Earlier this year, South Africans were again presented with crime data showing that certain interventions were “linked to” reductions in violent crime. Depending on where you sit politically, that phrase either confirmed what you already believed or sounded like spin.
But very few people paused to ask what “linked to” actually means in research terms. The assumption was immediate: if it is linked, it must have caused the decline.
This pattern is everywhere. A health behaviour is “linked to” a disease outcome. Load-shedding is “linked to” increased mortality. Screen time is “linked to” depression in adolescents. By the time the public encounters the claim, the authority of science has already been invoked. The phrase carries weight, it settles something.
The trouble is what “linked to” actually means. In research terms, it signals correlation – two things move together in data. In public debate, it reads as causation. The problem is obvious.
In observational research, especially in the social sciences and public health, confounding variables are everywhere: poverty, geography, education, policing intensity, reporting rates. When crime drops in one district after a new initiative is introduced, it may be because of that initiative. It may also be because of demographic shifts, economic changes, or simple regression to the mean. Untangling those threads is difficult. That is why researchers are cautious.
Once the finding is translated into public language, caution tends to disappear. “Linked to” becomes “caused by”. A tentative association becomes a vindication of policy, ideology, opinion. When later analysis complicates the picture, critics talk about incompetence or dishonesty. Sometimes those criticisms are justified, but often they are aimed at a stronger claim that the original research never actually made.
Rot starts earlier
Journalists get blamed for inflating findings, and sometimes they deserve it. But the rot starts earlier. Scientific careers are built on novel results, not on replicating old ones. Journals want surprising findings and funders want impact. Null results – studies showing that nothing happened – languish unpublished. The incentives push toward exaggeration before the mainstream society gets involved.
The replication crisis exposed how fragile some of those discoveries were. In psychology and parts of biomedical research, results that were widely cited and publicly discussed failed to reproduce when retested under stricter conditions. That failure was not primarily a media problem. It reflected weaknesses in methodology, statistical thresholds, and publication bias.
In other words, distortion does not begin only at the press release stage. It is sometimes baked in earlier, which complicates the story. It means that when people say “the science was wrong”, they are occasionally pointing to something real, not simply misunderstanding a technical term.
Covid magnified these tensions. Early observational data was treated as decisive. Mask efficacy, school closures, vaccine transmission effects, lab-leak hypotheses – each moved through stages of tentative evidence, strong assertion, and later revision.
Under political pressure, provisional findings were communicated with far more certainty than they warranted. When revisions came, trust in “the science” suffered. Critics interpreted change as proof of bad faith rather than as the normal process of updating under uncertainty.
Language and expectation
This is not an argument that nothing was known, or that all decisions were arbitrary. It is an argument about language and expectation. When scientific caution is compressed into political messaging, the nuance is the first casualty.
The anecdotes-versus-data dispute reveals the same fault line. During vaccine rollout, individuals who reported adverse effects were often told that their experiences did not count against population-level data showing safety and efficacy. That response was technically defensible at one level: large datasets are necessary to establish overall risk. But it was rhetorically disastrous. People do not accept that their own experience is irrelevant simply because it is statistically rare.
Anecdotes are not useless. They often signal the need for investigation. Clusters of reported side effects have led regulators to detect rare but serious complications. Whistleblowers have exposed institutional failures long before formal audits have done. Lived experience can be the starting point of serious inquiry.
But anecdotes cannot, on their own, tell us how common a phenomenon is, or whether it outweighs other risks. A handful of dramatic cases cannot establish a general pattern. That is what data is for. It aggregates, compares, and contextualises. It provides baselines, keeping “perception drift” in check.
The conflict arises when each side overreaches. Personal stories are treated as definitive proof of systemic truth. Statistical summaries are treated as a way of shutting down uncomfortable narratives. Both moves are common in public debate and neither is intellectually honest.
Not the same idea
Then there is the word “significant”. In everyday speech, it means important. In statistical practice, it refers to the likelihood that an observed effect is not due to random chance under a specified model and threshold, often expressed through a p-value. These are not the same idea.
With sufficiently large samples, even minuscule differences can reach statistical significance. A tiny reduction in crime or a marginal change in test scores can cross a formal threshold while being practically meaningless. Conversely, a genuinely important effect can fail to reach statistical significance in a small study, particularly where data is noisy or hard to collect.
When a headline declares that a finding is “significant”, readers infer importance. That inference is understandable, but it is also frequently wrong. The statistical label does not answer the policy question. It simply tells us something about the reliability of the observed difference under particular assumptions.
None of this ambiguity is politically neutral. Activists understand how these terms are heard. Corporations understand it too. “Linked to” sounds accusatory. “No significant effect found” sounds reassuring. The space between technical meaning and public interpretation is not an accident of language; it is often used strategically.
This is why phrases like “the science says” should prompt caution. Science produces findings under conditions of uncertainty, bounded by method and model. From there, judgement begins. How much risk is tolerable? Which trade-offs are acceptable? Whose interests should be prioritised? These are not scientific questions, even when they are informed by scientific evidence.
Scientists sometimes make policy recommendations. They may be right to do so. But a recommendation is not the same thing as a finding. Treating them as identical collapses the line between evidence and value. In a democratic society, that line matters. Scientific language should be used to structure argument, not finalise it.
Proxy for moral authority
The problem is not a handful of misunderstood words. It is the way scientific vocabulary has become a proxy for moral authority. If something is “supported by the science,” opposing it feels irrational. When that support later weakens or shifts, the backlash is severe. Trust erodes not because science is useless, but because it was presented as more final than it ever was. Science, at all times and in all places, is by definition provisional.
We would argue more clearly if we were stricter about categories. An association is not a mechanism. Statistical significance is not importance. Evidence constrains judgement; it does not replace it. These are modest clarifications. They do not resolve political disagreement. But they reduce the number of disputes driven by linguistic confusion rather than substantive difference.
Respecting science does not mean treating it as an oracle. It means understanding what it can legitimately claim – and refusing to let its language carry burdens that belong elsewhere.
[Image: Iris,Helen,silvy from Pixabay]
The views of the writer are not necessarily the views of the Daily Friend or the IRR.
If you like what you have just read, support the Daily Friend