A Beginner’s Guide to Irrational Behavior started yesterday. One might not immediately think that such a course would be relevant for statistical modeling. Well, it is if your statistical modeling uses people as informants. If the data come from individuals responding to inquiries, then it would help if we understood the response generation process.
For example, the reading list from the first week includes the article “How actions create – not just reveal – preferences.” For some reason this makes perfect sense in machine learning when we are programming our driverless car. Well, people were the first processors of big data, and our survival depended on our ability to structure input to serve our needs. We see what we must in order to act, and the rest is background. Our statistical models should do the same.
I recommend this course because it is a fun introduction covering many of the examples that you must work your way through in order to understand how much of what we take as immediate and direct requires learning and processing. What if that positive manifold, the halo effect, was not measurement error but a real manifestation of a greedy algorithm supporting action in an uncertain environment? If such were the case, then I would not want my statistical model to control for halo effects because the halo is the driver of behavior. Similarly, I would not use self-reports in my statistical model if the self-reports were after-the-fact justifications of choices made for other reasons. These are the issues raised by Dan Ariely’s Coursera course. They ought to be addressed in our model specification.