A correlation quietly becomes drove and one experiment becomes proves as the wording climbs

What it is

The chain from data to stated conclusion has several typical wording upgrades: an observed correlation becomes "drives" ("the new feature drove retention gains"), a single experiment becomes "proves" ("the experiment proved users prefer A"), a short sample becomes "trend" ("the growth trend is established"), a sample statistic becomes "users want." Each upgrade makes the statement punchier and easier to spread—and moves it further from what the data actually supports. Wording inflation is rarely deliberate falsification; it accumulates through retelling, reporting, and propagation—each retelling wears away a few cautious qualifiers ("possibly," "in this experiment," "directionally").

Why it happens

Inflation happens through language's economy and the selection pressure of transmission: qualifiers lengthen sentences and add cognitive load, so they are the first casualties in oral reports and social media; "proved" travels better than "is associated with," so it gets copied more. The upgrades follow stable patterns: causalization (correlational wording → causal wording, extending into the same group's juxtaposition hints), universalization (sample wording → population wording, ignoring sampling scope), permanence (point-in-time wording → trend wording, calling one quarter of data a "trend"), and mechanization (phenomenon wording → mechanism wording, sliding from "users clicked more" to "users prefer this design"). The anchoring defense is fixing "the boundary of data support" at the source: when each chart's conclusion sentence is written, tag it with its evidence grade (observational correlation / single experiment / replicated finding), so any reteller's upgrade immediately shows as a gap from the grade. The direct implication for chart makers: every sentence on the chart and caption should be able to answer "which part of the data supports this sentence?"—if none can, it has crossed the boundary.

Where it stops holding

The tension between precision and reach is real: wording strictly within evidential limits ("in this two-week experiment, variant A's click rate exceeded variant B's") is too verbose for most communication. The practical solution is not abandoning precision but layered wording—titles allow moderate summarization, but the direction of summary must match the data ("the new feature is associated with higher retention"), with precise qualifiers in the subtitle or caption. Another boundary is who bears the responsibility for detecting inflation: every link in the propagation chain erodes, and the chart maker controls only the source—the closer the source wording sits to the evidence boundary, the less room downstream inflation has; an already-inflated source only gets worse in transit. The operational test for boundary-crossing is the substitution test: replace causal words with correlational ones and "users" with "experiment sample"—if the conclusion collapses under substitution, the original sentence already depended on semantics beyond the data.

Applying it

• Tag every explanatory chart's conclusion sentence with its evidence grade (correlational / single experiment / replicated), and match the wording to the grade.
• Maintain a "boundary vocabulary" for writing: observational data uses "is associated with / accompanies / is more likely"; experimental data uses "caused / improved (in this experiment)"; qualifiers drop only after replication.
• Retelling check: before release, have someone uninvolved restate the conclusion from the chart alone; any inflation in their retelling (correlation became causation, sample became population) shows the source wording left room for it.
• Verification: run the substitution test on every conclusion sentence in the report (causal → correlational words, population → sample); any sentence that collapses is an overreaching statement.

Related

• Same group: U10.03.1 Juxtaposing graphics implies causation · U10.03.2 Uncontrolled variables must be stated explicitly
• Nearby: U9.02.3 A title's assertion must be supported by the graphic · U10.03.2 Uncontrolled variables must be stated explicitly
• Search terms: overclaiming · hedging · evidence grading