Similarly, a direct product between two covariance matrices would not work because it would result in variances squared on the diagonals. So, for model 1, in principle, we want to use residual idv(column).ar1(row)instead of residual id(column).ar1(row); the
difference is simply the character 乪v乫 appended to 乪id乫 to make it a covariance rather than correlation structure. In practice, however, ASReml will, if possible, adjust the components of the residual structure to produce an appropriate R structure. So, one can 乪get away乫 with writing residual id(column).ar1(row)in this case, but we recommend writing the correct structures as a habit, since there may be instances where one has to get the structure exactly correct, for example, when providing starting values for covariances vs correlations.