Console #17: Japanese Pizza Toast and Turtles
SDE @amazon | independent consultant @block_one_ | sjkelleyjr.com
Welcome to Console, a technology, business, and financial markets newsletter with an emphasis on rationality. If you’d like to sign up, you can do so here
The more I analyze and partake in business and economics, the more introspective I’ve become in the direction of history and culture. Business is inherently cultural because it is an inherently human endeavor.
Food is the perfect blend of culture and business, and Craig’s writing about Japanese Pizza Toast exemplifies this so well.
Two words: irrational exuberance.
I’m happy to see Toyota innovating, but this kind of seems like an executive’s pet project more than anything else.
The problem is, I’m also a fan of Nassim Nicholas Taleb, and have seen him eviscerate studies like these, so I’m always wary of the statistics used.
It looks like this particular study uses a Poisson regression, which likely means a lot of information is lost between reality and the model.* They found a p-value of .013, which is better than the often used p-value of .05. However, Taleb has shown p-values to be a flimsy indication of statistical correlation
. It seems the paper this article is written about is a case of the all too pervasive p-value hacking that’s found in the soft sciences today.
One of the projects in one of my statistics courses back in college was to collect a news story on a research paper similar to this one and analyze both the news story and the statistics used in the research paper.
The news organizations are incentivized to create content, particularly content with eye-grabbing headlines for SEO purposes. Because of this, what they’ll often do is find a research paper and interpret the results. The more extreme the result they interpret the better the headline, so they’re biased in that they would like their interpretation to align with the best outcome for the news organization’s revenue.
This involves a bastardization of the results of the research, which itself is already on faulty grounds due to its use of p-values and regression models. Often the news story is incorrect in what the research paper is asserting, and often the research paper itself is p-value hacking to get a paper into a journal. It’s turtles all the way down.
*I would say that about any model by the way, but that’s a discussion for another newsletter.
This week’s newsletter not your cup of tea? Check out some of the more popular weeks below!
Subscribe to get your daily round-up of top tech stories!