A multiple-step selection procedure with sequential protection of preferred treatments

Area of Science:

• Statistics
• Biostatistics
• Experimental Design

Background:

• Dunnett's one-step many-one test (1955) compares multiple treatment means against a control.
• Existing methods lack procedures for comparing treatments after the control is rejected.
• There is a need for methods that account for a pre-defined order of preference among treatments.

Purpose of the Study:

• To propose a multiple-step selection procedure extending Dunnett's test.
• To provide statistical protection for preferred experimental treatments.
• To guarantee high selection probability for the correct treatment when appropriate.

Main Methods:

• Developed a multiple-step procedure building upon Dunnett's one-step many-one test.
• Incorporated an assumption of decreasing preference among experimental treatments.
• Derived methods for sample size calculation for normal, binomial, and exponential data with random censoring.

Main Results:

• The proposed procedure offers enhanced protection for preferred treatments.
• It guarantees a high probability of selecting the correct treatment under specific conditions.
• Sample size calculation methods are provided for various data types and censoring.

Conclusions:

• The multiple-step selection procedure is a valuable extension of Dunnett's test.
• This method improves the analysis of experiments with ordered treatment preferences.
• The provided sample size calculations facilitate robust experimental design and data collection.