Have you heard the one where a veterinarian, an architect, and a physicist walk into a barn?

The dairy farmer asks them, “How can my farm produce more milk?”

The vet recommends feeding the cows a hormone to stimulate more milk per cow.

The architect recommends reducing each stall width by six inches to make room for more cows.

The physicist studies the situation carefully.

Two weeks later, the farmer receives the physicist’s long-awaited report. It begins: “Consider a spherical cow of radius 1 in a vacuum.”

This is the spherical cow conundrum. Some scientific models reduce a problem to its simplest form, even if the model isn’t compatible with reality, in order to allow for workable calculations. This Occam’s Razor approach is commonly stated as, “The simplest explanation is usually the best one.”

Although modeling — creating a tool to understand the past and predict the future — is essential to scientific inquiry, the spherical cow approach can lead us astray.

Take this hypothetical conversation: A clinic administrator asks a physician, “Look at this model. I’ve solved our clinic capacity problem. Each physician can see 64 patients per day.”

“How did you come up with that number?” asks the physician.

“I assumed each doctor spends 100% of their time seeing patients.”

“But that’s not feasible. What about time needed for staff meetings, phone calls, emails, chart review, consulting with colleagues, prior authorizations?”

“Stop right there!” interrupts the administrator. “I also assumed that each physician sees eight patients per hour.”

“That’s not realistic. And it’s not good care,” responds the physician.

“I’m not talking about what’s realistic. I’m talking about what I assumed. We can change the assumptions.”

“So you’re talking about hypotheticals, not real numbers.”

“Exactly,” says the administrator. “You’re stuck in reality.”

Welcome to Model Land! Statistician Erica Thompson, in her 2022 book “Escape from Model Land” warns that, “Taking models literally and failing to account for the gap between Model Land and the real world is a recipe for underestimating risk and suffering the consequences of hubris.”

Making an assumption that’s incompatible with reality can lead to significant mistakes in forecasting, such as basing clinic patient capacity — the number of patients a clinic can schedule — on 100% direct care time by medical staff. In addition, models can’t account for unquantifiable unknowns, such as a pandemic lockdown.

And consider this: every model, by its very nature, is biased. For example, a model that looks at clinic patient capacity without considering factors such as how many patients a physician can comfortably see in a day without burning out, or the centrality of the doctor-patient relationship, appears to value productivity more than staff or patient satisfaction.

The physician’s frustrated response to the administrator reflects this values gap. “Your hypotheticals are not based on real numbers. Perhaps the whole model is based on imaginary numbers!”

This example reflects a need to escape not from models, but from Model Land. Without models, Thompson reminds us, “Data would be only a meaningless stream of numbers.”

Thankfully, most models are well-grounded, starting with defining the model’s purpose: What questions can this model answer? What questions are beyond its scope?

For example, using the weather app on my phone, I can make a reasonable decision about whether to wear snow boots to work tomorrow. But the weather app is less reliable about whether I should wear snow boots 10 days from today. Perhaps 10 days is beyond the scope of this weather forecasting model.

In addition, most researchers make sure to acknowledge the value judgments implicit in the model. If they can’t identify value judgments, they look for what the model leaves out, which reflects what is less valued.

Thompson recommends using more than one model, not only to gain insight into the limitations and biases of each model, but also to facilitate communication and keep the models reality-based.

As artificial intelligence (AI) assumes a more central role in the modeling space, maintaining a connection to reality becomes even more crucial.

In terms of reducing a model to its simplest form, Albert Einstein put it best: “Everything should be as simple as it can be, but not simpler.”

As for the spherical cow with radius of 1 in a vacuum, she is with the spherical chickens, grazing in Model Land.

Diane Roston is on the clinical faculty of the Geisel School of Medicine at Dartmouth, Department of Psychiatry, and serves as a staff psychiatrist at West Central Behavioral Health in Lebanon.