(Co)-variances

Starting values, priors, “true” values

Co-variance matrices for all random factors in the models including the residual can be specified.
This can be directly in the driver file following the $PRIOR keyword, or in a file pointed to by the
fn option of the $PRIOR keyword line.

For analyse type 1 (AI-REML), the values specified are used as starting values. If no starting values are
specified, identity matrices of appropriate dimensions are assumed for all (co)variance matrices in the model.

For analyse type 2 (MCMC), the values are used as priors with a degree of belief as specified in the
additional input to MCMC, using the $MCMC keyword.

For tnalyse type 11 and 12 all non-zero elements in all (co)variance matrices must be specified,
and are used in the model as "true" values.

The matrices are numbered as in the RANDOM model directive line. The number for the residual (co)variance
matrix is always one larger than the last factor in the RANDOM line. DMU will print a summary of the assumed
covariance structure, which can be used to check that model and priors are correctly specified.

The prior variances and covariances must be specified in random factor number sequence i.e. priors for random
factor 1 must be specified before priors for random factor 2 and so on. Each line consists of 3 integers and
a real number (free format). The first integer is the random factor number followed by row-column (trait)
combination and finally the prior (co)variance.

For estimation of variance compoments (analyse type 1 or 2) in models with heterogenious residual variancei a
general starting values to used for all residual variance groups can be specified, or specific vales for each
variance group can be specified. In this case, the lines with starting values for the residual co-variances
must have an extra column containing the code for the heterogen residual variance group.

For prediction (analyse type 11 or 12), the "true" co-variance components must be specified for all heterogen
residual variance group, and values for the residual co-variances must have an extra column containing the
code for the heterogen residual variance group.