How do we know Hamilton wrote the other 51?
The history of the Federalist Papers as a discrimination task
Yesterday, like 90% of musical theater kids with internet connections, I spent yesterday watching the filmed version of Hamilton. (Yes, I am aware of a lot of the very legitimate criticism against Hamilton and Lin Manuel Miranda. I personally think it is a masterful work of fanfiction. Please don’t be that guy).
In the show, during finale of Act I, Aaron Burr delivers a passionately inflected statement about the breakdown of the authorship of the Federalist Papers.
Trust me, it’s a real show stopper.
Fictionalized Aaron Burr is actually only able to bellow authoratatively about the 85 annoymously published essays because of data science.
Prior to the 1960’s, historians had achieved consensus that John Jay had written 5, Hamilton had written 51, Madison had written 14, and Hamilton and Madison had written 3 collaborative. However, there were 12 essays whose authorship was disputed as being written by either Hamilton or Madison.
Bayesians, We Get the Job Done
In the early 1960’s, Frederick Mosteller and David L. Wallace determined that it was highly likely that Madison wrote all 12 of the disputed papers. First using Fischer’s Linear Discriminate Analysis (a frequentist technique) then using Bayesian methods.
The Bayesian methodology blew the traditional frequentist approach out of the water in terms of the strength of support it lended to the conclusion that Madison wrote all 12. At this time, Bayesian methods were not in common use, in large part due to them being computationally complex.
This study was also landmark in early natural language processing and found that function words like “upon”, “an”, and “any” proved to be better discriminators of authorship than content words
Thankfully, we now live in the age where we no longer use punchcards to power our computing and since then new and more sophisticated methods of text discrimination and Bayesian analysis have been and continue to be developed.
If you’re interested in reading the original 1963 paper, it’s availible here.