Abstract
Serial autocorrelation equal to zero. Precision for the estimation. The precision of nine procedures for the calculation of the autocorrelation was evaluated in a Groups x Occasions design. Whe have used six different structures of dispersion matrices (Σ) that have absence of serial autocorrelation. These were: Compound Symmetric, Huynh-Feldt, Unstructured (ε= .56 y ε= .75) and of Random Coefficients (ε= .56 y ε= .75). The results show that the Hearne, Clark & Hatch (1983) procedure reaches a right estimation independently of the sample size and of the number of series points (q) except when Σ is of Compound Symmetric or Huynh-Feldt. The rest of the procedures depend on q and on Σ, and only the Jones (1985) and Pantula & Pollock (1985) procedures depend significantly on sample size. The procedures of Wilson, Hebel & Sherwin (1981), Gill (1992) and Pantula & Pollock (1985), in this order, are the most affected by the marriage absence of autocorrelation-absence of sphericity.
