An examination of methods for sample size recalculation during an experiment

Area of Science:

• Biostatistics
• Experimental Design
• Statistical Inference

Background:

• Experimental design often faces challenges in accurately estimating response variance.
• Under-powered studies can result from relying on initial variance estimates.
• Existing methods for sample size recalculation include fully sequential and multi-stage approaches.

Purpose of the Study:

• To propose a novel quasi-sequential procedure for adaptive sample size determination.
• To combine the efficiency of fully sequential methods with the practicality of two-stage designs.
• To reduce the mean squared error of the final sample size compared to existing methods.

Main Methods:

• Developed a quasi-sequential procedure using multiple imputation for data beyond the initial sample.
• Applied the stopping rule from fully sequential procedures to the extended dataset.
• Conducted simulations to compare the proposed method with a two-stage approach.

Main Results:

• The quasi-sequential procedure demonstrated a considerably lower mean squared error for the final sample size, especially when the initial sample size was underestimated.
• Simulations showed improved accuracy in sample size recalculation compared to the two-stage procedure.
• Analysis of sample size distributions provided insights into the performance of alternative procedures.

Conclusions:

• The proposed quasi-sequential procedure offers an advantageous alternative for adaptive sample size determination in experimental design.
• This method balances efficiency and practicality, leading to more accurate final sample sizes.
• The findings suggest that quasi-sequential designs can mitigate issues associated with initial variance estimation errors.