Base Rates

In the early months of the COVID-19 pandemic, millions of people across the world rushed to buy home pulse oximeters after reports that low blood oxygen could signal dangerous pneumonia even in patients who felt fine. The devices were cheap and widely available. But a problem emerged that most buyers never considered: when a rare condition is screened across an enormous population, the math of false positives overwhelms the math of true detection. With a disease prevalence of, say, 1% in the general population and a device accuracy rate of 95%, a low reading was still far more likely to be a false alarm than a real warning. Emergency departments filled with anxious people whose oxygen was fine. The critical number they needed -- the base rate of actual illness in their demographic -- was almost never part of the conversation. The device's accuracy was beside the point without it.

What a Base Rate Is -- and What It Is Not

A base rate is the underlying frequency of an outcome in a relevant population, prior to considering any specific information about a particular case. It is the statistical starting point: how often does this type of thing happen, across all cases, before you know anything distinctive about the one in front of you? The base rate for startup failure is roughly 90%. The base rate for a new restaurant closing within five years is about 60%. The base rate for a randomly selected American adult having a given rare disease might be 1 in 10,000.

This is not the same as a prediction. A prediction incorporates specific evidence about a particular case; a base rate is what you should believe before any such evidence arrives. The distinction is crucial because the most common error in human judgment is not failing to gather specific evidence -- people are generally eager to do that -- but failing to start from the right statistical foundation before interpreting that evidence. A base rate tells you what is typical. A prediction tells you what is likely in this specific instance. Good judgment requires both, in the right order.

Why the Brain Discards Base Rates

The tendency to ignore base rates -- formally called base rate neglect -- was identified and documented by Daniel Kahneman and Amos Tversky in a series of experiments beginning in the early 1970s. In their landmark 1973 study, participants were told that a group consisted of 70 engineers and 30 lawyers (or the reverse), then given a personality sketch of a randomly selected member. The sketch was designed to sound stereotypically like an engineer. When the base rate favored engineers (70/30), participants judged the person was likely an engineer -- reasonable enough. But when the base rate was reversed to 30 engineers and 70 lawyers, with the identical personality sketch, participants still rated the person as likely an engineer. The base rate barely registered against the vivid, narrative detail of the description.

The mechanism is not stupidity. It is a deep feature of what Kahneman later described as System 1 and System 2 processing. The brain's fast, automatic system excels at pattern-matching and narrative construction. When it encounters a specific story -- a charismatic founder, a brilliant restaurant concept, a positive medical test -- it generates a judgment immediately, using representativeness (how closely the case resembles a prototype) rather than probability. The slow, deliberate system that could integrate base rate information requires effortful activation, and in most situations, the fast system's answer arrives first and feels right. Kahneman documented that even professional statisticians are susceptible when problems are framed in narrative rather than numerical terms, suggesting the bias is not a knowledge deficit but a processing default wired into the architecture of human reasoning.

Two Scales of Evidence

At the personal level, base rate neglect shapes judgments daily. Consider a founder leaving a stable job to launch a startup. She has a compelling idea, relevant expertise, eighteen months of runway, and an enthusiastic early customer base. Everything specific suggests viability. But the base rate for startup survival past five years is roughly 10%. A thoughtful founder would anchor on that 10%, then ask how much each specific advantage justifies adjusting upward. In practice, most founders skip the anchor entirely. They begin with their specific story and build confidence from there, which is how 90% of startups fail while nearly 100% of founders believed they would succeed.

At the systemic level, base rate neglect produced one of the most consequential analytical failures in modern medicine. In the 1990s and 2000s, widespread PSA screening for prostate cancer led to millions of men receiving positive results and subsequent biopsies. The test had reasonable sensitivity, but the base rate of clinically significant prostate cancer in the screened population was low. The U.S. Preventive Services Task Force eventually concluded that for men of average risk, the harms of screening -- unnecessary biopsies, overtreatment, anxiety -- outweighed the benefits. The test was not inaccurate. The base rate was simply so low that even a good test generated far more false alarms than true detections.

Where This Breaks Down

Base rate thinking is powerful, but it has specific failure modes that limit its application.

The most common misapplication is using an inappropriate reference class. The base rate for "restaurants" is different from the base rate for "restaurants opened by chefs with ten years of fine-dining experience in growing neighborhoods." Choosing which population to draw your base rate from is itself a judgment call, and selecting too broad a reference class can be as misleading as ignoring base rates entirely. Precision in selecting the reference class is where base rate thinking becomes genuinely difficult.

Base rates can also shut down legitimate reasoning about genuinely novel situations. When someone proposes a technology or business model with no close historical precedent, insisting on a base rate from the nearest available category can be a form of false precision. The base rate for "social media companies" was meaningless in 2004 because the category barely existed.

The concept can become a rhetorical weapon. Citing "90% of startups fail" to dismiss a proposal without engaging its specific merits is base rate thinking used as a substitute for analysis rather than as a foundation for it.

Base rate neglect is sometimes rational. In domains where you have genuinely strong diagnostic information -- an expert mechanic listening to an engine, a physician examining highly specific symptoms -- the base rate should be substantially updated. The error is not in moving away from the base rate; it is in moving too far, too fast, on the basis of information that is less diagnostic than it feels.

Finally, base rates themselves can be unreliable. Published statistics about failure rates or disease prevalence are often derived from specific populations and time periods that may not match your situation. Treating a base rate as a precise number rather than an approximate range creates a false sense of rigor.

Connections to Other Concepts

Base rates are the empirical foundation of Bayesian thinking. In the Bayesian framework, the base rate is your prior probability -- the starting belief before new evidence arrives. Bayesian updating then adjusts this prior proportionally to the strength and relevance of the new evidence. Base rate neglect is, in precise Bayesian terms, the error of setting the prior to an uninformative value and letting the likelihood ratio do all the work, which produces wildly overconfident estimates.

The relationship between base rates and cognitive biases runs deep. Base rate neglect is itself a cognitive bias, but it interacts with several others. The availability bias makes vivid examples -- the friend whose startup succeeded, the lottery winner on the news -- disproportionately salient, overwriting the statistical background with anecdotal foreground. Confirmation bias compounds the problem: once you form an impression based on specific details, you selectively attend to supporting information while discounting the base rate as "just statistics."

Base rate thinking also connects to second-order thinking. First-order analysis asks "will this specific plan work?" Second-order analysis asks "what typically happens to plans like this one, and what would need to be true for this one to be an exception?" The base rate is the second-order question made quantitative -- it forces you to situate your specific case within the broader pattern of outcomes before evaluating its distinctive features.

There is an important link to survivorship bias. The base rates we informally absorb from the world are systematically distorted by the visibility of survivors. The entrepreneurs who succeed give talks and write books; the ones who failed are invisible. Our intuitive base rates are biased upward by disproportionate exposure to successes, making formal base rate research a necessary corrective to lived experience.

The Diagnostic Question

The self-test is a single question: "What usually happens in situations like this one?" Before engaging with the specific details of the plan, pitch, or opportunity in front of you, force yourself to answer this first. Find or estimate the base rate. Then, and only then, evaluate how the specific evidence should adjust that number.

The internal experience of doing this correctly feels counterintuitive and slightly deflating. You encounter an exciting opportunity, your mind races ahead to the specific reasons it will work, and then you pause and ask the question. The base rate arrives like cold water: most of these fail. The temptation is to push the number aside and return to the warmth of the specific narrative. Resisting that temptation -- holding the base rate in view while you evaluate the evidence -- is the discipline. It does not feel like insight. It feels like restraint.

The trigger situation is any moment when you are evaluating a specific case and feeling increasingly confident. Rising confidence in the presence of a vivid, detailed narrative is the exact condition under which base rate neglect is most dangerous. That is when the question needs to fire.

The Oximeter, Revisited

The pulse oximeters were not defective. The problem was that millions of people interpreted readings without reference to the single most important number: how likely they were to actually have the condition in the first place. A positive reading in a population with a 1% disease base rate means something radically different from the same reading in a population with a 30% base rate. The device cannot tell you which population you belong to. Only the base rate can do that. The best thinkers do not ignore specific evidence in favor of statistics, nor statistics in favor of evidence. They start with the base rate, engage seriously with the particulars, and update proportionally. That is not pessimism or optimism. It is calibration.