Last year I was a member of the Data Science and Statistics club on campus. The group was project-based. At the beginning of the semester, any interested students could present a data science project idea they had, then the rest of the students could join whatever group they found interesting. Most of the projects had a scope of about a semester, and even completely inexperienced students were allowed to join.
The team I joined set out to do an analysis of NBA data. From the start, there was some conflict on whether the NBA had enough easily accessible data for us to use to complete the project. Some members of the group thought that college football would be an easier sport to analyze. They said that we should do college football as a way of getting our toes wet, then we could move onto the NBA.
We started out using NBA data for the first couple of weeks, but the college football supporters were pretty adamant in their idea. We finally relented and started using football data. As the semester went on fewer and fewer people came to meetings. Most people were not as interested in the analysis of college football data as NBA, so they decided to just drop the club. By the end of the semester we only had a half-done project, so we all just decided to give up.
I think that our project was doomed from the start. It seems as though we failed at even establishing some core values from the Model 1 theory of Argyris and Schon. After the first meeting, we did not have any sort of game plan. We didn't have a well defined (or even badly defined) end goal. We just had an idea of doing sports analytics. This was not a recipe for success. We also did not minimize expressing negative feelings, as from the start some showed that they did not trust in the idea and wanted to do their own. (Why they did not just form another group I have no idea).
There were also plenty of informal networks in the groups. Several pairs or triplets of friends joined the group together so that led them to support each other in decision making, which created further division in the group. I think this is another reason the football idea was so popular, as several of those expressing the idea were friends from before they joined the club.
Finally, and probably most egregiously, we also did not really come up with any method of managing the group. Most of the time ideas or strategies were presented, and those that presented them were assumed to do that strategy. We did not really divide up tasks. This led to people not presenting any ideas and then not having anything to do until the next meeting. We also did not have any real way of enforcing deadlines, as most did not care about the project past a resume builder. In the end, the lack of management and care by the group is what killed us.
This project was not all bad, though. I learned a lot about how to effectively start student groups from this experience. I joined the club again this year, and we used the first meeting as a way to decide concrete goals on a timeline, and assign people to different deliverables. I think had we done this to start of the project last year we would have been far more successful. The management style of group decision could have worked, and this year I am trying again and with the division of tasks, we have decided it is working far better. Finally, the informal networks were also a big part in killing the group last year. I think had we split up the group based on sport from the start we would have been far more successful.
The team I joined set out to do an analysis of NBA data. From the start, there was some conflict on whether the NBA had enough easily accessible data for us to use to complete the project. Some members of the group thought that college football would be an easier sport to analyze. They said that we should do college football as a way of getting our toes wet, then we could move onto the NBA.
We started out using NBA data for the first couple of weeks, but the college football supporters were pretty adamant in their idea. We finally relented and started using football data. As the semester went on fewer and fewer people came to meetings. Most people were not as interested in the analysis of college football data as NBA, so they decided to just drop the club. By the end of the semester we only had a half-done project, so we all just decided to give up.
I think that our project was doomed from the start. It seems as though we failed at even establishing some core values from the Model 1 theory of Argyris and Schon. After the first meeting, we did not have any sort of game plan. We didn't have a well defined (or even badly defined) end goal. We just had an idea of doing sports analytics. This was not a recipe for success. We also did not minimize expressing negative feelings, as from the start some showed that they did not trust in the idea and wanted to do their own. (Why they did not just form another group I have no idea).
There were also plenty of informal networks in the groups. Several pairs or triplets of friends joined the group together so that led them to support each other in decision making, which created further division in the group. I think this is another reason the football idea was so popular, as several of those expressing the idea were friends from before they joined the club.
Finally, and probably most egregiously, we also did not really come up with any method of managing the group. Most of the time ideas or strategies were presented, and those that presented them were assumed to do that strategy. We did not really divide up tasks. This led to people not presenting any ideas and then not having anything to do until the next meeting. We also did not have any real way of enforcing deadlines, as most did not care about the project past a resume builder. In the end, the lack of management and care by the group is what killed us.
This project was not all bad, though. I learned a lot about how to effectively start student groups from this experience. I joined the club again this year, and we used the first meeting as a way to decide concrete goals on a timeline, and assign people to different deliverables. I think had we done this to start of the project last year we would have been far more successful. The management style of group decision could have worked, and this year I am trying again and with the division of tasks, we have decided it is working far better. Finally, the informal networks were also a big part in killing the group last year. I think had we split up the group based on sport from the start we would have been far more successful.
Let me ask some questions about the overall club. Were there other project groups that were successful the year your group didn't produce anything? What happens with a successful project? Is some document written? Are the day published online? Then there is the relationship between the individual projects and the overall club. How did that work?
ReplyDeleteNow, getting to the team that you did join, did it have some name or some purpose? In other words, was it definitely looking at sports data? This is not an area I know a lot about, but a friend is really big on Bill James and what he's done for baseball analysis. I'm not aware whether there is equivalent information available to the public for other sports. But, let's say there isn't. In the book Moneyball (but I don't think in the movie) Bill James is discussed quite extensively. I'm wondering whether the other members of your club wanted to do a Bill James imitation, but then with the data from other sports.
That would be an understandable yet vague source of motivation. What Bill James did took years and years of careful data analysis, and as you said, these projects were supposed to take a semester. So I'm wondering whether Argyris and Schon are relevant here and if the real issue is the expectation of doing something really interesting with the reality that it likely wasn't going to happen in the time period allotted for the task.
Now, just to amuse you and to suggest a possible data project to consider, when Table PCs were something of a think on campus I hand wrote this document on Favorites and Underdogs, which gives a theoretical derivation that favorites should be risk averse and do things to lessen risk, while underdogs should be risk seeking and do things that increase risk. This is the type of theoretical result that one could confront with data and see: (1) does it predict behavior, (2) if predicts behavior sometimes but not always, what happens to the likelihood of winning when the it does predict behavior? Does it go up or go down?
So, as a final note, I love the idea of looking at data to tell you something, but I was schooled on theory first. You need some model. That the data can tell the full story without any model, that makes me ill inside. Alas, I'm afraid that's becomeing the majority view.
Hi Professor,
ReplyDeleteThere were other project groups that were in fact successful. A popular thing was to present at engineering open house. After that though, most projects kind of just wallowed on github only to be used as a resume item.
Moneyball was a big inspiration for our team. We talked about it a few times at meetings. I think you are right though that we took on too much of a project for a semester. We also didn't really have the actual industry knowledge required either.
Looking at underdogs was actually the first thing we did. We made a formula to determine an upset value so we could group upsets by how 'upsetting' they were. We also tried a k means clustering approach but I think we liked the formula we made up better. We wanted to see if team strategy or individual player performance was very different in these games apart from the other season games, but we did not actually get to that part of the project sadly.
In regards to your final paragraph, I think that that view is becoming shared amongst experts. I went to the big data summit in research park last week and there was a presentation on just that idea. I think that most students think that the data will solve every problem, and that is what the presentation said as well. I have to admit that I was a more data forward idealist before I heard this. I'm interested in seeing how this paradigm shift will have me approach ideas in the future.