Concept Architecture
Concept
Theoretically, Autoregressive Correlation is a statistical correlation structure that describes the dependence between repeated observations measured on the same individual or unit over time. It assumes that observations closer together in time are more highly correlated than observations further apart, with correlation declining systematically as the time interval increases. The concept is founded on time series analysis and longitudinal modelling and exists to represent temporal dependence within correlated data.
Mathematically, Autoregressive Correlation is most commonly represented by a first-order autoregressive, AR(1), correlation structure in which the correlation between two observations depends on the distance separating them in time. The correlation declines exponentially according to a single autoregressive parameter, producing a valid positive-definite correlation matrix suitable for repeated measures and longitudinal analyses.
In practice, Autoregressive Correlation is estimated from longitudinal or repeated-measures data using maximum likelihood or restricted maximum likelihood estimation within mixed-effects models, generalised estimating equations and repeated-measures analyses. It is widely applied in clinical trials, health economic studies and observational research involving serial measurements of costs, utilities, biomarkers and clinical outcomes.
Purpose
Used to model temporal correlation among repeated observations, improve estimation efficiency, obtain valid standard errors and support longitudinal analyses in clinical and health economic research.
Mathematical Formulae
Primary Formula
Corr(Y?, Y?) = ?^|i?j|
where:
- Corr(Y?, Y?) = correlation between observations i and j
- ? = first-order autoregressive correlation parameter
- |i ? j| = time lag between observations
Supporting Formulae
R =
[1 ? ?� ?]
[? 1 ? ?]
[?� ? 1 ?]
[? ? ? ?]
?1 < ? < 1
Cov(Y?, Y?) = ��?^|i?j|
where:
- �� = variance of repeated observations
Related Mathematical Methods
- AR(1) Correlation Structure
- Mixed-Effects Model
- Generalised Estimating Equations
- Longitudinal Analysis
- Repeated Measures Analysis
- Restricted Maximum Likelihood Estimation
- Covariance Structure Modelling
Example
A clinical trial measures patient utility scores every three months for two years.
An AR(1) correlation structure is specified with:
? = 0.80
The estimated correlation between adjacent measurements is:
Corr(Y?, Y?) = 0.80
The estimated correlation between measurements two visits apart is:
Corr(Y?, Y?) = 0.80� = 0.64
The model therefore reflects progressively weaker correlation as the interval between observations increases.
Excel Implementation
| Function | Example Formula | Health Economics Application |
|---|---|---|
| POWER | =POWER(B2,ABS(C2-D2)) | Calculates AR(1) correlation for a specified time lag. |
| ABS | =ABS(C2-D2) | Calculates the time lag between repeated observations. |
| POWER | =E2*POWER(B2,ABS(C2-D2)) | Calculates covariance using variance and autoregressive correlation. |
| MMULT | =MMULT(A2:D5,E2:H5) | Performs matrix operations when constructing autoregressive covariance matrices. |
VBA (Optional)
A VBA macro can automatically construct autoregressive correlation and covariance matrices for repeated-measures and longitudinal health economic models.
Sources
- Diggle PJ, Heagerty P, Liang KY, Zeger SL. Analysis of Longitudinal Data. 2nd ed.
- Fitzmaurice GM, Laird NM, Ware JH. Applied Longitudinal Analysis. 2nd ed.
- Pinheiro JC, Bates DM. Mixed-Effects Models in S and S-PLUS.
- Verbeke G, Molenberghs G. Linear Mixed Models for Longitudinal Data.
- Briggs A, Claxton K, Sculpher M. Decision Modelling for Health Economic Evaluation.
Related Concepts (2)
Library
Publications
1
Bayesian Methods in Health Economics — Gianluca Baio, 1st Edition ed., 2012 (Chapman & Hall / CRC Press)
An overview of Bayesian statistical methods for the analysis of health economic data, covering economic evaluation concepts, statistical cost-effectiveness analysis, Bayesian computation and MCMC, and applied health economic evaluation.
BookView source →
Frequently Asked Questions (6)
What is autoregressive correlation?
A statistical structure describing how repeated measurements on the same individual are correlated over time, closer measurements typically more strongly correlated.
Source: Box & Jenkins 1970
How does an autoregressive correlation structure model repeated measurements?
An autoregressive correlation structure models how repeated measurements taken on the same person relate to one another, assuming that observations closer together in time are more strongly correlated than those far apart. It captures the intuitive pattern that a reading today resembles yesterday's more than one from months ago, with correlation fading as the gap widens. Building this into an analysis of longitudinal data avoids treating repeated measures as if they were independent, which would understate uncertainty. Decaying correlation over time is what it represents. Kirkwood and Sterne (2003) describe such structures.
Source: Kirkwood & Sterne 2003
How does an autoregressive correlation structure work?
An autoregressive correlation structure works by specifying that the correlation between two measurements on the same individual depends on the time, or number of steps, between them, decreasing as that separation grows, commonly so that each step multiplies the correlation by a factor less than one. Adjacent measurements are thus most correlated, and distant ones least. So an autoregressive correlation structure works by modelling the declining correlation between repeated measurements with increasing time apart, which captures the intuition that recent observations are more related than distant ones, and it is used in longitudinal models to represent the dependence among a person's repeated measurements over time.
Source: Box & Jenkins 1970
When is autoregressive correlation used?
Autoregressive correlation is used in analysing longitudinal or time series data, where repeated measurements on the same individual over time are correlated, and where the correlation plausibly weakens as measurements become further apart in time. It is common in models for repeated measures, such as some mixed models and generalised estimating equations. So autoregressive correlation is used when modelling repeated measurements whose correlation declines with the time between them, which fits many longitudinal settings better than assuming all repeated measures are equally correlated, allowing the analysis to account appropriately for the pattern of dependence in data collected over time on the same units.
Source: Box & Jenkins 1970
Why does accounting for autoregressive correlation matter?
Accounting for autoregressive correlation matters because repeated measurements on the same individual are not independent, and ignoring their correlation can bias the standard errors and lead to incorrect inference, such as confidence intervals that are too narrow. Modelling the correlation structure appropriately gives valid estimates of uncertainty. So accounting for autoregressive correlation matters for correct analysis of longitudinal data, since treating correlated repeated measures as independent understates the uncertainty and can mislead, and choosing a structure such as autoregressive that reflects the declining correlation over time helps ensure the standard errors and tests properly reflect the dependence in the data.
Source: Box & Jenkins 1970
How does autoregressive correlation differ from other correlation structures?
Autoregressive correlation differs from other structures in how it specifies the correlation among repeated measures: it assumes the correlation declines with the time between measurements, whereas an exchangeable, or compound symmetry, structure assumes all pairs are equally correlated regardless of timing, and an unstructured form estimates every pairwise correlation separately. So autoregressive correlation is one of several structures for repeated measures, distinguished by its assumption that closeness in time means stronger correlation, which suits data where dependence fades over time, while other structures suit different patterns, and the choice affects the model, so it is guided by the nature of the data and the plausibility of the assumed pattern.
Source: Box & Jenkins 1970
Trust Record
Verified by Dr Darrin Baines
British health economist
Professional identity: darrinbaines.org
Verification date: 11 Dec 2025
Content version: 1.0.0
Canonical Identity
- Term code
- HE-ES-SA-008
Stable URI · Machine-readable · Resolvable · CC BY 4.0