As Gen Z embraces a host of optimization trends — ranging from sleepmaxxing to looksmaxxing — a new craze has taken hold among statisticians: likelihoodmaxxing.
The fad involves subjecting likelihood functions to a series of exacting statistical procedures to identify values that make them as large as possible. Originally a classical method pioneered by R.A. Fisher, likelihoodmaxxing has taken off online. Its aficionados tout multiple benefits, including improved point estimation, clearer skin, valid statistical inference, and a smug sense of self-satisfaction that many describe as euphoric.
Statistics and machine-learning subreddits are increasingly filled with likelihoodmaxxing content. On a recent Saturday, a top post, from user “MLE_Dickinson”, showed a Binomial likelihood, its apex labeled with the Edmund Hillary quote: “It is not the mountain we conquer, but ourselves.” Commenters were delighted. “It's giving apogee,” wrote one fan below the post. “Very log-concave, very mindful” remarked another.
Cultural critics said the likelihoodmaxxing is not inherently dangerous, but that the practice can induce antisocial or even monomaniacal behavior. “As with all optimization strategies under capitalism, likelihoodmaxxing is about inculcating a false sense of control,” explained Jia Tolentino, staff writer for the New Yorker. “If you can find that Poisson MLE, maybe you can find your place in an increasingly uncertain economy.”
Assistant Professor of Statistics David Hoffman doesn't see it that way. The 31-year-old describes his favored statistical manipulations as part of a “small-batch, artisanal-numerical” experience. In a series of TikTok videos, Hoffman detailed the strenuous routines to which he subjects his likelihoods. He begins by computing their logarithms — or “logging,” in the parlance of online likelihoodmaxxers. He then takes their derivatives (known as “Frenchy-ating”), sets them equal to zero (“null-rizzing”), and — if the resulting system is linear — uses row reduction to solve the resulting equation (“slay[ing] the Gauss-found roots”).
Hoffman takes particular pleasure in certifying that he has, in fact, likelihoodmaxxed successfully. “It's a multi-step process,” he explained to his followers in a recent video. “Bone broth. Alkaline ice bath. Take the second derivative and confirm it's negative. And gratitude — always gratitude.”
In her practice, Oakland Psychologist Verna Wentworth-Currier regularly sees Bay Area likelihoodmaxxers who have gone to extreme lengths. “Likelihoodmaxxing should not imply sanity-minning,” she explains. Wentworth-Currier stresses moderation, exercise, and frequent bathroom breaks, even for veteran likelihoodmaxxers. “I've seen optimization routines you wouldn't believe: line searches, branching, bounding, quasi-Newton...” she says, her voice cracking. “At some point, you have to just say, 'I'm high enough.'”
To Hoffman, the Assistant Professor, it's all about consistency — in every sense of the word. “My MLE approaches the truth in the asymptotic limit,” he explains. “Maybe, with hard work and a little self-discipline, I can too.”
And with that, he popped in his headphones, blasted Kate Bush's “Running Up That Hill”, and got back to work.