CRITERIA Subcommand (LOGISTIC REGRESSION command)

CRITERIA controls the statistical criteria that are used in building the logistic regression models. The way in which these criteria are used depends on the method that is specified on the METHOD subcommand. The default criteria are noted in the description of each keyword below. Iterations will stop if the criterion for BCON, LCON, or ITERATE is satisfied.

BCON(value). Change in parameter estimates to terminate iteration. Iteration terminates when the parameters change by less than the specified value. The default is 0.001. To eliminate this criterion, specify a value of 0.

ITERATE. Maximum number of iterations. The default is 20.

LCON(value). Percentage change in the log-likelihood ratio for termination of iterations. If the log-likelihood decreases by less than the specified value, iteration terminates. The default is 0, which is equivalent to not using this criterion.

PIN(value). Probability of score statistic for variable entry. The default is 0.05. The larger the specified probability, the easier it is for a variable to enter the model.

POUT(value). Probability of conditional, Wald, or LR statistic to remove a variable. The default is 0.1. The larger the specified probability, the easier it is for a variable to remain in the model.

EPS(value). Epsilon value used for redundancy checking. The specified value must be less than or equal to 0.05 and greater than or equal to 10-12. The default is 10-8. Larger values make it harder for variables to pass the redundancy check—that is, they are more likely to be removed from the analysis.

CUT(value). Cutoff value for classification. A case is assigned to a group when the predicted event probability is greater than or equal to the cutoff value. The cutoff value affects the value of the dichotomous derived variable in the classification table, the predicted group (PGROUP on CASEWISE), and the classification plot (CLASSPLOT). The default cutoff value is 0.5. You can specify a value between 0 and 1 (0 < value < 1).

Example

LOGISTIC REGRESSION VARIABLES = PROMOTED WITH AGE JOBTIME RACE
 /CATEGORICAL RACE
 /METHOD BSTEP
 /CRITERIA BCON(0.01) PIN(0.01) POUT(0.05).
  • A backward stepwise logistic regression analysis is performed for the dependent variable PROMOTED and the independent variables AGE, JOBTIME, and RACE.
  • CRITERIA alters four of the statistical criteria that control the building of a model.
  • BCON specifies that if the change in the absolute value of all of the parameter estimates is less than 0.01, the iterative estimation process should stop. Larger values lower the number of required iterations. Notice that the ITER and LCON criteria remain unchanged and that if either of them is met before BCON, iterations will terminate. (LCON can be set to 0 if only BCON and ITER are to be used.)
  • POUT requires that the probability of the statistic that is used to test whether a variable should remain in the model be smaller than 0.05. This requirement is more stringent than the default value of 0.1.
  • PIN requires that the probability of the score statistic that is used to test whether a variable should be included be smaller than 0.01. This requirement makes it more difficult for variables to be included in the model than the default value of 0.05.