A correlation is a pattern: two measurements moving together across a set of observations, more often than chance alone would produce. A studio’s revenue and its marketing spend rising in the same quarters, or a game’s player count and its patch cadence tracking together over a year, are correlations in exactly this sense, real, checkable patterns in real data. What a correlation doesn’t establish, on its own, is which of the two things, if either, is causing the other to move. NIST’s own statistical handbook, a standard reference for the field, states the limit plainly: “causality implies association [but] association does NOT imply causality.” A cause will show up as a correlation. Plenty of correlations have no causal relationship behind them at all.
The handbook’s discussion of scatter plots, the simplest tool for spotting a correlation in the first place, is direct about what the tool can and can’t do: a scatter plot “reveals relationships or association between two variables,” showing whether they move together, in a straight line or a curve, and whether the pattern holds across the whole range of the data or only part of it. But, the handbook continues, “a scatter plot can never ‘prove’ cause and effect”, that conclusion requires “the researcher (relying on the underlying science/engineering)” to reason about the mechanism connecting the two variables, something the plot itself has no way to show. Two variables can move together because one causes the other, because both are being driven by some third factor neither chart shows, or because of nothing more than coincidence over a short enough window.
The industry’s own common confound is size. A larger studio tends to have both a bigger marketing budget and higher revenue than a smaller one, for reasons that have little to do with marketing spend actually driving revenue: it has more games in its catalog, more platforms, more existing audience. Comparing marketing spend to revenue across studios of wildly different sizes without accounting for that will produce a strong-looking correlation that’s mostly just restating that big studios are big on every measure at once, not a discovery about what marketing spend specifically buys. The handbook’s own prescription for actually separating a cause from a correlation is a designed experiment, deliberately varying one factor while holding others constant and measuring what changes, precisely because that’s the setup that lets a third factor’s influence be ruled out rather than merely hoped away. Most industry reporting doesn’t have the luxury of a designed experiment; two companies making genuinely different choices under genuinely different circumstances is the normal state of the industry, not a controlled trial.
That gap is exactly why a plausible causal story in industry reporting deserves the same scrutiny a bare correlation does. “Company X’s revenue rose after it adopted engine Y” describes an association in time, not a demonstrated cause; the same quarter could carry a new game launch, a platform expansion, or a marketing push that had nothing to do with the engine switch, any of which could be doing the actual work the headline credits to the engine. None of this means correlations aren’t worth reporting, a real pattern is real information, worth naming as exactly that: a pattern, not yet an explanation. It means treating “X happened, then Y happened” as the start of a question rather than the end of one, and reserving language that implies a mechanism, caused, because of, drove, for cases where an actual mechanism has been shown, not merely for cases where a chart happens to show two lines moving in the same direction.