice. We also provided small grants to groups that needed them. This grant process was not very smooth this year, and we think participants spent more time thinking about it than was necessary or useful. Here’s a proposed system for next year. During the Summer Institute it should be very easy to get a micro grant (hundreds of dollars) to run a quick, proof-of-concept pilot. Then, participants could build toward a full grant proposal submitted a few weeks after the Summer Institute. To receive a micro grant, participants could submit a one paragraph description of what they want to do and why at lunch time each day. Then, the organizers should be able to provide an answer by dinner time, if not before. Participants can create a special email account for their project, and this email address can be used to create an account with the service provider (e.g., Amazon Mechanical Turk) and money can be deposited into this clean, new account. This system requires creating new accounts, but it avoids any accidental comingling of projects. We also plan to spend more time clarifying the scope and purpose of the group project during the final session on Saturday morning– in particular, we want to communicate that the participants are not expected to produce a polished research project by the end of the week, but rather a strong research proposal (or a proposal to create infrastructure of other tools that will help expand the field).
During the second week we also had two guest speakers present remotely. In terms of logistics, one of these went smoothly and the other had poor audio and video quality. For speakers presenting remotely in future years, we should try to get slides ahead of time, and we should test the entire set-up. This would help us find problems with the software or internet connection (or both). This year we tended to use free services such as Google Hangouts and Skype. In future years, we could try one of the many paid video conferencing services.
On Friday afternoon participants presented their group research projects right before the closing dinner. We think that participants benefited from seeing the presentations of the other groups, and it provided us a chance to ask some questions that connected the specific projects to the more general themes of the Summer Institute. Next year, we will begin these presentations slightly earlier in order to facilitate final group meetings where the participants can discuss the next steps for their projects while they are together in person.
The most common problem during the second week was that some of the group research projects would struggle to get started or would get stuck. Some of this should be expected when there are people from different fields trying to work quickly on a new project. But, we did have a few strategies for trying to get the groups moving forward. First, we tried to remind participants about disciplinary differences in research goals; some of being stuck was just people not understanding each other. Second, we tried to turn a disagreement into a research question. For example, if one participants want to use data from Yelp and one wanted to use data from Google Places, we suggested that they use both and compare the results. Third, we sometimes tried to lower the stakes. That is, some of the times that groups got stuck, it felt like some of the participants were viewing this as something with very high stakes. We were reminded everyone that they can just think about this as a learning experience, it seemed to help people try new things. Finally, if a group was stuck, we also encouraged them to consider switching teams. Overall, we think that a huge amount of learning happens during the group research projects, but everyone should remember that trying to do interdisciplinary group research projects in a week is difficult.
Finally, we would make one small logistics change next year: we don’t think it is a good idea to schedule a visiting speaker the night before the group research projects are due.
After everyone has left, there are a number of things that need to be done to archive what happened this year and to prepare for next year. These wrap-up activities include: archiving all slides, code, data, and videos; collecting evaluations from participants; reviewing and funding research grant proposal; thanking speakers; creating ways to stay in touch (e.g., Facebook groups); and planning a meet-up. Also, final book-keeping of expenses related to the event should be conducted promptly so as to come up with a final estimate of the amount of money available to participants for research via post-Institute grant proposals.
The Silberman School of Social Work, part of Hunter College at the City University of New York (CUNY) hosted a satellite SiCSS site from June 18th – July 29th, 2018. Student feedback indicates the institute was extremely successful in its first implementation, and there was a unanimous call for a bigger, better institute to be hosted in 2019. This post-mortem is organized chronologically, beginning with the application process and ending with student feedback. The final section offers the main organizer’s reflections concerning institute strengths and deltas.
Hunter SiCSS was organized as a domain specific site, and was further limited due to main organizer time constraints. As such, applications were open only to CUNY affiliated graduate students and faculty. We had 8 applicants (and an additional 3 queries from other prospective students) and accepted 8. A copy of the application can be found at the Hunter SiCSS site: https://compsocialscience.github.io/summer-institute/2018/hunter-nyc/apply.
General Event Organizing
The students were comprised of 5 current CUNY doctoral students (all Silberman affiliated), 1 junior faculty member, 1 senior faculty member, and 1 MSW student. CUNY does not have dormitories, thus we were unable to offering lodging to any students: however, all were already NYC/NY area residents and did not require lodging. We organized Breakfast and lunch for all days during week 1, and the first and last day of week 2. We asked folx if they required any additional assistance to realize their full participation and only came across one issue: one participant had an ancient computer that ultimately couldn’t perform some of the more intensive computations. We resolved it by loaning them a laptop during the morning lectures, and partnering the participant up with others during the afternoon exercises.
Nearly all participants had little to no exposure to programming beforehand. As such, we structured our week 1 days a bit differently than main site SiCSS. Afternoons during days 1-3 were focused on learning basic R interfacing: understanding where the idea for R came from, Rstudio, version control with git/gitub/rstudio projects, and doing basic analysis with R. Students reported this was a very useful part of the institute because it gave them a solid foundation from which to dive into the Day 3 group activity and beyond.
Because of the nature of commuting in NYC, we adjusted our schedule so that we 1) did not meet during the weekend and 2) met via Slack during Tuesday-Thursday of week 2. Students reported this schedule to be most helpful, since some commuted from as far away as Baltimore.
Though the application asked participants to commit to all mandatory institute days, and made the three middle days of week 2 optional, we still had 3 students who did not participate in the final day of presentations. This attrition was due to work and school obligations. I don’t know that there is a direct way to address this, as I note other sites had similar issues. Perhaps it may be best to accept more students to account for a small percentage dropping out?
Final projects were predominantly individual, though there was 1 two-person group. Importantly, we had an intergenerational presentation day with 2 children of students (ages 5 months and 15 years old) joining us on the last day. The 15 year old actually served as his dad’s TA and gave a small portion of the final presentation – it was pretty cool.
Main Organizer Reflection
Below is a summary of the strengths and deltas of the Hunter SiCSS site form the main organizer’s perspective.
Students noted that having a roomful of social works tackling an area of inquiry that excited them but currently has little exposure within the field was immensely inspiring and affirming for their own desire to pursue these methods and data within their own scholarship. My main goal for this institute was to begin to build a body of computational social worker scientists, and that seems to have begun taking shape.
Small Group Size
Though the number of students was small, the size was useful in two ways. One, as a junior faculty member, I wanted to experience running the institute with a smaller pool to understand what the challenges and opportunities were and in order to not take on too much too fast. Second, students reported a quicker sense of camaraderie amongst each other and thus were better able to learn from each other during the group exercises.
Focus on the Basics
Recognizing the limited exposure to programming, it seemed appropriate to provide incentive and justification for the use of R/Rstudio (to say nothing of Python) for research and, in the end, was a key piece of the institute that students reported appreciating the most.
Each day ended with a 30 minute debrief in which participants described what went well, what was challenging, and how they wanted to carry over themes into the next day. As a result, each day was altered slightly to accommodate student demand and this worked well to maximize learning.
There was uneven uptake of the Datacamp lessons, despite multiple email reminders. Student feedback was that the emails used the term “suggested’ and future institutes should be clear that these are mandatory. Further, it was unclear that main site TAs were available to support students from other sites during the lead up to the institute with programming. Having clear main site TA availability denoted in a more prominent way and making clear DataCamp is required for all novice programmers would likely go a long way toward increasing the use of Datacamp for novice programmers prior to SiCSS.
Students reported using Slack during week 1 was cumbersome, generally as a result of having to split attention between the computer and the lecture. There was no exposure to Slack before the institute, and I wonder if there is a way to present it earlier in the application/acceptance process to help gain some facility.
While inviting participants only from social work was a strength, it’s also something I’d like to change for two reasons. One, student feedback indicated that having more computer science/programming expertise beyond the TAs and main organizer would have been extremely beneficial. Second, during the institute Matt summarized Sociology as the study of groups of people. Social Work is the study what happens when people are excluded from the group AND mitigating the impacts of that exclusion for BOTH groups. Social Work is applied sociology, yet this is little understood or described. Thus, the second reason I’d like to expand the applicant pool is to increase exposure for social work scholarship and theory as a matter of import for creating social science for social good.
Unfortunately, we didn’t have anyone who wanted to engage with the Mturk group exercise, but I believe I should have made it mandatory, if nothing else than to show what is possible and not with it. Social Work is a population specific field of inquiry, and students felt that Mturk may not be useful in social work inquiry, so I’ll need to think about ways to argue for its (albeit limited) use.
The application didn’t go live until early-mid May, which was not ideal. February/March would have been preferable to garner more applications.
I declare the Summer Institute for Computational Social Science at NYU a great success. To be sure, the week was not without its hiccups. There are many things that could have been done to improve it. Next year will no doubt be better. But none of my worst fears were realized, and the feedback from attendees was overwhelmingly positive. Below I summarize the week’s events, what went well, and what could be done better next time around.
About 20 graduate students and a couple of postdocs attended SICSS 2018 for the first week of lectures and activities. We met in the main seminar room of the NYU Sociology department from 9:00am to 5:30pm. The mornings were devoted to lectures by Matt Salganik and Chris Bail, beamed live from Duke by projector; the afternoons were devoted to group activities building on the morning’s material, conceived by SICSS HQ at Duke and supervised by Barum Park and myself, Adaner Usmani (and on Monday and Tuesday by Taylor Brown, the global TA).
We had planned to continue the institute remotely, during Week 2. Indeed, on the final day of the first week, participants organized themselves into five groups each working on different projects. They expressed considerable enthusiasm about making use of the funds allotted to these research projects. The plan was to work via Slack and then share some kind of deliverable with me (and SICSS HQ) by the end of the second week. Unfortunately, not all of this energy came to fruition. One group made progress, but they did not spend anything like the allotted amount of money. Other groups have plans to spend more of their allotted money in the near future, all of which should be available to them through NYU for a while. But the second week lacked the coherence and energy of the first week, which was unlike the experience that I had at SICSS 2017. Below I reflect on why this happened.
What Went Well
My main goal in conducting the partner location at NYU Sociology was to expose graduate students who would not otherwise get any training in Computational Social Science. The goal, to be clear, was not exactly to train these students. In my view, that takes more time and effort than one week would ever permit. Rather, it was to introduce them to new methods and ways to learn these new methods that will, at some point, stand them in better stead. Put another way, the real purpose of the institute was make unknown unknowns into knowable knowns. In this aim, the week totally exceeded my expectations.
When I conceived doing something like this last summer, my target audience was 5-10 NYU Sociology students. As it happened, these 5-10 students attended, but so did 15 people from other disciplines and institutions who were similarly keen to learn what their grad programs were not yet able to teach them. Note that all of these people were largely recruited despite my advertising efforts, not because of them — because I anticipated constraints on space, I did not advertise beyond the NYU Sociology department. My sense, then, is that there is enormous demand for what SICSS NYU did. The fact that we served this demand competently and for a large number of people makes me proud to have been associated with the inaugural NYU satellite.
What else went well? The lectures were excellent — not just the material, but also the logistics involved in screening them. The livestream was always of excellent quality and the SICSS-wide Slack enabled participants at NYU to ask the speakers and each other questions. In their evaluations all participants remarked on the quality of the lectures. Similarly, the afternoon activities were very useful: they were digestible, well-conceived, and exciting. Every afternoon, the NYU Sociology conference room was buzzing with activity, which was very fun to watch.
One of the dividends of attending the inaugural SICSS at Princeton was that I met (and remain friends with) a number of people with whom I share methodological and substantive interests. In my estimation, SICSS NYU has done the same for those that attended. I was not sure that this was going to be the case, on Day 1. I had not established any kind of barrier to entry except affiliation to an academic institution, which meant that I did not curate the pool of attendees at all. I worried that the interests of those attending were too diverse and their skill-sets too incommensurate. But I no longer worried about this on Day 6. In the final analysis, SICSS NYU put a lot of interesting, motivated social scientists in touch with a lot of other interesting and motivated social scientists. This camaraderie was greatly facilitated by the group activities .
 Two things we did well that improved the group exercises, in my view: (1) we surveyed attendees about their prior R experience, and stratified the randomization process based on these categories; (2) we varied the size of groups from day-to-day. I would encourage future organizers to follow suit.
What Could Be Better
Of course, while there were many successes, there were invariably many challenges. Many of these admit of easy improvement. Here I discuss the main things that might be done differently, in the future. These are roughly ordered from least to most difficult to implement.
There were a handful of logistical snafu’s — concentrated on the first day, but peppered throughout the first week — that are probably unavoidable parts of the inaugural edition of something like this, but which are easily addressed.
First, at 9am on the first morning, it had not occurred to us that attendees would need Wifi. NYU makes it difficult but possible for guests to access Wifi, so this took 15 minutes to organize. Future organizers should remember that everyone needs Wifi access.
Second, by 12pm on the first day, it became clear that we had 25 laptops in the conference room but not space for 25 power chargers. This was easily remedied by the next day with a trip to CVS, but it would be better to plan for this in advance.
While the livestream from the main institute was crystal-clear, our own livestreaming capabilities were remedial. Being in NYC, we had access to an excellent list of guest speakers. But other institutes were not able to profit, as our video was grainy the audio terrible. This was a disappointment, but one that I did not have the bandwidth to address during the institute itself. Next time I would suggest that future organizers spend some of the grant money on livestreaming equipment. Even minimal investment ($250-$500) would yield enormous rewards in image and sound quality.
The main challenge in the group exercises came from the distribution of attendee experience. Two facts about this distribution were challenging.
First, there was wide unevenness in prior R experience. We tried to address this problem by stratifying the groups and making sure that there was at least one person with significant R competence in each group . But invariably the unevenness of R abilities meant that group dynamics were a function of the compassion shown by the most competent coder — if the most competent coder was impatient or more interested in learning the skill themselves than teaching it (which we tried to discourage but could not effectively police), the group exercise was not useful for those without much coding experience.
Second, the median of the distribution of abilities was low — lower than I had anticipated (lower than the distribution at SICSS Princeton 2017). This meant that most groups made little progress in these group exercises.
In my view, there are two ways to address this problem.
First, we could tailor the group exercises to different ability levels. Those without much coding experience might learn much more if they worked alongside those of similar abilities on a less difficult coding task than if they are stuck watching over the shoulders of a not-so-compassionate, experienced R user.
Granted, this would probably not work for those who have zero R experience whatsoever. Given this, I think it is imperative that future organizers also try to compress and shift upwards the distribution of attendee skills. This could be done by making the DataCamp courses mandatory or even just strongly emphasizing them in pre-institute communication. I regret not doing this.
 We were also fortunate that Barum Park, our superb teaching assistant, was an excellent user of R and was able to go from group-to-group to help.
Attrition After Week 1
Week 1 went very well, but Week 2 less so. Throughout Week 1, I reminded attendees that they had $2500 in research funds to use. On the last day, they formed five different research groups, each planning to work remotely via Slack. I asked that by the end of the second week they submit some kind of short deliverable (presentation slides, a description of the research they did, etc.). Yet as of my drafting this report (around 10 days after the end of the second week), only one group has submitted a summary of their research (and they spent only $50). I guess that up to $500-$1000 more of this money will probably be spent, but it is very unlikely that we will use the whole $2500.
Why did the second week lack the focus and coherence of the first? The principal reason is that we did not hold the second week on-site. When I first conceived SICSS NYU, I thought that the satellites were only going to be one week long, so I had planned to fly out of NYC at the end of the first week. As it turned out, I am aware that some of the other satellites met into the second week with great success. Thus, if future organizers are keen to see attendees work on research projects, I would urge them to budget the full two weeks and to ask their participants to commit to being on-site for both weeks. In retrospect, it was too difficult to ask attendees to coordinate research projects over Slack in just one short week.
In my view, this experience raises some questions about the structure of and funding for the satellites. There were several students at SICSS NYU 2018 who probably would not have attended if I had asked that they commit to being at NYU for two consecutive weeks. How should we deal with this class of applicants? One possibility would be to admit only those who can stay on-site for two weeks. Another possibility would be to admit two kinds of students: those who stay for one week, and those who stay for two. Only the latter set would be eligible to apply for research funds through SICSS. This is the model I would prefer. A third possibility would be to have some satellite institutes that meet just for one week, and others that meet for two. In any case, I do not believe it makes sense to transfer research funds from SICSS HQ to those institutes that meet on-site for just one week, as our experience suggests that much of this money will go to waste. Much of the SICSS money that was transferred to SICSS NYU 2018 for this purpose lies unused.
Attrition Before Week 1
While 20 people attended SICSS NYU for the full week, I had expected (and made plans for) about 30 people. About half of these people pulled out in the weeks prior to the start-date, and about half in the few days leading up to the institute. Had I known the exact numbers earlier, we could have saved some money on catering. Future institutes may be able to put this money to use, so it is worth considering whether this kind of attrition can be stemmed.
My guess is that this attrition was a result of the fact that participants who signed up for SICSS NYU did not have to offer anything beyond a verbal commitment of their time (unlike some other satellites, we had no formal application). While I emphasized that they were expected to attend if they signed up, there was no way for me to enforce this rule. If my guess is right, future organizers might consider increasing the costs of applying by requiring a formal application . I would expect the experience of submitting a formal application to reduce this kind of attrition.
 One could still accept all applicants, of course, thereby preserving the accessibility of the SICSS partner locations.
Guest lectures at SICSS NYU 2018 went well, overall, but there was not enough communication with the guest speakers about the content of the summer institute and the background of our audience. This was particularly an issue at SICSS NYU because we had our most advanced lecture on the first afternoon (Monday), before many participants had even been introduced to CSS. It may not be easy to address this, since speaker availability greatly governs the timing of guest lectures. But at the very least, I would recommend not scheduling a speaker on Monday, and trying to build from the simple to the complex — if possible.
About half of the respondents found the day’s schedule too packed, even though we let participants go by 5pm (i.e., we didn’t have group dinners, as we did at SICSS Princeton 2017). This may be because lectures were live streamed. It is tiring to listen to lectures in-person, no matter how excellent they are. But I venture that it is even more tiring to listen to lectures over a live stream. I do not have a good explanation for it, but I do have small-sample evidence that it might be true .
If this is true, there are two things that future organizers might consider doing. First, we could cut the number of guest lectures. By 4pm, attendees were tired, but they were not exhausted, as they were by 5:30pm. Simply reducing the amount of time spent sitting and listening to lectures might be sufficient.
Second, more ambitiously, SICSS HQ might consider abridging some of the lectures, as was suggested by some respondents to our post-institute survey. Particularly now that past lectures are all archived, attendees of SICSS 2019 could be required to watch some of the past lectures before attending, thus freeing up time during the week.
 That is, I refer to a comparison of my estimate of attendees’ tiredness during SICSS 2017 and my estimate of attendees’ tiredness during SICSS NYU 2018. Of course, there are confounders. The most significant confounder is that SICSS NYU 2018 was held in a poorly-ventilated windowless room, whereas SICSS 2017 had windows and much better ventilation.
This post-mortem uses the “strengths and challenges” approach to feedback. Strengths reflect practices that worked well, and challenges reflect practices that can be improved. For both strengths and challenges, associated strategies are brainstormed to indicate how the strength can be harnessed or the challenge can be addressed.
These points are not listed in any particular order.
The Summer Institute in Computational Social Science (SICSS) held a two-week training workshop from June 18 to June 29 at the University of Cape Town, run by Dr. Vissého Adjiwanou, Senior Lecturer in Demography and Quantitative Methods at the Center for Actuarial Research (CARe) at the University of Cape Town (UCT)  and the chair of the Scientific Panel on Computational Social Science at the Union for African Population Studies (UAPS). This year’s Summer Institute is an institutional collaboration with the main site at the Duke University and six other sites in the US and Finland. The Cape Town workshop has provided African scholars and students from various disciplines the opportunity to learn new quantitative methodologies related to digital trace data and to hone their research capabilities.
The Summer Institute in Cape Town has adopted a dual teaching approach which combined livestream lectures from Duke University with onsite presentations. During the two weeks of the Summer Institute, several lectures and talks were given by experts in the field of Computational Social Science. At the opening, Vissého Adjiwanou presented the importance of the Summer Institute for African scholars and led discussions on issues about estimating about causality in Social Science, and considered how new methods and data can contribute to a better understanding of social phenomena in sub-Saharan Africa. The lectures given at the SICSS then covered topics ranging from ethics, surveys and experiments in the digital age, by Matthew Salganik, to digital trace data collection and analysis (text analysis, topic modeling and network analysis) by Chris Bail. 2 by livestream from Duke University. Due to the six-hour time difference between Cape Town and Duke University, participants in Cape Town were unable to listen directly to the guest speakers at Duke University but watched recorded versions. The videos of these talks are freely available online to participants.
Onsite in Cape Town, the participants attended various lectures and talks on diverse topics. Nick Feamster delivered a lecture on machine learning and its application to detect ”fake news” from the web. Marshini Chetty gave a talk on ethics and on human-computer interaction. Tom Moultrie (director of CARe) introduced the participants to the opportunities provided by Computational Social Science for monitoring the Sustainable Development Goals and presented some challenges with working with digital trace data. Marivate Vukosi from The Council for Scientific and Industrial Research (CSIR) spoke of how data science should be used for public good. He presented various research projects he conducts in the field from road security to safety and crime incident detection on social media. The Summer Institute has also benefited from the participation of Kim Ingle from the Southern Africa Labor and Development research Unit (SALDRU) at UCT who introduced the National Income Dynamics Study (NIDS) to the participants. Finally, by Skype, Ridhi Kashyap, associate professor of social demography at the University of Oxford, presented her work on sex selection in India using Facebook Data. These various and highly qualitative research presentation made the Summer Institute one of its kind in sub-Saharan Africa, where participants learned from specialists from Africa and around the world.
In the second week, the participants worked in five small groups on various applied project, and they presented their results on the last day of the Institute. The first group worked to map research on gender based violence (GBV) from what lay people think about the issue by conducting text analysis from research articles and Twitter users. The second group project focused on measuring depression and stress-related issues by analyzing tweets in South Africa. They considered three sets of people; those who tweet about depression and have experienced it (Treatment Group), those who just tweet about it (Control Group 1) and those who didn’t tweet about it (Control group 2). The prevalence of depression among control group 1 was then predicted by comparing their characteristics with those of the treatment group. The third group worked to understand the environment of published news in South Africa. To do this, they analyzed publishers’ social Twitter network, and tweets from a topic and network analysis perspective. Finally, two other groups took an usual social science research approach to working on divorce and school dropouts in South Africa, based on panel data from the NIDS. They combined traditional statistical approaches with machine learning to better understand and predict these outcomes. These research projects should continue after the end of the workshop, and we hope to publish their findings in a special edition of the African Population Studies journal.
For many participants, this summer institute constituted both their first use of the R Software and their introduction to digital trace data methodologies. This proved to be a challenge for several participants, and for next year’s Summer Institute workshop, we plan to notify the participants much earlier of their admission, so as to permit them to get up to speed prior to the workshop by taking recommended online courses. We hope also that several participants will start using R in their own courses, making it easier for their best students to enroll in next year’s Summer Institute workshop. This summer institute has been a real opportunity for establishing new collaborations for many participants. The group project work developed during the second week proved to be an extraordinary and successful experience. There are no other opportunities offered elsewhere on the continent on the analysis of digital trace data. The success of the Summer Institute is obvious as the unsolicited comments we received and the final evaluation of the training. About 85% of the participants responded to that survey that they will definitely advise others to apply for the Summer Institute next year. Almost all of the participants also expressed their intention to make use in future of the methodologies learned during the Summer Institute.
A final source of satisfaction is to bring many scholars from outside Cape Town and to give opportunities to Master and PhD students together to attend the Summer Institute. Half of the 26 participants of the Summer Institute came from outside Cape Town, with foreign participants coming from 9 countries in both French-speaking countries (Benin, Burkina Faso and Togo) and English-speaking countries (Ghana, Malawi, Nigeria, Swaziland, Uganda, UK). This was made possible in large part by financial support provided by the International Union for the Scientific Study of Population (IUSSP). In addition, as the participants came from diverse disciplines, the workshop allowed participants the opportunity to exchange with others on issues of common interest and thus to break down disciplinary boundaries.
 Vissého Adjiwanou is now Professor (Professeur régulier) at the Université de Québec à Montréal (UQAM) and appointed as an Adjunct Senior Lecturer at the University of Cape Town.
The Summer Institute in Computational Social Science at the University of Colorado, Boulder ran from August 13th through 17th, and was organized by Brian Keegan and Allie Morgan. In addition to our sponsorship from the Russell Sage and Sloan foundations, we received support from the Center to Advance Research and Training in the Social Sciences (CARTSS), Institute of Behavioral Science (IBS), and Center for Research Data and Digital Scholarship (CRDDS). Our satellite operated later than the six other satellites, which gave us the opportunity to build and slightly adapt the materials presented at the primary SICSS. However, we remained very much in keeping with the five overall goals of the program:
(1 & 2) “Challenge disciplinary divides” & “Reach a broad audience”
Nineteen of our thirty-one participants were from the University of Colorado, with twelve coming from universities across the U.S. and Canada, and over 60% of our participants identifying as female. We spanned nineteen disciplines: political science, sociology, computer and information science, history, archaeology, criminology, nursing, education, and others. Presentations from our guest speakers touched on computational social science research across many of the fields mentioned above.
All of the materials for our satellite, including our guest speakers’ presentations, can be found on our website. Additionally, we live-streamed all of our lectures during the program. The availability of these materials was routinely positively mentioned on daily surveys administered throughout the program, and was ranked by some as the most important part of the program. In the future, like the primary SICSS site, we would also like to record our lectures for future playback.
(4) “Teach the teachers”
Last year’s SICSS inspired many alumni to bring the program to their institutions across the world. Through last year’s instruction, and with the commitments of Matt and Chris to open-source their teaching materials, Allie lead a technical lecture on web scraping, helped facilitate two network science group activities, and lead several group discussions. So with respect to teaching the teachers, last year’s program has been a great success. Among our participants at CU this year, five were early career faculty or postdocs, with the rest being PhD students at various stages in their programs. Time will tell whether our satellite will inspire participants to teach this material at their departments and institutions, but the overall positive response from our satellite is an encouraging sign.
(5) “Provide state-of-the-art training”
Lectures were led by local experts, and touched on details of computational social science methods, as well as applications to current research. The general response from our participants was that the lectures were very informative and provided new methods helpful to their research. After leaving SICSS, our participants expressed a strong desire to continue working in computational social science, and many indicated they would encourage others to apply to future versions.
In this short summary of our satellite, we now plan to discuss the strengths and weaknesses of our program in the format of our daily surveys (“keep”, “stop”, “start”), and highlight possible ways forward for future SICSS satellites.
Keep – What we did that is worth continuing.
Stop – What we did that is not worth continuing.
Start – What we did not do, but should in the future.
The Helsinki Context
The SICSS Helsinki partner site was organized in Helsinki, Finland. We organized it as a two-week institute: the first week was focused on lectures and second week on group projects. We had a total of 19 participants, an instructor and two TAs. Participants were both from Finland (University of Helsinki, Aalto University, Tampere University of Technology, Turku University) as well as from other European countries (Netherlands, Poland, Germany, Denmark) and rest of the world (India, Turkey). The overall net promoter score (based on the after-course evaluation) was 9 – indicating that the participants considered the course successful and would recommend it to their colleagues. Similarly, the textual feedback suggested that the course was found helpful, engaging and even fun.
We had made a deliberate decision at Helsinki to take the student group to an offsite location for the first week in Tvärminne. Beyond providing a fabulous venue in terms of scenery, outdoor activities and food, it also forced participants to have an on-campus experience for the first week, and socialize. Based on my random observations, this seemed to be happening during evenings, after the classes. Similarly, we chosen to organize after work activities during the second week, which aimed to help mitigating some of the challenges of lack of residential accommodation and social activities which emerges from that collocation.
What I found challenging and surprising was the need to facilitate group processes during the second week focused on group work. This was not extensive, and rather often groups seemed to manage on their own with this. We were rather pressed for time during the second week, but I’m thinking should we organize dailies similar to agile software production: each team member would speak for a few minutes what they are doing next, what have they done and if they have any major challenges. It might help us to intervene quicker in the group processes facilitation to potential problems and help the groups to manage the time and workload of such short project. Ideally, over the week, we should move the responsibility of organizing these dailies to the groups.
Something I was disappointed myself was lack of a “global” community of SICSS in the course Slack community. I believe that majority of comments from Helsinki to the public channel came from me or our TAs. Reasons for this may include time difference between Europe and US – but maybe also some failures to motivate and incentive this correctly by my side. Maybe next year, the Slack community management could consider establishing smaller channels for particular topics (“communities of practices”) to help people finding smaller venues where they can collaborate and share ideas and comments. Another aspect which I think might be helpful would be to start the online community building way before the event, also content wise. For example, maybe we could create for some of the pre-readings cross-national reading groups or other activities which would encourage them to speak with people from other communities?
Instruction and activities during first week
Something we spend a lot of time discussing with my TAs (in our debrief after SICSS) was the scope of the instruction. We covered many different things during the first week to provide a general overview of several different method families in computational social science. However, the question was if there should be more extensive discussion about some methods to ensure that students are able to fully understand them and not just rush through them. It was even proposed that we should really go through some basics of algorithms and computer science for the students.
However, I believe that the current idea of providing a rich overview of different methods is a good choice. It will familiarize students with many opportunities and help them to rethink social research. However, this is a communication challenge: this type of pedagogical choice needs to be explained and articulated. I did a few rounds of talks focused on this topic, but having that mentioned already on day one would be nice. (For example, one student commented in the feedback form that we could have a separate institute on any of these topics – which is true – and thus indicating that we should have been much more clear on the scope and idea behind that scope).
Second aspect which we discussed with my TAs and some participants explicitly commented: the course was a good crash course for people entering computational social science from a social science background, but for people coming from computer sciences and familiar with data science things many aspects of these lectures were rather boring and even useless. We tried to introduce to some activities (ad hoc) refocus on aspects such as teaching qualitative research methods to those not familiar with them or directing them to consider social science theory of their data analysis. These types of aspects could be more baked into the activities material next year (and I’m happy to help editing them towards this goal).
I will discuss video lectures later, but we had a mix of content coming from Helsinki (i.e., I was instructing the content) and some parts we choose to take as video lectures. Something I found difficult when instructing was to find the correct balance between audience participation and me lecturing topics. (To be honest, I actually don’t usually enjoy lecturing that much.) In some of my regular computational social science classes at University of Helsinki, I usually ask students to read a case study before the class which applies the approach or method we are learning. It serves two aspects: first, we can use that article to develop fruitful discussions with the students and therefore, I don’t need to lecture as much. Second and more importantly, the case studies provide the students an opportunity to see how “social science theory” (whatever that means) is integrated in the computational work – which I also believe to be a core skill in computational social sciences. If I’m organizing a class like this next year, I would integrate a component like this to balance the me-speaking – students-speaking a bit.
The video lectures
Overall, given the time difference organizing the video lectures was somewhat challenging. For the guest lectures (given at Duke in the evening) we opted to watch them delayed the next day. This reduced some aspects of their liveliness, which many students commented in somewhat negative tone. We tried to stop these lectures when possible to discuss the lectures in our group (a proposal by one of the students) and this seemed to make them somewhat more engaging. I would recommend similar idea to other locations which must follow lectures in non-live format.
On the instruction (given at Duke during morning), we chose some topics where we followed the Duke stream while on other topics I chosen to held the lectures myself. The student feedback suggested that they liked these locally provided lectures more than following live streams from Duke, so I think we made the right decision to develop some instruction on our own. Naturally the challenge with this is that the quality and content between the institutes may vary somewhat (for example, for text analysis, I chose to start the lecture by speaking about traditional qualitative research methods). However, I think that some quality bonus we had from organizing these locally – such as high level of interactivity and ability to react to local situations – was worth of this extra investment. In future years, I would examine to replace even more of these lectures when possible and to support on that, produce the materials early enough to allow discussions of them within the instructor community to find potential areas of improvements.
The project week
My only concerns which emerged from the project week related to the group dynamics and lack of proper theoretical reflections during them. I think our group creation process could have been clearer for people; we asked everyone to list topics they find interesting and in collaborative manner, mark which of them they might consider working on. Initial groups were formed based on that and even while I tried to encourage participants to not stay in what seemed like the local optimum, these groupings were set. I think next time, I would force people to change the groups and have similar discussions once or twice to show them the range of opportunities. Furthermore, facilitation of the project management as discussed above may have helped in this. Similarly, on the theoretical reflection, I did ask people to produce a mind map on the first day about theoretical concepts and literature and relationship between those and what they planned to do. Sadly, this itself did not seem to help students enough to engage in this thinking and follow that throughout the week. Again, scaffolding and facilitation may provide helpful in this.
While the first week facilities were excellent, the second week facilities were in our use daily only from 8am to 4pm. This limited some activities and influenced our scheduling. As the summer institute is during summer, many spaces at University of Helsinki just closed a bit too early. Next year, I would reconsider the second week location to have a few more hours of shared time. Also, I would have a clear single location for all non-Helsinki visitors and recommend they stay there to reduce some extra coordination efforts.
Negotiating interdisciplinary and cultural boundaries
Something I tried to bring you in fishbowl discussions was the interdisciplinary nature of computational social science and some of the challenges and problems related to this. Sadly, while I enjoyed these discussions, their intention may not have been as clear for participants as it was to me. I have been working across different academic communities for a long time and thus, rather familiar in interdisciplinary collaborations. They take time and often require a lot of flexibility and openness. However, I think that I understand the difficulties of these jumps as I’m so familiar with them and thus, didn’t provide as much support as one could have provided during the two weeks. For example, the fact that I rarely addressed traditional social science theories and methods in the instruction could have helped participants to follow the teaching more and made the classes more engaging to people from computer science background as well.
Similarly, the problem with teaching interdisciplinary groups is their internal heterogeneity in terms of skills. One solution worth of consideration could be to separate the group based on skills or provide even more modular learning activities, where we could assign different participants and groups slight variations of the same tasks to make them more engaging or to allow participants to enter zone of proximal development. This again would most likely push us to reconsider the role of lectures and instruction.
Finally, people did come from different cultures (not only country wise, but also academic cultures), which meant that their understandings of – among other things – research contributions, value of group work and ideas of good instruction differed. These were not something major challenges in the project. However, for me better managing this boundary work in the future is critical and having tools and approaches to facilitate students with these is necessarily. Sadly, I don’t yet have a clear and good solution for this problem.
The aim of this reflection has been to pinpoint potential areas for improvement, both for myself as well as other SICSS communities and their organization. Therefore, I have aimed to address challenges and problems and discuss them in extensive manner. However, as said in the beginning, most participants had a positive and engaging experience with this summer institute. The ideas and comments throughout this text may help to further improve the learning and clarify some of the difficulties observed.
We thank support from Russell Sage Foundation, Alfred P. Sloan Foundation and Helsinki Institute for Information Technology HIIT for their generous financial support.
The core of the Seattle partner location for SICSS, and the key to its success this year, was our set of 23 participants. These participants came from all stages in their academic careers: 2 Masters students (9%), 14 PhD students (61%), 2 postdocs (9%), and 5 faculty (22%). While many of the postdoctoral fellows and graduate students had some exposure to computational social science, we found a general trend: faculty often applied to SICSS-Seattle with less hands-on experience with computational methods but expressed great enthusiasm to gain exposure to computational methods, apply to research, and share with their students.
Our participants spanned a diverse range of disciplines, including sociology, political science, economics, communication, geography, archaeology, education, public health, transportation engineering, and linguistics. Throughout the week, participants expanded each other’s views of research through interdisciplinary conversations. Of the participants, 17 came from the University of Washington, while 6 came from other universities, primarily in the Pacific Northwest.
SICSS-Seattle participants brought a heterogeneous set of skills to the program as well. Some came with extensive coding experience in R and Python, some brought familiarity with novel data sources and other resources, and some drew on knowledge of critical theory and digital media studies. Participants shared their various skills through tutorials and discussion.
Finally, we had a great teaching assistant in Chuck Lanfear. He’s a PhD student in the Department of Sociology at the University of Washington, who teaches a popular Introduction to R for Social Scientists course. His R expertise came in handy for our group exercises.
The main week of SICSS-Seattle was hosted at the eScience Institute at the University of Washington, where we took over nearly half of the institute’s WRF Data Science Studio. We are grateful to have had their support — in addition to providing space, they handled the logistics of ordering and setting up food and coffee, managed our funding from the Sloan Foundation, and providing supplies needed.
For the optional second week, SICSS-Seattle participants interested in collaborating on projects were provided space at the University of Washington Center for Studies in Demography and Ecology.
A special thank you to Rachael Murray, Sarah Stone, Micaela Parker, and Robin Brooks from the UW eScience Institute, Sara Curran, Matt Dunbar, and Kristian Haapa-Aho from the UW CSDE, and our faculty sponsor Tyler McCormick for all of the help in making SICSS-Seattle a success.
Content shared with the main site
A large part of the content for the SICSS Seattle partner site was shared with the main Duke University site. In particular, we tuned in via livestream for lectures and guest speakers, and followed the main site’s group activities from the first week. However, these had to be compressed into a shorter amount of time — all Seattle activities and content were between 9-5PM Pacific time.
Since Seattle is on the west (and some would say, best) coast here in the US, we were behind the Duke University feed by three hours. The morning and afternoon lectures given by Matt and Chris were run on a delay for the Seattle site, while the guest speakers were (mostly) in real-time. Delayed streaming gave an advantage of skipping breaks and some content when trying to compress the content in our shorter programming time, while real-time video from guest speakers allowed us to send in questions to the speaker via Slack which were answered on video.
The group activities appeared to be a big hit at the Seattle site despite compressed activity time on some days. The participants were engaged, talkative, and eager to show their results at the end of each day. On one day, we had a hard time convincing any groups to even break for lunch! Many of our participants had prior programming experience, so most groups were autonomous and made great progress.
Seattle-specific lectures and activities
In addition to exclusive lectures on ethics and digital trace data, the Seattle site also held lightning talks from local participants and a networking opportunity with data scientists working on social good issues.
On Monday and Tuesday, we allocated time for in-person lectures to complement the main SICSS content. The first such lecture was on ethics and data violence from Anna Lauren Hoffmann, Assistant Professor at the UW iSchool; the second on the demographics of digital trace data from Nina Cesare, researcher at the Institute for Health Metrics and Evaluation at UW. The local speakers were a hit! We believe the opportunity to ask questions in person benefited our local participants greatly; consider promoting this to all partner locations in 2019.
Friday was the final day of lecture content for our site, where we saved time to schedule six 15-minute lightning talks for SICSS-Seattle participants. This was a great opportunity to hear a little about the research and open source software that participants have worked on. Perhaps it even spurred some of those presenters to advertise projects for the second week, as a number of those signing up for lightning talks later signed up for second week projects. (Though we know better than to infer causality, it could have been the more outgoing or prepared participants signing up for both.)
As luck would have it, the other half of the eScience Institute’s space hosted summer participants in UW’s annual Data Science for Social Good program, modeled after the larger program at the University of Chicago. All three projects at UW have strong computational social science components and the shared space provided opportunities for SICSS-Seattle participants to network and discuss.
Second week projects
The official program for the Seattle partner location ran for one week, but encouraged participants to collaborate on projects during the following week. Participants shared their project ideas on a Google spreadsheet, allowing others to read and choose to participate. This lead to at least two collaborations, both spearheaded by participants from the summer institute.
What could be better next time
While we were pleased with this year’s event, there are a number of things we could do better for next time based off of our experience in 2018 at the Seattle site.
While many had some experience with computational methods, we did get some feedback that a participant felt ill-prepared for the programming aspects of the workshop even after going through DataCamp. One option is to align timing with an in-person introduction to R, such as Software Carpentry or Data Carpentry. These are already offered at UW eScience quarterly.
Much of the project planning and collaboration development occured on Friday afternoon. It’s likely that this was not enough time for participants to discover common interests, form groups, and become sufficiently excited to entice them away from their normal duties to work on a shared project.
To facilitate collaborative projects, we can consider treating the second week as a requisite part of the SICSS-Seattle. With the second week optional, it was easy for participants to delay joining a project group until the beginning of the second week. By then, the pull from our normal duties is much stronger. We should note that requiring a two-week commitment would be in tension with accessibility.
We had some feedback that the month between sending acceptances and the start of SICSS placed major constraints on a small number of participants in arranging childcare, etc.
Overall, the Seattle partner location of the Summer Institute for Computational Social Science was a success this summer! In addition to our local support, we want to thank the Russell Sage Foundation and the Alfred P. Sloan Foundation as well as remote support from Matt Salganik, Chris Bail, and the main SICSS organizers and TAs on the east coast.
As the Traveling Teaching Assistant for SICSS in 2018, I visited five sites between June 18th and 30th. These sites were:
In this report, I aim to add my observation to the steps taken and setbacks experienced by partner institute organizers, and to collect a list of their suggestions for innovation and improvement going forward. I will also make recommendations for future Travelling Teaching Assistants. While the partner reports provide more detailed overview of experiences at each site, this report is unique in its ability to speak to the relative experiences across sites. The report is divided into sections according to the various tasks undertaken by partner institute organizers. Within each section, I provide a summary, notes on what to keep, and suggestions for how to improve.
Overall, the Summer Institute in Computational Social Science partner institutes ran very smoothly. All institutes reported near full attendance for the first week of the institute, and the primary educational resources afforded by the institutes—both local and streaming from the core site at Duke—were delivered without significant setback. Indeed, the central message of this report is that the level of educational instruction gained by participants at partner institutes was on par with that gained by participants at the core institute. I confirm this as both a previous Teaching Assistant for the core site at Princeton University in 2017, and as a three-day participant of the core site at Duke University in 2018.
By in large, the equality of instruction across institutes resulted from the near-flawless streaming of lectures from Duke to partner institutes, the hard work of partner institute organizers to provide quality in-house instruction, and the opportunity for all participants to engage in real-time with each other and with speakers via Slack.
This said, there are several differences and discrepancies that exist between the core institute and the partner institutes. The differences can generally be seen as trade-offs between the two formats, whereas the discrepancies present opportunities for improvement at either the core or the partner institutes. Many of these are discussed throughout the remainder of this report and in partner reports, but I begin by noting the three factors most apparently responsible for differences and discrepancies.
First, the uniquely short timeline provided between this year’s funding notice and the partner institute events made it difficult for partner institute organizers (henceforth PIOs) to attend to all of the necessary logistics. PIOs included graduate students and post-docs generally not experienced in organizing such events. While they did an excellent job, their short notice for planning resulted in small discrepancies between the core and partner institutes, mostly related to the presence of guest speakers and the deployment of group projects. These are discussed further below. Future SICSS already aim to lengthen the preparation time provided to PIOs, and a central purpose of this report is to diminish PIO learning curves by gleaning lessons from past experiences.
Second, while educational instruction at partner institutes was on par with that afforded at the core institute, the “study-abroad” or “summer camp” experience of being housed collectively on campus provided a unique social advantage to the core institute. In general, participants of partner institutes remain at home and commute to/from the institute each morning and evening. In one respect, this independence from the event is a benefit to participants with families or other local commitments that require them to remain involved in their normal lives. Moreover, given the locality of the partner institutes, it is not uncommon for participants to already know and work with one another. This makes social activity less necessary to foster relationships. However, in the majority of cases, partner institute participants did not know one another prior to the event. In these cases, the lack of social activity outside of formal SICSS hours seemed to diminish the strength of bonds formed between participants, the number of chance opportunities for informal instruction, and the investment of time and energy given to group exercises and projects at the end of the day. There seems a simple solution to this: more social activities at the partner institutes, as discussed in the “social” section below.
Third, a large advantage of the core institute (e.g. Duke) is the opportunity to develop a relationship with and gain personalized feedback from Chris Bail, Matt Salganik, and the cadre of guest speakers who attend the event—all of whom are leaders in the field of computational social science. It seems difficult to afford this same advantage to partner institutes, as even senior guest speakers are not available long enough to develop relationships. However, PIOs should be further encourage to provide such guest speakers, and perhaps also to invite them as group-project advisors. This not only gives participants a chance to meet senior scholars in person, but also builds the local community of computational social scholars.
In the remainder of the this report, I discuss other lessons learned, best practices, and opportunities for improvement, specifically with regard to the deployment of SICSS partner institutes.
Recruitment and applications
Partner institute organizers (PIOs) set up the application process in several ways. Seattle and Chicago both held competitive application processes that were open to all disciplines; Hunter limited applications to social workers at Hunter; NYU admitted participants on a first come, first serve basis up to the capacity of the facility they used for the institute. PIOs also advertised their respective institutes using a combination of social media, email listservs, and word-of-mouth.
The online recruitment process, social media advertising, and Google Forms application.
Several PIOs suggested that applications and acceptances be synchronized across institutes, so that information about changes to office hours, resources, etc. can be shared in real-time. There appeared to be confusion at several partner institutes with regard to which information was most up-to-date at a given point. Similarly, participants expressed that a centralized application website, where applicants can express a hierarchy of preferences for which institute to attend, could streamline the process. For example, interested parties could apply to the core institute and, if not accepted, be passed on to their top choice of partner institutes.
Some participants expressed that the application process was slightly confusing because it was unclear which institutes allowed for which types of participants. For instance, SICCS-Hunter College was only open to social workers at Hunter College, and several institutes did not accept international participants. Such information was not explicitly listed on application sites, however, so participants were required to contact each institute via email to find out whether they were eligible. Next year, PIOs should be clear on their application what the parameters of the institute will be.
There were two concerns expressed repeatedly by PIOs with regard to recruitment. The first is that the core institute has or may become so prestigious relative to the partner institutes that it will be hard for partner recruitment and retention. For instance, participants may rather wait to apply to the core institute again than go to a partner institute in a given year. The core institute is indeed a unique opportunity, so PIOs did not feel that applicants should be discouraged from trying to attend. If there is a way to discourage applicants from applying to a partner institute if they are not intent on attending, however, that may mollify this problem. The second concern express by PIOs regarding recruitment and application was that the computational social science community of scholars is not very large, so local regions may quickly exhaust their pool of applicants. This could be resolved by rotating partner locations to allow applicant pools to recover in off-years, and by not allowing for partner institutes in close proximity unless the population of applicants is sufficient to support them.
Although NYU used a first-come-first-serve application policy, the PIO expressed that this may have created less commitment from participants to attend. There were several participants who dropped out at the last minute, and there was thought that having a competitive process may result in more committed attendance. Part of the reason this is important is that the number of participants impacts other logistics of the event, such as daily catering and group project funds.
As partner institutes aim to foster a regional community of computational social scientists, they each had several participants traveling from a local but reasonably far distance. For example, at SICSS-Seattle two participants travelled from the University of Oregon and one traveled down from Vancouver; at SICSS-Chicago, several participants traveled from Chicago to Evanston, which takes about an hour or two on both ends of a long workday. PIOs expressed that it would be nice to have a small grant to provide housing and/or commute subsidies to a handful of such participants.
Finally, very few participants had children. At several institutes this was discussed and the conclusion was drawn that two-weeks is a lot of time for parents to commit to being away from their family. This is a strong argument for the benefit of partner institutes: that it allows parents to attend while staying close to home. To aid parent participants further, a clause could be appended to the application that addresses relevant issues (e.g. leaving early to pick up children from school, the availability of nursing rooms and campus daycare, etc). Informing participants of their acceptance to an institute earlier will also help parents prepare for the event. This year, one parent participant was informed too late to easily obtain childcare.
Overall, the daily instructional and work facilities at all partner institutes were on par with that of the core institute. In every case, facilities were on a university campus, with substantial seating, tables, and a large projector screen for live streaming and presentations. In most cases, there was ample natural light—which is important for participants sitting long hours for two weeks. PIOs should consider the layout of their facility space prior to the event, and adapt as necessary. After the first day, one institute decided to switch from a room with just chairs in rows, to a room with chairs at tables. In general, the best setup will be tables arranged in a u-shape facing the projector—as it allows all participants to see and engage with one another during lectures. Some element of facility that future PIOs should consider include:
If possible, refreshments should be served so as to encourage breaks to be an opportunity for participants to leave the room or even the building (particularly if the institute is housed in a highrise where fresh air is hard to access). Two weeks indoors from 8am to dusk can get difficult and frustrating, and a small commute from workspace to break space can be a good refresher, as well as a chance for participants to get to know each other instead of remaining at their computers checking email, etc.
PIOs could also be encouraged to do as the core institute and have topics of discussion for mealtime. This seems to act as a moment for continuing education where participants can draw on their own expertise to attend to topics relevant to computational social science but not on the curriculum of SICSS.
This year was an exception. Because of the late notice of funds from the granting institutes, PIOs were on a short timeline to prepare their event. With that said, the planning at each institute was generally smooth and several PIOs expressed that a second year of the institute would be even smoother, given the several lessons learned and the establishment of social networks for accomplishing various logistics. For instance, several PIOs noted that sorting out how to get money from Duke took a surprising amount of time, but that they now know what the process entails and could do it better next year.
According to one PIO, Bit-by-Bit is a great pre-conference introduction to the institute because it is unifying, in that it focuses attention on the principles of computational social science, rather than a particular discipline or coding language.
Slack office hours were helpful for those participants in need and aware of this resource. Participants with less coding experience should be strongly encouraged to utilize Datacamp and the pre-conference Slack TAs.
Almost none of the PIOs understood that they had Teaching Assistant resources at their disposal prior to the event. They also didn’t know about Slack office hours for Datacamp. These resources should be made more explicit next year. It may also be useful to develop a curriculum for the pre-conference Slack office hours. It appeared that many participants did not utilize these office hours because they either didn’t know about them or didn’t know what to ask about. If each office hour had a specific theme (perhaps following along with Datacamp) that could provide scope for questions.
Datacamp was a useful resource that seemed underutilized. For those partner institutes in which participants had significant coding background, Datacamp seemed almost irrelevant, while for those partner institutes with less coding background, Datacamp was necessary and perhaps could have been utilized even further. In the future, PIOs should strongly consider the needs of their participants and encourage Datacamp as necessary. PIOs should also make clear to participant what level of coding experience is expected for the institute to be beneficial. At three separate institutes, at least one participant was very discouraged by their lack of coding ability. One participant skipped a day because they felt too lost. PIOs expressed that perhaps the level of coding required to get the most out of the institute was not communicated clearly enough, and that Datacamp was too casually encouraged to be taken seriously for these participants.
Streaming was very smooth across all partner institutes. Likewise, all institutes had sufficiently large screens to accommodate an impressive streaming experience for participants. Importantly, it was also easy for institutes to pause or completely delay the stream. This was necessary for a handful of reasons. At NYU, for example, there was an unexpected fire drill. At Hunter College, streaming was often paused to help explain coding to beginners and to adapt lessons to the unique contexts of social work. At Seattle, streaming was often delayed due to time differences.
There does appear to be much that should be changed about the streaming process from Duke. PIOs and participants especially liked the opportunity to watch the Duke guest speakers live, and to ask questions via Slack.
While streaming from Duke was very smooth, only the partner institute in Chicago was able to set up streaming from their institute to all others. At least one organizer mentioned that they felt overwhelmed with so much else to do that they put guest speakers and live-streaming at a lower priority. In the future, with more time to prepare, PIOs should be strongly encouraged to setup streaming from their institutes. They should bear in mind that YouTube requires 24-hour notice to approve a live-streaming account. One institute relied on another’s YouTube account to live-stream because they were unable to broadcast a speaker due to this restriction.
As previously mentioned, several PIOs expressed that recruiting speakers took lower priority, given how much time they had to prepare for the event. They all felt that more lead time would have allowed them to set up guest speakers locally, and they all agreed that live, local speakers would help to build their regional community of computational social scientists.
Both speakers at the core institute, and speakers at the local institutes. Also, streaming of all speakers who give permission.
PIOs and partner participants agree that, while local speakers helped build their regional community, they did not like the idea of missing streams of prestigious speakers at the core institute. While it’s now clear that speakers need the flexibility to bounce around their schedules , it would be ideal to give PIOs a preliminary schedule of core institute guest speakers at least one month prior to the event. This would allow PIOs to plan their local speakers accordingly.
Given that the PIOs are often graduate students and junior faculty, it may be helpful to have a template invitation document for them to send to potential guest speakers, with Chris and Matt as co-signatures at the salutation line. This may help with speaker recruitment.
PIOs expressed that it was unclear whether there were honorariums available for local speakers. This should be communicated to PIOs prior to their recruiting guest speakers. And PIOs should be reminded to set up guest speaker one-on-one meetings for their participants.
There was significant engagement in the exercises at the partner institutes. At several, it was even difficult to get participants to regroup because they preferred to keep talking about the exercises. Essentially the same level of engagement was witnessed at partner institutes as in the core institute, aside from the few factors discussed below.
Exercises were generally well-received by the partner institutes, so the provision of exercise templates that can be adapted at each location should be maintained.
Because participants at the core institute plan to eat dinner together and sleep in the same location, their engagement in the exercises tends to sustain until the end of the day. By contrast, participants in the partner institutes begin to check out as the end of the day approaches, assumedly because they need to make plans for dinner, commute home (sometime a rather long distance), and attend to any other local engagements, etc. Finding ways to keep participants engaged until the end of the day is something that PIOs should prepare for in coming years.
Some PIOs expressed that it may be nice to provide a list of the programs that will be referenced in the institute exercises, so that participants can take a look if they are eager or able prior to the institute. Several participants were unfamiliar with mTurk, for example, and noted that the majority of their time in the mTurk exercise was spent figuring out the platform and not learning the principles (i.e. what’s a HIT?). Other programs that could provided in a list to participants prior to the event include: Google Forms, All Our Ideas, Slack, RStudio. It seems sufficient to just provide the list and let participants learn as they wish.
One partner institute explicitly mentioned that they would likely get rid of an exercise next year to allow for a catch-up day (i.e. “Pick one exercise from the past few days that you did not finish or that you wanted to spend more time on, and do that.”).
Not all partner institutes formally participated in the second week, which is group projects. Some PIOs felt overwhelmed by the planning of the first week and did not build sufficient awareness or excitement about the second week. In these cases, PIOs decided to conduct group projects very informally. Other PIOs felt that the second week was too much to ask of participants who are still going about their daily lives–compared to participants at the core institute, who have opted out of their daily lives for the duration of the two-weeks. Of those partner institutes that did participate in the second week, group projects appeared to be just as rewarding, ambitious, team-building, and productive as at the core institute.
Allow the second week to be optional for partner institutes.
SICSS-Chicago did a very good job of encouraging their groups to think not only about the technical/data-driven side of the group project, but also about the social theory motivating it. They asked participants to write small proposals to receive funds and were explicit that theory was part of the judgment criteria. Such a process should be held in the future.
PIOs should be encouraged to advertise–early and often–that the second week of the institute will be group projects, and that participants should think specifically about what resources grant money could be most productively applied to. This is not to suggest that money must be spent, but given the unique opportunity that it is for participants to easily access funds for research, it was unfortunate to see so many pass it by simply because they felt too pressed for time to think about spending responsibly. Perhaps if participants are on-boarded earlier onto Slack, they can share interests and begin to express project ideas even before the institute begins.
Across all sites, there was significant confusion with regard to Slack early on. This seemed to stem from three issues. First, the use of different Slack workspaces for pre-conference office hours and the event itself. Second, participants general inexperience with Slack. Third, the late-notice communication of how each partner institute Slack workspace/channel would be setup.
The use of Slack as the communication technology across all institutes. This program allows not only for messages, but file sharing, group formation, and live guest-speaker questions as well.
Do not have separate Slack workspaces for pre-conference office hours and the event itself.
Provide participants with a brief explanation of how Slack works prior to the event. Nothing detailed, as people generally catch on eventually by using the application. Introduction need only be pithy. For example:
For those of you unfamiliar with Slack, it is a way for groups to collaborate online. Follow the link below to join our SICSS “workspace.” We encourage you to download the desktop application as well. Once you join our workspace, you will see that there are already conversations going on and resources being shared across different “channels.” For example, in the #general channel, we have shared information about the schedule of the upcoming institute. You will also see a channel associated with you particular SICCS location (e.g. #duke). Please feel free to introduce yourself and get the conversation started! For the most part, you learn Slack by using it. However, if you’d like to take a tutorial, follow this link: .
This is somewhat of an unavoidable issue, but important to bear in mind: last-minute shifts in the schedule–while small at the core institute–can have larger consequences for partners, particularly if they are on a different time zone. These changes can interrupt breaks and cause food to go cold, for example.
With a combination of social activities and on-site accommodations, SICSS-Helsinki seems to have most closely approximated the social environment provided by the core institute. Other partner sites provided one-off social events that served as the primary opportunity for participants to get to know one another. SICSS-Seattle, for example, held a joint happy hour with the Data Science for Social Good community during the first week, and SICSS-Chicago held a closing picnic for their participants.
In the future, partner institutes should be encouraged to organize at least a few off-campus evening activities, daytime social events (e.g. tourist site seeing), or informative trips to CSS-related companies in the area. These should be planned early on in the event, as the friendships formed can improve discussion and debate during SICSS, as well as investment and buy-in for the second week of group projects. A common take-away from SICSS-2017 is that, in addition to the instruction gained, the event’s primary good over the long duration is the establishment of network between both participants and their respective institutions.
Across SICCS institutes, it was expressed that some form of networking would be useful. One idea mentioned was to gather a list of participants across all sites and match on interests for introduction via Slack or Skype prior to the event. This may further establish and encourage the benefits of the worldwide community of SICSS participants and alum. It may also foster more discussion–especially from more isolated communities–in Slack during the event.
Few partner institute appeared to have an established and consistent form of feedback from participants. In the future PIOs should be encourage to set up a feedback procedure early (e.g. the Google form used at the core institutes) and to ask for feedback at the end of each day. Next year it may also be useful to have a centralized exit survey for all participants across all sites, to get their feedback on what went well and how to improve the experience.
PIOs should also be encouraged to keep a record of the events of each day as they happen. This can be accomplished by end-of-day or start-of-day meetings between all organizers an TAs, where they can recap prior experience and review feedback forms from participants.
PIOs suggested that, instead of incurring the cost and effort of sending t-shirts to each partner institute, the core institute should simply send a template that can be printed on shirts or other paraphernalia locally.
Advice for future Traveling Teaching Assistants
I highly recommend that Traveling Teaching Assistants (TTAs) be experienced with flying and comfortable navigating new cities. If possible, it would be very useful for the TTA to have TSA pre-check, as it can save upwards of one hour at each airport visit. There were at least two airports in which TSA-precheck saved more than an hour of commute time, which allowed for more time to be spent at institutes, and also removed stress from an already full travel schedule over two weeks.
At a minimum, TTAs should plan to spend a day and a half at each site they visit, and should aim to attend at least two meals and several breaks with participants—as these are the best opportunities to speak learn, answer questions, and gain feedback.
In the weeks prior to their visit, TTAs should inform PIOs of their visit and set up one-on-one meetings to discuss how the institute is going. These meetings provided an invaluable, causal, real-time feedback mechanism that is likely to extract information that would otherwise be forgotten by the end of the event.
TTAs should consider whether a hotel or an Airbnb will be more comfortable during their travels. While hotels offer a level of formality that may ease logistics, Airbnbs are often cheaper, closer to campus, and feel more like a home during a long two weeks of travel alone. If the TTA chooses to stay in Airbnbs, it is strongly encouraged that they select only experienced, highly-rated Airbnbs, as newer accommodations are more likely to present problems related to checking in, internet, etc. Below is a record of the Airbnbs used by the TTA in 2018. Both were close to campus and comfortable, with reliable internet and easy check-in/check-out. They were also cheaper than local hotels:
Although many cities have good public transportation, it was much easier to take Lyft/Uber/etc directly between airports and accommodations. In general, travel between accommodations and institutes should be no more than a short walk or an easy public transit commute.
We hope that this blog post described a) what we did, b) what we think worked well, and c) what we will do differently next time. We also hope that this document will be useful to other people organizing similar Summer Institutes, as well as people who are organizing partner locations for the 2019 Summer Institutes in Computational Social Science. Finally, we wish the thank the main funders of the Summer Institutes in Computational Social Science: the Russell Sage Foundation and the Alfred P. Sloan Foundation.