When I went to graduate school I did not realize I would have to take statistics. This is super silly because I went to school for a technical degree and I also know how to code. I started grad school freshly 22, with an extremely limited idea of what a technical grad program entailed.
Now that I am several years removed from grad school, I feel safe enough to reflect on those times with a bit more insight. A recent therapy session inspired me to look through my online archives of everything I had ever coded, regressed, and analyzed in school. I wrote and supported policy papers on everything from domestic violence (how staying in abusive relationships for the short term enhances one’s chances of leaving in the long term) from policy memos addressed to dead world leaders. Policy masters programs are very strange. What I thought would be a fun two years filled with slow mornings of coffee at my desk, messy academic hair buns (you know the Pinterest photo I’m talking about?) with pencils holding my hair together and an environment of learning bliss was really, really wrong. Grad school was essentially a beer fueled two years full of not getting more than 5 hours of sleep a night and submitting homework a few minutes before the deadline. Broke and tired was the reality. And statistics.
The best professor I have ever had taught the last statistics class I ever took. Recently, they took to Instagram to talk about the experimental methods if one were to date them. I remember rewatching these stories over and over again, trying to remember why Propensity Score Matching might be labeled as a stereotypical Gemini and what the exclusion criteria was for certain methods. I stopped short of opening R and regressing for fun. Instead, I landed at this blog post. Caught in a space between trying to remember my degree and feelings of wanting to forget it. While I try to understand what those feelings mean, I’m going to have fun with my earned degree. That means labeling statistical methods Sun and Moon signs, because why not?
Randomized Control Trials
RCTs have some bullshit fixed signs that I am totally not envious of (I am very envious). These methods are basic, a bit conceited because they are the gold standard and probably have a Leo sun. They have a Cancer moon, which makes you a little jealous of them because they know how to feel their feelings but also have fun. RTCs are trials in which subjects are randomly assigned to one of two groups: one (the experimental group) receiving the intervention that is being tested, and the other (the comparison group or control) receiving an alternative (conventional) treatment. They carry a Louis Vuitton Neverfull bag with a Roxanne Gay book sticking out of it to show the world they are really woke. They are the type to date you but have someone else lined up when they are ready to exit the relationship. You’ve dated an RTC, everyone has.
What’s the sign that wants to be a Leo but isn’t? Scorpios. Lotteries are more complicated than RTCs. Maybe they did a stint in Americorps or worked at a local nonprofit. They may have taught abroad for a year before coming home to live with their parents again. The strong advantage of a lottery-based design is that it results in two groups that differ from each other only by chance. They also have a Scorpio or a Capricorn moon and are pretty vocal about it. If you take them on a date they’ll brag about using crystal deodorant and come outfitted in Free People (thrifted, of course).
Fixed Effects/Diff in diff
Gaslight, girlboss, gatekeep. To use diff-in-diff, we need observed outcomes of people who were exposed to the intervention (treated) and people not exposed to the intervention (control), both before and after the intervention. AKA: they need to creep on your ex’s social media accounts and play the comparison game, but they’ll never admit it. Diff-in-diff has a Gemini sun, Libra rising and an Aries moon. This is my personal favorite method because it was fun to code.
Regression Discontinuity Designs
Too good to be true, but they are limited. Think Aquarius sun and moon. (RDD) is a quasi-experimental evaluation option that measures the impact of an intervention, or treatment, by applying a treatment assignment mechanism based on a continuous eligibility index which is a variable with a continuous distribution. At its best, RDD can be almost as good as Randomized Control Trials. This would be like dating a Scoprio and thinking they are a fire sign. Not quite an Aries, but close.
This method is used to account for unexpected behavior between variables. An instrumental variable is when a third variable, Z, is used in regression analysis when you have endogenous variables—variables that are influenced by other variables in the model. In other words, you use this method to account for unexpected behavior between variables. Instrumental Variables account for the unexpected, which means they have a Sagittarius sun, Gemini rising and a Leo moon. Nothing is where you’d expect it to be. If you were to date Instrumental Variables they would absolutely be a catfish.
Propensity Score Matching
“Even if that region of common support is met, you’re still dating a Gemini” comes from Professor Nelson’s Instagram and perfectly sums up this method. PSM creates an artificial control group by matching each treated variable with a non-treated variable of similar characteristics. Only then can you estimate the impact of an intervention. I like this method, they scream the earth sign to me. I’m giving them a Taurus moon to mirror my own.
I felt satiated in grad school. I actively felt like I was learning and applying it. I could touch my research, code, and presentations. Reflecting on the time spent in Bloomington, I am always glad I went. Though I may not be ready to dive further into the repressed memories I have from grad school, it feels good to flex these mental muscles again, even if it means assigning experimental methods half assed moon signs.