This episode of Spencer Writes in the Library took place Thursday, April 9th around 9:30am.
Where am I working today?
Today, I am on the second floor of the west wing, at a table near the periodicals section. I’ve worked in this general area before, but never quite at this spot.
What’s a perk of this spot?
Things just really seemed to click today. Chair was comfy, table was at a good height, lots of light coming in through the windows, everything was dandy!
What’s a problem with this spot?
Being at a table rather than a desk made it easy for me to spread my stuff all over the place instead of being tidy and professional.
What have I learned in this spot?
I was really impressed today with the amount of electronic resources available through the MSU library. Early on in my explorations, I found a reference to a book that might be helpful. When I looked it up to see where I would need to go to flip through, I discovered that I could just download the PDF of the book because of my MSU affiliation. This was the easiest it’s ever been to find what I was looking for!
How would I rate this spot?
7 out of 5 dentists. (Why dentists?)
What am I working on today?
My exploration of principal components analysis and factor analysis are still moving forward, but I’ve started looking into something else that’s worth exploring. Many of the variables that I’m considering in this study only appear a fraction of times when compared to the number of cases (i.e., games) that are involved. In other words, when I’m looking at a list of ~13,000 games, there are some game features that are only present in 5, 10, or 100 of those games. This creates the possibility that those 5, 10, or 100 games have a disproportionate amount of influence on the numbers that I’m crunching, so I spent today trying to find a reasonable cutoff point. That is, how many games does a feature need to appear in before I should really consider it? It’s easy to come up with an arbitrary number, but it’s harder to find something that I know how to defend to, say, a practicum committee. I spent several hours today trying to find some guidelines for this sort of decision.
What’s the highlight from today’s work?
Well, I found something, but I’m not quite done yet. One of the trickiest things about answering this sort of question is not immediately knowing what sort of words to use to describe what you’re looking for. I found some references to “near zero-variance predictors,” which sounded (and still sounds) very promising. These references offered some helpful guidelines for what kind of cutoff point to use when deciding whether or not to keep a variable… the only problem is that it doesn’t offer a lot of evidence to back up these guidelines. So, the search continues!