CRITERIA Subcommand (MIXED command)

The CRITERIA subcommand controls the iterative algorithm used in the estimation and specifies numerical tolerance for checking singularity.

CIN(value). Confidence interval level. This value is used whenever a confidence interval is constructed. Specify a value greater than or equal to 0 and less than 100. The default value is 95.

HCONVERGE(value, type). Hessian convergence criterion. Convergence is assumed if g' _k H _k ^-1 g _k is less than a multiplier of value. The multiplier is 1 for ABSOLUTE type and is the absolute value of the current log-likelihood function for RELATIVE type. The criterion is not used if value equals 0. Specify a non-negative value and a measure type of convergence. The default is .00000001 RELATIVE.

DFMETHOD(type). Specify how degrees of freedom are computed for significance tests. This keyword can specify the options RESIDUAL, SATTERTHWAITE (default), and KENWARDROGER.

SATTERTHWAITE: The Satterthwaite method has a field degrees of freedom across tests. It is useful if your sample size is small, or the data are unbalanced, or the model uses a complicated covariance type; for example, unstructured. This is the default.
RESIDUAL: The residual method has a fixed degrees of freedom for all tests. It is useful if your sample size is sufficiently large, or the data are balanced, or the model uses a simpler covariance type; for example, scaled identity or diagonal.
KENWARDROGER: The Kenward-Roger method offers a more precise small-sample estimator for the variance-covariance of the fixed effects parameters and the approximate denominator degrees of freedom in t-tests and F-tests. The method introduces a scale factor for the F-statistic and estimates it and the denominator degrees of freedom by using a Taylor series expansion for the estimated random structure within the data. The value of the estimated denominator degrees of freedom is the Kenward-Roger degrees of freedom in GLMM. The corresponding derived F- and t-statistic are used in EMMEANS and Customs Tests for statistical inference.
Note: The Kenward-Roger method is used in the model based covariance (instead of robust covariance). When both the Kenward-Roger method and robust covariance is selected, the Kenward-Roger method is applied to model based covariance, and the following warning is presented: “Since Kenward-Roger method is selected, the robust covariance method is changed to model-based covariance method”.

LCONVERGE(value, type). Log-likelihood function convergence criterion. Convergence is assumed if the ABSOLUTE or RELATIVE change in the log-likelihood function is less than value. The criterion is not used if a equals 0. This criterion is not used by default. Specify a non-negative value and a measure type of convergence.

MXITER(n). Maximum number of iterations. Specify a non-negative integer. The default value is 100.

PCONVERGE(value, type). Parameter estimates convergence criterion. Convergence is assumed if the maximum ABSOLUTE or maximum RELATIVE change in the parameter estimates is less than value. The criterion is not used if a equals 0. This criterion is not used by default. Specify a non-negative value and a measure type of convergence.

MXSTEP(n). Maximum step-halving allowed. At each iteration, the step size is reduced by a factor of 0.5 until the log-likelihood increases or maximum step-halving is reached. Specify a positive integer. The default value is 10.

Note: When a spatial or Kronecker product covariance structure is specified on the REPEATED subcommand, the SOLUTION keyword is specified on a RANDOM subcommand, RANDOM is specified without either REPEATED or REGWGT, or KENWARDROGER or RESIDUAL is specified for the DFMETHOD on the CRITERIA subcommand, an alternative method using cubic interpolation is used to determine step sizes during iterations, and the maximum number of cubic interpolation steps is fixed at 20. The MXSTEP specification may, or may not, be used in these situations, but the number of cubic interpolations displays in the Iterations History table.

SCORING(n). Apply scoring algorithm. Requests to use the Fisher scoring algorithm up to iteration number n. Specify a non-negative integer. The default is 1.

SINGULAR(value). Value used as tolerance in checking singularity. Specify a positive value. The default value is 10 ^-12.

Example

MIXED SCORE BY SCHOOL CLASS WITH AGE
/CRITERIA = CIN(90) LCONVERGE(0) MXITER(50) PCONVERGE(1E-5 RELATIVE)
/FIXED = AGE
/RANDOM = SCHOOL CLASS.

The CRITERIA subcommand requests that a 90% confidence interval be calculated whenever appropriate.
The log-likelihood convergence criterion is not used. Convergence is attained when the maximum relative change in parameter estimates is less than 0.00001 and number of iterations is less than 50.

Example

MIXED SCORE BY SCHOOL CLASS WITH AGE
  /CRITERIA = MXITER(100) SCORING(100)
  /FIXED = AGE
  /RANDOM = SCHOOL CLASS.

The Fisher scoring algorithm is used for all iterations.