Quantification of the impact of ensemble size on the quality of an ensemble gradient using principles of hypothesis testing