Estimation of the Smallest Acceptable Sample Size in Bilateral Approaches to Coefficient Estimation and Accuracy Prediction

Abstract

This study is about smallest acceptable sample size determination in experimental design studies involving a driving simulator. The smallest acceptable sample size should be specified so researchers can make accurate inferences about their studied populations. However, the number of samples typically collected is largely subject to the expense of data collection. Working out the methodology of estimating the required number of subjects based on an initially small number is a better way for researchers to determine the smallest acceptable sample size in the experiment. Predictor estimate precision and prediction accuracy are major factors for conducting experiments. Accordingly, this study estimates the smallest acceptable sample size, with emphasis on coefficient estimation and prediction accuracy for selected significant variables. The smallest acceptable sample size is chosen to be the maximum value returned by both coefficient estimation calculation and accuracy prediction calculation approaches. This methodology is flexible and scalable, and can be tailored to other experimental situations. To validate the appropriateness of this procedure, a more than sufficient sample of 50 drivers was recruited. The smallest acceptable sample size was determined backwardly, based on the variable coefficient convergence trends of the mean squared error (MSE) curves of the significant variables. Both the clear converging trends of the MSE curves and the proposed method indicated that 30 was an acceptable sample size.