How to setup an experiment

The Coffee Experiment!


When I first started in graduate school my process for performing/presenting experiments was more of a ‘shoot from the hip’ approach rather than one that followed a systematic protocol. Fortunately, I was taught a systematic protocol that is not only easy to follow, but also easily translates to an experimental outline for a research paper. Without further ado, here is an example of the protocol on a groundbreaking study (spoiler alert: yes, I just needed to be reminded of the scientific method from 5th grade…).

Note, the remainder of this post is taken from a document that has been circulated in the HazyResearch lab for years. It is unclear to me who wrote it, but they should get the credit.


Caffeine makes graduate students more productive.


To measure the productivity of graduate students, we use the time it takes to complete their PhD as the proxy. To measure the caffeine that each student takes, we use the number of cups of coffee a students drinks as the proxy.

Note that we do not count caffeine that comes from other sources such as tea or soda.


Out of the 100 students in our school, have them report the mean cups of coffee they drink each week. Also, record the date they start their PhD and the date they get their PhD. Only consider students who have left the school with a PhD.

Note that if a student is very productive, but dies of caffeine toxicity days before they finish their PhD, they will not be counted.

Note that since we record mean cups of coffee per week, we do not capture information about increases or decreases throughout their PhD.

Expected Results

We expect that students without caffeine take a long time to get their PhD; students who drink a lot of caffeine finish their PhD very quickly. But drinking too much caffeine would cause caffeine toxicity and would length the time to get a PhD due to psychosis. In short, we expect a convex curve like the one below.


Since it is convex, this implies there is some optimal amount of coffee to drink to minimize the time spent working on a PhD.


The results match our expected results. We observe that the knee of the curve is at approximately 30 cups of coffee per week with a PhD in 36 months; this seems to be the optimal amount of coffee to drink to be a productive graduate student.


Note that there seems to be an outlier who drinks about 60 cups of coffee a week but finished their PhD in 19 months. We tracked down this student and found out that they drink 5 large iced Americanos a day. Upon further testing, we discovered that this student is immune to caffeine toxicity.

Raw data for results are stored here: