When most people think of labs, they imagine scientists in white coats staring into microscopes, carrying around beakers of bubbling chemicals, and holding test tubes over Bunsen burners. In social science, the reality is much more mundane. It’s usually just a room full of computers with software that may or may not be useful and may or may not be up to date. Even less compelling are the labs associated with statistical methods classes. The last couple years my own classes have been the worst case scenario–I just get up and lecture about how my students should use some particular piece of software to apply the methods we’ve been learning in the “lecture” part of the class. It doesn’t have to be this way.
Over the next few months I will have the opportunity to teach two new methods classes and completely re-invent how I incorporate labs. I had lunch with Mayur Desai the other day and I think he does a great job with labs in his classes and he’s inspired most (but not all) of the ideas here. This is what I’m thinking:
No lectures. None. Students enter the lab and get their assignment and spend the rest of the class trying to complete it.
Each assignment starts with a data set (preferably real) and a blank screen–that is, I don’t give them any code. Their job is to answer a substantive question by applying methods we’ve covered recently to the data.
Students work in pairs and take turns driving. I think this keeps students focused and they can teach each other. It also means only half the class has to have laptops if I want to implement a lab in a regular classroom.
I’m around to answer questions. In this way, it’s very different from a problem set where getting stuck on something dumb for hours at a time is a common occurrence. Struggling with problem is good for learning, but banging your head against a wall isn’t an efficient use of time.
The end product should be similar to results they might find in a published paper. Sometimes I’ll provide an empty table they must fill in and other times they will produce their own tables of results from scratch.
There should be opportunities for quicker/more advanced students to do more. One size does not fit all.
While it’s possible to use any statistical analysis tool in a lab successfully, I do think some packages are better than others. Most students already know Microsoft Excel and doing basic analyses (even regression) using it is easy, but you really hit a wall when you want to do anything even a little sophisticated. SAS is powerful, but there is a steep learning curve. My plan is to use Stata. You can browse your data in a spreadsheet style interface. You can play with commands through the menus and when you choose one, it shows you the command-line equivalent. You can work interactively at the command-line or build programs (using those same commands) in an editor. And the documentation is excellent and available online.
I’ll let you know how it goes!
- Reblogged from highvariance