Base Rates: The Numbers We Forget When We Judge
Written by The Pilgrim ·
When you meet someone and form a quick judgement about them, what exactly are you drawing upon? Are you reasoning from the particular details in front of you, or are you, without realising it, ignoring a vast ocean of relevant information that exists before the encounter even begins? These are not merely technical questions for statisticians. They are questions about how all of us think, every single day, and why we so often think badly.
The base-rate fallacy is, at its core, a failure of proportion. When we are presented with a specific case, a person, a symptom, a news story, a remarkable coincidence, we tend to focus almost entirely on the vivid details of that case. We reason from the inside, absorbed by the particulars. What we neglect, almost reflexively, is the base rate: the background frequency with which the thing in question actually occurs in the broader population or context. We forget to ask, before anything else, how common is this, really?
Consider a relatively simple example. Imagine that a medical test for a rare condition is ninety-nine percent accurate. A person takes the test and receives a positive result. Most of us, hearing this, would feel a surge of alarm. The test is almost never wrong, after all. But if the condition itself affects only one person in every ten thousand, the calculation changes dramatically. Among ten thousand people tested, we would expect roughly one genuine case and approximately one hundred false positives. A positive result, therefore, is far more likely to be wrong than right, even from a test that sounds nearly infallible. The base rate of the condition, the sheer rarity of it in the general population, must be the starting point for any honest interpretation. Without it, we are reasoning in the dark, dazzled by a statistic that feels precise but tells us far less than we suppose.
The psychologists Daniel Kahneman and Amos Tversky spent much of their working lives documenting this tendency. Their experiments, conducted over decades, demonstrated that people reliably neglect base rates when presented with colourful, individuating information. In one famous class of problems, participants were told that a group consisted overwhelmingly of one type of person and only a small number of another. When then given a vivid but vague personality sketch, participants would identify the person as belonging to the rarer group simply because the description seemed to fit a stereotype. The background proportion, the simple numerical truth about the group, was effectively overridden by narrative. We are creatures who find stories more compelling than frequencies.
This is not an obscure academic finding. It touches everything from medical diagnosis to criminal justice, from investment decisions to the way we read the news. When a dramatic event is reported, we rarely ask how often such events occur relative to the ones that are never reported. The very act of reporting selects for the unusual, the striking, the deviant. We are then left with a distorted sense of frequency, convinced that certain dangers are everywhere when in truth they are rare, or conversely that certain reassurances hold when the underlying rates tell a grimmer story.
Why is the neglect of base rates so persistent? One explanation lies in the way human cognition evolved. For most of our history as a species, the relevant information for survival was immediate and particular. The rustle in the grass mattered; abstract proportions did not. We became exquisite readers of individual signals and rather poor intuitive statisticians. The mind that served us well on the savanna turns out to be somewhat ill-equipped for reasoning about populations, probabilities and background rates in the modern world. This is not an excuse for the error; it is an explanation of why correcting it requires deliberate effort.
There is also something psychologically seductive about the specific. A face, a name, a story, these things engage our sympathy and attention in ways that a percentage never can. When we are told that a disease kills forty thousand people annually, we absorb it rather coolly. When we are shown the photograph of one victim, the emotional and cognitive response is of an entirely different order. This is sometimes called the identifiable victim effect, and it is the mirror image of base-rate neglect. Both involve a failure to hold the particular and the general in proper relation to one another.
Good reasoning, on this account, requires a kind of double vision. We must attend to the particulars of the case before us while simultaneously keeping the background frequency in view. This is harder than it sounds. It means, in practice, asking a set of unglamorous questions before reaching for a conclusion. How often does this kind of thing happen? What proportion of similar cases turn out this way? If I had no specific details and only the base rate, what would I believe? These questions feel cold when set against the warm urgency of the particular, but they are the questions that honest reasoning demands.
The discipline required here is not one of suppressing intuition altogether. Intuitions carry genuine information, and a wholesale distrust of them would be its own kind of error. The goal is rather to integrate the base rate with the specific evidence, weighting each according to its actual reliability. Bayesian reasoning, loosely understood, is precisely this: beginning with a prior probability derived from background frequency, then updating it in light of new evidence. The background is the anchor. The particulars are what we adjust from.
Perhaps the deepest lesson of the base-rate fallacy is that humility about our judgements is not merely a moral virtue but an epistemic one. To judge well is to remember how much we do not know about what we do not see: all the cases that did not reach us, all the outcomes that were never reported, all the people who resembled the one before us and turned out quite differently. Can we train ourselves to feel the weight of those invisible numbers? Can we make the background frequencies as vivid and present to ourselves as the particular case demanding our attention? These remain open questions, and sitting with them honestly may itself be the beginning of better judgement.