The CAUSALMED Procedure

computes bootstrap-based confidence intervals for the effects and percentages of effects, and displays them in the "Summary of Effects" table. This table includes a column that indicates the number of bootstrap samples used to compute the confidence intervals; this column is not displayed but is available if you save the table as an output data set by using the ODS OUTPUT statement. You can also display this column by modifying the corresponding template.

You can specify one or more of the following types of bootstrap confidence intervals separated by spaces:

The ALPHA= option in the PROC CAUSALMED statement sets the confidence level for constructing bootstrap intervals. For more information about how bootstrap-based confidence intervals are computed, see the section Bootstrap Methods. By default, PROC CAUSALMED produces bias-corrected confidence intervals (BOOTCI(BC)) based on 1,000 bootstrap samples.

ignores the case-control design and performs a regular bootstrap sampling that draws observations from the entire sample with replacement. When you specify a case-control design by using the CASECONTROL option and omit the IGNORECC option, the default bootstrapping draws samples separately from the case and control groups. As a result, the bootstrap samples maintain the same numbers of case and control observations as those of the original sample. However, when you specify the IGNORECC option, it overrides the default bootstrap sampling method for case-control studies, and the numbers of case and control observations generally vary from (bootstrap) sample to sample.

If you do not specify the CASECONTROL option, the procedure performs regular bootstrapping automatically and the IGNORECC option is irrelevant.

specifies the minimum number of converged bootstrap samples that is required for conducting any bootstrap estimation, where n is between 30 and 10,000. By default, n=30.

specifies the number of bootstrap sample data sets (replicates), where n is between 50 and 10,000. By default, n=1000.

includes in the bootstrap estimation bootstrap samples that have inconsistent class levels in either the mediator or outcome model. By default, for any classification variable in a bootstrap sample data set that does not contain all levels that have been used in fitting either the mediator or outcome model for the original input data set, PROC CAUSALMED treats the corresponding bootstrap sample estimates as nonconvergent and skips the estimation of the mediation effects. This option overrides the default and continues the estimation of the mediation effects for bootstrap samples that have inconsistent class levels.

provides the seed that initializes the random number stream for generating the bootstrap sample data sets (replicates). If you do not specify this option or if you specify a value for n that is less than or equal to 0, the seed is generated from reading the time of day from the computer’s clock. The largest possible value for the seed is .

You can use the SYSRANDOM and SYSRANEND macro variables after a PROC CAUSALMED step to query the initial and final seed values. However, using the final seed value as the starting seed for a subsequent analysis does not continue the random number stream where the previous analysis ended. The SYSRANEND macro variable provides a mechanism to pass on seed values to ensure that the sequence of random numbers is the same every time you run an entire program. To reproduce the random number stream that was used to generate bootstrap estimates, you must specify the same value for the SEED= option and the same value for the THREADS= option in the PROC CAUSALMED statement.