Cohort retention metrics are critical to evaluate product-market fit in consumer tech. If a product is satisfying user needs in the relevant market, retention metrics will reflect that users continually return to the product over time.

I think the quality of a movie is highly correlated — but not perfectly correlated — with how rewatchable the movie is. There are great movies that I never want to watch again (Raging Bull, A Separation) and there are lesser movies that I find difficult to turn off once I catch a snippet (yes, the Austin Powers trilogy…).

Over at FiveThirtyEight, this week’s Riddler Classic question truly is a “classic” varietal: a problem about drawing colored balls selected from an urn. We are told:

I’ve long admired the writing of Philip Roth. Though I’ve read ten of his novels, it somehow took me until this year to pick up what many regard as his finest work, American Pastoral, but like dessert it sat waiting. The book follows the turmoil in the life of Swede Levov, former star athlete and successful businessman, after his daughter commits an act of political terrorism in 1968 in protest of the Vietnam War.

Consider the responsibilities of a typical Analyst in a technology company — whether in Product Analytics, Business Analytics, Marketing Analytics, or Data Science. A simplified model can classify their work into offense and defense:

[This post is intended for practitioners of A/B tests and assumes some familiarity with A/B testing concepts.]

After I handed in the final copy of my senior thesis, an event that signified the intellectual summation of my four years of college, my advisor told me it would likely be the longest thing I’d ever write. Until last month, I would have taken his side of the bet; my recent scribbling had been restricted primarily to technical subjects along with some miscellaneous blog posts. I couldn’t imagine anything that would eclipse the 22,000-word mark I had previously set.

I’ve come across a number of SQL style guides and had major points of disagreement with most of them. My grumbling tends to come from, in general, their reliance on tradition rather than updating standards for the workflow of the modern analyst or engineer. To be more formal, we might say they are path dependent, and optimized for a set of conditions that no longer exist.

The computer scientist Ken Thompson has an unusual distinction: his research has changed the rules of chess. Thompson has a rich resume, but it’s quite a feather in one’s cap to lay claim to changing a game that’s existed for hundreds of years. Specifically, Thompson’s computer-aided work on chess endgames carved out several exceptions to the long-standing fifty-move rule. At least, his work did until Thompson’s discoveries became too numerous, and chess’s overlords threw up their hands and said they were getting rid of the exceptions entirely.

My preferred formal definition of politics is the authoritative allocation of resources. Resources — money, healthcare, oil — are subject to scarcity, and thus politics becomes the authoritative allocation of scarce resources.

Note: This post is intended for practioners of experiment and A/B test evaluation in technology settings. I argue that proposed product changes that increase a target metric while reducing a participation metric (e.g., daily active users) should require elevated scrutiny.

What I really want to talk about is thank-you notes after job interviews and why I believe you shouldn’t send them – but first, we’ll dive into “Pascal’s wager” to develop the decision framework for later use.

In 2014, Benjamin Morris published “The Hidden Value of the NBA Steal” on FiveThirtyEight. In his article, Morris argued that both traditional NBA player evaluation and analytics had underrated the value of a player getting steals; he included the eye-catching assertion that “a steal is ‘worth’ as much as nine points.” (More on that later.) With its provocative headline, the article generated a fair amount of discussion in the sports analytics community. Morris ended up writing a four-part response and rebuttal to his critics [1,2,3,4].

Recently, I started learning more about the Minecraft speedrunning community. For anyone unfamiliar with “Minecraft” or “speedrunning”: Minecraft is a video game based on exploration, world-building, and construction; and speedrunning is an attempt to complete a video game or part of a video game as quickly as possible. Speedrunning has grown into a popular subculture across many games: some of the most popular speedruns have upward of 10 million views on YouTube, and players frequently livestream speedrun attempts to substantial audiences on Twitch and other platforms. The website speedrun.com serves as an archivist and provides leaderboards for many communities. As of March 2021, the site indicates that Minecraft is the most popular game for speedrunning, with over 1,400 registered and active speedrunners. The size of the full Minecraft speedrunning “community,” including watchers, likely numbers at least in the hundreds of thousands, based on YouTube and Twitch view counts.

I posit the sparkling water conjecture in the nascent field of calorific arithmetic. The conjecture is as follows:

How “close” was the 2020 U.S. presidential election? There are several ways to approach the question: for instance, one can cite the popular vote (Joe Biden won by 7,052,770 votes, or by a margin of 4.4%) or the electoral college margin (Biden won 306 electoral votes to Trump’s 232). But if we focus on the closeness of the outcome of the election — who ended up in the White House — neither of these metrics tell us very much alone. The popular vote is irrelevant in a system governed by the institution of the electoral college (popular vote is not how the election is decided), and the overall electoral college margin can obscure the within-state vote margins that provide tipping points for the election’s outcome.

On January 30, I posted an observation that plotting Collatz sequences in polar coordinates produces a cardioid and nephroid. I didn’t have an underlying intuition as to why.

This week’s Riddler column over at FiveThirtyEight sets us up with some classic fodder for mathematical puzzles – the “The Tower of Hanoi,” also known as Lucas’s tower. As the column notes, the tower…

The Collatz conjecture is a well-known math problem that concerns the following sequence for a given starting integer N:

Originally published over on my Medium page, I took a look at why colleges’ ‘average class size’ statistics misleading, and suggest a better alternative to give students the right data to make any decisions.