Catching up: “Statistics: Making Sense of Data” on Coursera
When you’re trying to figure out how best to use all these wonderful tools we now have for the computational analysis of literary texts (such as the stylometry scripts, MALLET, WEKA, and many more), you tend to get into situations where you wish your knowledge in statistics was more solid, recent, broad, and thorough. (I did study psychology as a minor, and that included a little bit of statistics, but don’t remind me of when that was).

Image by user IllonkaTallina on Flickr, published under a Creative-Commons BY-SA-NC license. http://www.flickr.com/photos/92641139@N03/8423301722/
So, in order to catch up with statistics, and to finally see what the talk about MOOCs is all about, I enrolled in Coursera’s class “Statistics: Making Sense of Data“, together with about 50,000 others (several of them local colleagues!). After a little more than four weeks into the class, it’s time to ask myself whether a) this will help me with my text analysis research and b) whether MOOCs are the right thing for me.
Frankly, I think my answer to both of these questions is something of an emphatic “yes, but”. YES, what I’m learning about statistics is extremely relevant to computational text analysis: basic notions like random sampling, standard deviation, normal distribution, or significance tests are more than useful, they are essential. And the accompanying R tutorials are really good too, because they enable me to move from theory to action. BUT what I’m freshening up right now is really just the basics, and although doing this now will certainly help me get started, there is much more work to do.
And YES, the MOOC format has some really cool features: a series of short video lectures, nicely prepared with graphs and plots, some images, some formulas, is where the meat is. And you really do learn quite a bit more by doing the little quizzes in the video lectures (to let you check whether you really understood), the more elaborate weekly quizzes and the first longer assignment. Since we are 50,000, we will have to rate each others’ assignments this week, as well. BUT there are limits to online interactive learning, at least in this one implementation of a MOOC: it would be great to have something like an “ipython notebook” for the R statistical environment built into the MOOC. This would allow you to actually try out statistical computations as part of the quizzes and assignments.
Of course, every course will be different. This one has an extremely relevant topic, and the two instructors, Alison Gibbs and Jeffrey Rosenthal, are doing a great job keeping everyone motivated. The idea to look at the same datasets several times from different angles and with various questions is also very good. One thing that bothers me a bit is that they tend to spend a little too much time on the easy stuff (watching Jeff flip a coin ten times is concrete, but not very instructional) and then speed over the harder stuff (for example, how to actually calculate the standard deviation or a p-value). Not that I mind looking things up on Wikipedia and doing some additional thinking myself – but the real test for a medium of instruction (or anything else, I guess) are the hard problems.
Still, I’m going to continue with this course and hopefully earn my “statement of accomplishment” (there is one). And I will try out some more massive open online courses soon – there are many more exciting classes out there which are also relevant to computational text analysis – just on Coursera, there is also Machine Learning (already running, and looking good) as well as Programming with Python (not so much focused on text manipulation).
And of course, there are other platforms: There is the popular Khan Academy which has plenty of stuff in Computer Science. And, as you may have heard, Openuped has just been launched, with a more “open” approach in terms of who gets to offer a class, what types of classes are on offer (with more humanities topics: The Modern Middle East or Cultural Heritage, for instance), and what languages the classes are in (Turkish, Russian, Arabic, Gaelic, …). So yes, the digital age also means continuous learning is at the reach of more and more people!
I’m just learning how to do visualizations as you describe in your blog, and I’m experimenting with MALLET this summer. I agree wholeheartedly about the challenges of learning statistics in order to do work in humanities with texts! Your blog is both a source of help and encouragement: thank you! :)