Why Facts Need Context To Be Useful

"This drug cuts your risk of heart attack by 22 percent" is a true statement that has been used, more or less verbatim, to sell cholesterol-lowering statins to the public. It is also, without the number sitting next to it, a badly incomplete one. The absolute risk reduction behind that 22 percent relative figure is often closer to 1 percentage point, meaning roughly 100 people need to take the drug for years to prevent one heart attack in one of them. Both numbers are correct. Only one of them tells you what you're actually deciding.

A fact is a coordinate, not a map

A single fact tells you that one point is true. It doesn't tell you how that point relates to the points around it, which is exactly what you need to predict what happens in a situation the fact never explicitly covered. Knowing that "water boils at 100 degrees Celsius" is a fact; knowing that boiling point depends on atmospheric pressure, and therefore water boils at a lower temperature at high altitude, is context that lets you predict an outcome (undercooked food at altitude) the original fact never mentioned.

This distinction matters because facts are what content optimized for speed tends to preserve, headline numbers, social captions, single-line trivia, while context is what gets cut first when information gets compressed. A stat that took a research team years to produce, with a defined population, a comparison group, and a measurement window, gets reduced to nine words and a percent sign. The number survives the compression. The conditions that made the number meaningful do not.

Cognitive scientists studying how people hold scientific knowledge have documented this exact failure pattern in learners. Physics education researcher Andrea diSessa's "knowledge in pieces" framework, developed through classroom studies at UC Berkeley starting in the 1980s, found that students frequently hold isolated, technically correct facts about physical phenomena that never connect into a usable model (diSessa, "Knowledge in Pieces," in Constructivism in the Computer Age, 1988). A student can correctly state that heavier objects don't fall faster and still predict, in an unfamiliar setup, that they will. The fact is stored. The context that would let the fact generalize is not.

In the most-cited public-health illustration of this gap, meta-analyses of statin trials for primary prevention have reported relative risk reductions in cardiovascular events in the 20 to 30 percent range, alongside absolute risk reductions closer to 1 percentage point over roughly five years (see Table 1 below). The same trial data supports both descriptions; only one of them tells a patient how many people like them would actually need treatment to prevent one event.

Relative risk vs. absolute risk: the same number, two very different pictures

This is the single most studied and most consistently mishandled context gap in health and science journalism, and it has a name: the difference between relative risk and absolute risk. Relative risk is a ratio, how much a rate changed compared to its own starting point. Absolute risk is the plain difference between two rates, stated as a share of the whole population being studied. They can both be true of the same data set and paint wildly different pictures.

Take a simplified but realistic version of a statin trial. Suppose 3 people out of 100 in a placebo group have a heart attack over five years, and 2.3 people out of 100 in the treatment group do. The relative risk reduction is (3 minus 2.3) divided by 3, which is about 23 percent, a number that sounds like it belongs in an ad. The absolute risk reduction is just 3 minus 2.3, or 0.7 percentage points, a number that sounds a lot less impressive and is a lot more useful for deciding whether the drug is worth taking.

Public health researchers have been raising this exact concern for decades. A widely cited 2003 discussion of statin communication in the Journal of the Royal Society of Medicine walked through precisely this arithmetic, showing that "22 percent risk reduction" and "roughly 1 percentage point absolute reduction" describe the identical clinical trial outcome, and argued that reporting only the relative figure systematically inflates how beneficial an intervention appears (Skolbekken, J R Soc Med, 2003). More recent meta-analyses of statin trials for primary prevention (no prior heart disease) have similarly found absolute risk reductions in all-cause mortality around 0.8 percentage points alongside relative risk reductions closer to 9 percent for the same pooled data, a gap that widens further depending on which outcome and time horizon researchers report.

Neither number is fabricated in these cases. Neither is "wrong." A relative risk reduction is genuinely useful information if you already know the absolute baseline it's built on, because it tells you how strong an effect is independent of how common the underlying condition happens to be. The problem is that relative risk, stripped of its baseline, is the number that gets quoted in headlines and drug marketing, because a percentage in the twenties reads as far more dramatic than a percentage point below one.

Why does relative risk get reported more often than absolute risk?

Relative risk numbers are almost always larger and more dramatic-sounding than absolute risk numbers for the same underlying data, because they're expressed against the smaller baseline rate rather than the whole population. A headline built on a 22 percent reduction performs better than one built on a 0.7-percentage-point reduction, even when both describe the identical trial result.

Base rate neglect: why our intuitions actively resist context

There's a deeper reason context gets dropped so easily: human probability judgment is measurably bad at holding onto background rates once a vivid, specific piece of information shows up. Psychologists call this base rate neglect, the tendency to underweight general statistical information (how common something is overall) in favor of specific-seeming evidence about the case at hand, even when the specific evidence is less reliable than it feels.

The classic demonstration is the taxicab problem, introduced by Amos Tversky and Daniel Kahneman in their research on judgment under uncertainty. In the setup, a city has two cab companies, Green (85 percent of cabs) and Blue (15 percent of cabs). A hit-and-run involves a cab, and a witness who is correct 80 percent of the time under similar test conditions identifies it as Blue. Most people, asked for the probability the cab really was Blue, answer around 80 percent, matching the witness's reliability. The Bayesian answer, once you fold in the base rate of 85 percent Green cabs, is close to 41 percent (Tversky and Kahneman, "Evidential Impact of Base Rates," in Judgment Under Uncertainty: Heuristics and Biases, 1982). People weren't ignoring the base rate entirely, they were dramatically underweighting it relative to the vivid, case-specific witness testimony.

This matters for how facts travel because a fact delivered without its base rate reads as more diagnostic than it actually is. Tell someone "a new screening test correctly flags 90 percent of cases," and most people intuitively treat that as meaning a positive result is 90 percent likely to be real. If the condition being screened for actually affects 1 in 1,000 people, and the test also produces false positives in even a small fraction of the healthy population, a positive result can still be far more likely to be a false alarm than a true case, exactly the kind of arithmetic that trips up practicing physicians, not just laypeople, in published studies of diagnostic reasoning.

How does base rate neglect connect to facts losing their context?

A base rate is a form of context, it tells you how common something is before any new evidence arrives. When a fact gets reported without its base rate (a test's accuracy without the disease's prevalence, a risk reduction without the starting risk), readers aren't just missing a detail. They're primed by a well-documented cognitive bias to overweight the vivid number they were given and underweight the background rate they weren't.

Why context determines whether a fact generalizes

A fact learned in one specific setting, "the study found a 40 percent reduction," carries an implicit scope: reduction in what, measured how, compared to what baseline, in which population, over what time period. Strip that scope away and the fact becomes a floating number readers will apply far beyond where it was actually shown to hold. This is a persistent failure mode in health and nutrition reporting, where a finding from a narrow study population, often small, often short-duration, gets generalized into blanket advice the original research never supported.

The table below shows what the same underlying statistic looks like with and without its context, using realistic figures modeled on the statin literature discussed above.

Version	What's stated	What's missing	What a reader concludes
Bare statistic	"Cuts heart attack risk by 22 percent"	Baseline risk, absolute reduction, population studied, time horizon	The drug is highly effective for almost anyone
With context	"Reduces five-year heart attack risk from about 3 in 100 to about 2.3 in 100 in adults with elevated cardiovascular risk, based on pooled trial data"	Nothing critical	The benefit is real but modest, and depends on starting risk
Bare statistic	"Test detects 90 percent of cases"	Disease prevalence, false positive rate	A positive result is very likely a true case
With context	"Test detects 90 percent of cases, but the condition affects roughly 1 in 1,000 people, so most positive results in a general screening population are false alarms"	Nothing critical	A positive result needs confirmation before it's acted on

The number in each pair is identical or nearly so. Only the context changes what a reasonable person would conclude from it.

How can you tell if a fact you've encountered has enough context to act on?

Ask whether you could explain, from the fact alone, what would make the outcome different in a related but distinct situation. If changing an obvious variable, baseline rate, population, timeframe, measurement method, leaves you unable to predict whether the fact still holds, you likely have an isolated statistic rather than something you can safely act on.

Context is what makes a fact falsifiable in the right places

A bare fact is hard to challenge productively because it doesn't specify the conditions under which it should be true or false. "This drug reduces heart attacks by 22 percent" isn't something you can meaningfully argue with beyond disputing the trial itself. But "this drug reduces the five-year heart attack rate from 3 percent to 2.3 percent in adults with elevated LDL cholesterol, based on pooled randomized trials" gives you something specific enough to test: does that hold in populations with different baseline risk, does the absolute benefit shrink or grow over longer follow-up, does it hold for all-cause mortality the way it holds for cardiovascular events specifically. That specificity is what lets a claim be checked, refined, or overturned, rather than simply repeated indefinitely in whichever direction serves the person repeating it.

This is also where context does double duty for skepticism. A statistic with a named source, a named year, a stated sample size, and a stated comparison group is a claim you can go verify. A statistic floating free of all four is, functionally, an assertion dressed up as data. Readers trying to evaluate whether a science claim is too good to be true, a skill covered in more depth in how to tell if a science claim is too neat, are really asking a context question first: does this number come with enough surrounding detail that I could independently check it, or is it a bare assertion wearing a percent sign.

Where this goes wrong at scale: a documented case

The relative-versus-absolute risk gap isn't a hypothetical classroom example. It has shaped real public behavior. One of the most-cited real-world illustrations is the United Kingdom's 1995 "pill scare," in which the UK Committee on Safety of Medicines issued a warning that certain third-generation oral contraceptive pills roughly doubled the risk of venous thromboembolism (blood clots) compared to earlier formulations, a genuinely accurate relative risk figure. What the initial warning did not prominently convey was the absolute risk: venous thromboembolism in young, healthy women taking oral contraceptives was already rare, on the order of about 1 in 7,000 users per year for older pills, rising to roughly 2 in 7,000 for the newer ones under scrutiny, an absolute increase of about 1 in 7,000. UK researchers subsequently documented a sharp, measurable spike in unintended pregnancies and abortions in the months following the warning, as large numbers of women stopped taking oral contraceptives entirely in response to the "doubled risk" framing rather than switching pill types or discussing the small absolute change with a physician (Furedi, "The public health implications of the 1995 'pill scare'," Human Reproduction Update, 1999). The relative number was accurate. Without the absolute baseline sitting next to it, it produced a public reaction badly out of proportion to the actual change in individual risk.

This case is frequently cited in medical statistics and risk-communication literature precisely because both numbers involved, "doubled risk" and "about 1 additional case per 7,000 users," were true simultaneously. The scare wasn't caused by a false statistic. It was caused by a true statistic delivered without the context that would have let people weigh it correctly against their own circumstances.

What people get wrong about "just give me the number"

A common reaction to all of this is to assume the fix is simple: always report absolute numbers instead of relative ones, and the confusion goes away. That's an improvement, but it's not a complete fix, because absolute numbers carry their own context requirements. "1 additional case per 7,000 users" still needs you to know that this is a yearly rate, not a lifetime rate, what population it was measured in (young, otherwise healthy women, not everyone), and what the comparison pill actually was. Drop any of those and the absolute number can mislead in its own way, just less dramatically than the relative version usually does.

The other common mistake is treating "more context" as synonymous with "more words." Context isn't bulk, it's the specific handful of variables that determine whether a fact generalizes: baseline rate, population, timeframe, comparison group, and measurement method. A sentence can add fifty words of hedging and caveats without supplying a single one of those five things, and a sentence can supply all five in under twenty words if it's precise. The goal when reading or writing a statistic isn't maximal caution, it's making sure those specific coordinates are present, because those are the ones that let a reader correctly extend the fact to their own situation, or correctly recognize when it doesn't apply.

It's also worth being honest about what statistical context can't do. Even a fully contextualized statistic, baseline rate, population, timeframe, source, and comparison group all present, doesn't tell an individual reader with certainty what will happen to them personally. Population-level risk reductions describe average outcomes across a study group; they don't predict any one person's outcome, and reputable reporting on medical statistics should distinguish "this changes the average risk across a population like yours" from "this is what will happen to you." Context reduces the chance of misapplying a fact. It doesn't eliminate individual uncertainty, and treating a well-contextualized statistic as a personal guarantee is its own kind of error.

What this means

A fact without context is a trivia answer: correct, memorable, and largely inert for actually reasoning about the world or making a decision with it. The relative-versus-absolute risk gap and the well-documented tendency toward base rate neglect aren't edge cases, they're the two most common, most studied mechanisms by which an accurate number gets misread at scale, from cholesterol drug marketing to a national contraceptive scare. Context, the baseline rate, the population, the timeframe, the comparison group, is what converts a bare statistic into something you can extend to a new situation, challenge with evidence, or correctly recognize as not applying to you. The number rarely lies. What usually goes missing is everything standing next to it.

---

Read next:

• How to build a better mental model
• How to tell if a science claim is too neat
• How to read a research paper without getting lost

FAQ

Q: What's the difference between relative risk and absolute risk?

Relative risk describes how much a rate changed compared to its own starting point, expressed as a percentage. Absolute risk is the plain difference between two rates as a share of the whole population studied. A drug can genuinely cut relative risk by 22 percent while only lowering absolute risk by about 1 percentage point, because the relative figure is measured against a small baseline rate.

Q: What is base rate neglect?

Base rate neglect is a documented cognitive bias, first demonstrated in depth by Amos Tversky and Daniel Kahneman's taxicab problem, where people underweight general statistical background information in favor of vivid, specific-seeming evidence, even when that background information is more diagnostic. It explains why a test's accuracy rate feels more informative than it actually is once you factor in how rare the condition being tested for actually is.

Q: Why is a technically true fact sometimes still not useful?

A true fact only tells you one specific point is accurate; it doesn't tell you the mechanism, scope, or baseline conditions that connect it to related situations. Without that context, you can't reliably predict whether the fact still applies when circumstances, like the population or measurement window, change even slightly.

Q: Did a real public health warning cause harm by leaving out context?

Yes. The UK's 1995 oral contraceptive "pill scare" warned that certain pills doubled the relative risk of blood clots, which was accurate, but did not clearly convey that the absolute risk increase was only about 1 additional case per 7,000 users per year. Researchers documented a measurable rise in unintended pregnancies afterward as many women stopped taking the pill entirely in response to the relative-risk framing alone.

Q: What's a practical test for whether a statistic has enough context to act on?

Check whether it includes a baseline rate, the population studied, the timeframe, and what it was compared against. If you can't say what would happen to that number if one of those variables changed, the fact you're holding is likely a bare statistic rather than something you can safely apply to your own situation.

Q: Does more context always mean more words?

No. Context means supplying five specific coordinates, baseline rate, population, timeframe, comparison group, and measurement method, not adding general hedging or caveats. A short sentence can supply all five and be more useful than a long, cautious one that supplies none.