Earlier this year, I posted the first part of a write-up for a project I worked on during a project I worked on last year. As a reminder, my project involved taking 4 different types of correlation, test the baseline and 8 different data shifts, generating 1000 simulations with 100 points of data each. I then compared how three different multivariate control charts, Hotelling T^2, MEWMA, and MCUSUM, were in detecting a shift while holding the false rate fixed over all simulations. For this I use the pretty standard for in control average run length (ARL) of 300. This means, on average when the process is in control, you can expect to see an out of control data point in any 300 sequential points.

Time to mix it up a bit from my usual game probability related simulation. This week I want to talk about a project I worked on last year in doing simulations for Statistical Process Control. For a brief primer, Statistical Process Control or SPC is a method used for monitoring something like a manufacturing process. It’s a form of time series analysis where you monitor results of some statistic taken from observing the process and then compare that statistic to a set of limits on a control chart to determine whether or not the distribution of that statistic has likely changed.