Home Statistical Dictionary Lurking Variable
Lurking Variable
An unmeasured or unconsidered variable that influences an apparent association between two other variables, often invoked specifically in the context of correlational data to warn against inferring causation from a bivariate relationship alone.
In Plain English
A lurking variable is a hidden factor, one that wasn't measured, or perhaps wasn't even thought of, that's quietly responsible for an association you're observing between two other variables. It's essentially the same underlying idea as a confounding variable, but the term 'lurking variable' is used especially often in introductory statistics teaching, specifically as a memorable warning attached to correlation, whenever two variables are correlated, there's always a possibility that some unconsidered lurking variable, rather than either one directly causing the other, is what's actually driving the relationship. The classic phrase 'correlation does not imply causation' is, in large part, a warning about exactly this possibility.
Definition
A lurking variable is an unmeasured, or simply unconsidered, variable that influences an observed association between two other variables under study, effectively synonymous with a confounding variable in its statistical mechanism, but conventionally used, especially in introductory statistics and correlation analysis, as a cautionary term emphasizing that an observed bivariate correlation may not reflect a direct causal relationship between the two measured variables at all, but may instead result from both being driven by some third, lurking factor that was never explicitly identified or measured. Because a lurking variable is, by definition, one that was not considered or measured, its influence cannot be directly detected or adjusted for using the observed data alone, which is exactly why the possibility of a lurking variable is treated as an ever-present, irreducible caveat whenever causal claims are drawn from purely correlational or observational data, rather than something that can be checked and ruled out from within the dataset itself.
Properties
- The term lurking variable is used almost interchangeably with confounding variable in casual and introductory usage, but carries a specific pedagogical emphasis, it foregrounds the idea that the responsible third variable might not even have occurred to the researcher, as opposed to a confounder that was identified but perhaps not adequately controlled for.
- Because a lurking variable was, by construction, never measured, its existence and influence generally cannot be confirmed or ruled out using the dataset at hand alone, this irreducibility is precisely why 'correlation does not imply causation' remains sound cautionary advice regardless of how large or statistically significant an observed correlation is.
- Randomized controlled trials neutralize the threat of lurking variables in the same way they neutralize confounding variables generally, by randomly assigning treatment, any lurking variable, known or unknown, measured or unmeasured, is balanced across groups in expectation, which is exactly why experimental evidence is considered much stronger support for causal claims than a purely observed correlation.
At a Glance