What does autocorrelation mean?
Autocorrelation is a statistical measure of how strongly a sequence of data points relates to shifted versions of itself. Rather than comparing two separate variables, it compares a variable with its own past or neighbouring values: if today's temperature resembles yesterday's more than chance would suggest, the series shows positive autocorrelation. Analysts compute it across various lags using tools such as the autocorrelation function, and it is central to time-series analysis in economics, meteorology, and signal processing. Its presence is often treated as a diagnostic warning, because many classical statistical methods assume independent observations, which autocorrelated data violate. The term combines the Greek prefix auto-, meaning self, with correlation, and first gained currency in the early twentieth century alongside the development of modern time-series theory. Understanding autocorlation-like structure in data allows researchers to build better forecasting models and avoid overstating the certainty of their results.
nounThe correlation of a signal, time series, or dataset with a delayed or shifted version of itself, used to measure how strongly its values at one point relate to its own values at other points.
- (statistics and signal processing) The correlation of a series with a delayed copy of itself as a function of time lag.
- (more generally) The degree to which values in a sequence are similar to nearby values.
"The autocorrelation of the monthly rainfall data revealed a clear seasonal pattern repeating every twelve months."
"Significant autocorrelation in the residuals suggested the regression model was missing an important variable."
"High spatial autocorrelation means neighbouring regions tend to have similar values, violating the assumption of independent observations."
The plural is common when referring to autocorrelation coefficients computed at multiple lags, or to autocorrelation structures across several variables.
"The table reports the autocorrelations at lags one through ten for each series."
Autocorrelation is how a dataset talks to itself — and it's the mathematical reason you can predict tomorrow's weather from today's.
Reviewed by Deb Chak, Editor. AI-assisted content curated by RJS Tech Solutions LLP.
Etymology of autocorrelation
The word autocorrelation joins auto-, from Greek autos meaning 'self', with correlation, formed from Latin com- ('together') and relatio ('a carrying back'), via correlatio in Medieval Latin. It emerged in the early twentieth century within statistics and time-series analysis, as mathematicians formalised ways to describe serial dependence in data. Its cognates include other auto- compounds such as autonomy and automatic, and fellow correlation derivatives such as correlate.
Related word forms
How autocorrelation is actually used
Autocorrelation is a technical term used chiefly in statistics, econometrics, signal processing, geostatistics, and climate science. It is neutral in connotation but often carries an implicit warning in statistical contexts, since unaccounted-for autocorrelation can invalidate standard significance tests. In British English both /ˌɔː-/ and /ˌəʊ-/ initial vowel variants are heard; the stress falls consistently on the fourth syllable.
Easily confused with autocorrelation
Correlation measures the relationship between two different variables, whereas autocorrelation measures the relationship between a single variable and lagged copies of itself.