A Note on Implementing a Special Case of the LEAR Covariance Model in Standard Software

07/26/2017 ∙ by Sean L. Simpson, et al. ∙ 0 ∙ share

Repeated measures analyses require proper choice of the correlation model to ensure accurate inference and optimal efficiency. The linear exponent autoregressive (LEAR) correlation model provides a flexible two-parameter correlation structure that accommodates a variety of data types in which the correlation within-sampling unit decreases exponentially in time or space. The LEAR model subsumes three classic temporal correlation structures, namely compound symmetry, continuous-time AR(1), and MA(1), while maintaining parsimony and providing appealing statistical and computational properties. It also supplies a plausible correlation structure for power analyses across many experimental designs. However, no commonly used statistical packages provide a straightforward way to implement the model, limiting its use to those with the appropriate programming skills. Here we present a reparameterization of the LEAR model that allows easily implementing it in standard software for the special case of data with equally spaced temporal or spatial intervals.

READ FULL TEXT