One of the two classes that I’m taking this semester is in design of experiments, which is more fun that it might sound. I’ve done these before, and I think that the one that I’m doing now is pretty cool.
I ran an OFAT experiment back in middle school using paper airplanes and a contraption that ensured that they’d be propelled with the same force each time. Dad and I built a wooden box, entrapped a rubber band inside of a staple on the front of the box, and placed a clothespin at the back of the box that would hold the tail of the plane such that the nose of the plane a standard length from the front edge of the box. Opening the clothespin launches the plane. I ran five trials for each plane, averaged them, found the variance, and tabulated my standings. I didn’t get first place, but I did have fun.
This experiment is more interesting and advanced. I’m running a fractional factorial experiment, which means that I can test four factors in just eight runs without losing anything but higher-order interactions that aren’t likely to matter very much. I’m using the coffeeshop that I spend a lot of time in, largely because I have the trust of the baristas that I’m not going to screw them over. I have four factors:
- Time of day of the shift. There are two shifts each day.
- The size of the tip jar.
- The opacity of the tip jar. These two factors require that we have four tip jars.
- Whether the tip jar is seeded or not.
My premise is this: tipping baristas is a social phenomenon. “Do I tip her? All she did was make me a latté,” is a valid question. Anything that we can do to shake up the social norm and show that, yes, people tip baristas is a good thing. But I have no idea what factors will work. That’s why you experiment, people.
I’m sitting here waiting on the end of the third shift under experiment. #4 and #5 happen tomorrow, and the rest will conclude by Thursday. I’m having fun, and so are the baristas, especially since I’m giving them money to participate. I got some very interesting results yesterday, and I’m getting some predictable ones today. Data! I want more data!
I’ll get to do this one more time, either as a full 2^3 factorial that would test every factor combination, or by doing the alternate fraction of this 2^(4-1), which would test the other eight combinations. I’d get to do the former if I have an obvious factor that has no value in the analysis of the first experimental run; if I don’t have conclusive data, I’ll do the latter. I could also run another fractional factorial by dropping a factor and adding another one.
The ladies already want me to run more experiments. I guess that I know what I’m going to be doing on my own time this summer.