The SEQDESIGN Procedure

Boundary Variables

The boundaries created in group sequential trials depend on the type of the alternative hypothesis and the early stopping criterion. Table 89.5 shows the boundaries created with various design specifications.

Table 89.5: Boundary Variables

Specifications

 

Boundary Variables

Alternative

   

Lower

 

Upper

Hypothesis

Early Stopping

     

 Alpha 

 Beta 

     

 Beta 

 Alpha 

Lower

Accept $H_{0}$

   

X

     
 

Reject $H_{0}$

 

X

       
 

Accept/Reject $H_{0}$

 

X

X

     

Upper

Accept $H_{0}$

       

X

 
 

Reject $H_{0}$

         

X

 

Accept/Reject $H_{0}$

       

X

X

Two-sided

Accept $H_{0}$

   

X

 

X

 
 

Reject $H_{0}$

 

X

     

X

 

Accept/Reject $H_{0}$

 

X

X

 

X

X


Up to four boundaries can be generated in a group sequential design:

  • the upper $\alpha $ boundary, to reject the null hypothesis for the upper alternative

  • the upper $\beta $ boundary, to accept the null hypothesis with an upper alternative

  • the lower $\beta $ boundary, to accept the null hypothesis with a lower alternative

  • the lower $\alpha $ boundary, to reject the null hypothesis for the lower alternative

For a two-sided design, the null hypothesis is accepted only if both the hypothesis is accepted with an upper alternative and the hypothesis is accepted with a lower alternative.

For a one-sided design with a lower alternative, only the lower boundaries are created. Similarly, for a one-sided design with an upper alternative, only the upper boundaries are created. For example, Figure 89.10 shows the boundary plot for a one-sided test with an upper alternative.

Figure 89.10: Boundary Plot for One-Sided Test

Boundary Plot for One-Sided Test


Figure 89.10 corresponds to a one-sided sequential design with early stopping to reject or accept the null hypothesis. For a sequential test with early stopping only to reject the null hypothesis, there are no acceptance boundary values at interim stages. The acceptance boundary value and its associated acceptance region are displayed only at the final stage. Similarly, for a sequential test with early stopping only to accept the null hypothesis, there are no rejection boundary values at interim stages. The rejection boundary value and its associated rejection region are displayed only at the final stage.

For a two-sided design, both the lower and upper boundaries are created. For a design with early stopping to reject the null hypothesis, $\alpha $ boundaries are created. Similarly, for a design with early stopping to accept the null hypothesis, $\beta $ boundaries are created. For a design with early stopping to accept or reject the null hypothesis, both the $\alpha $ and $\beta $ boundaries are created.

For example, Figure 89.11 shows the boundary plot for a two-sided test.

Figure 89.11: Boundary Plot for Two-Sided Test

Boundary Plot for Two-Sided Test


Figure 89.11 corresponds to a two-sided sequential design with early stopping to reject or accept the null hypothesis. For a sequential test with early stopping only to reject the null hypothesis, there are no acceptance boundary values at interim stages. The acceptance boundary value and its associated acceptance region are displayed only at the final stage. Similarly, for a sequential test with early stopping only to accept the null hypothesis, there are no rejection boundary values at interim stages. The rejection boundary value and its associated rejection region are displayed only at the final stage.