Statistical Techniques Applied to Aerial Radiometric Surves (STAARS): Percentile Estimation with the Normal and Lognormal Distributions

摘要

Because of the prominence of percentile estimation in the analysis of NURE aerial radiometric data, a study was conducted to compare two methods of estimating percentiles. The two methods of estimation considered are (1) the normal distribution assumption method and (2) the lognormal distribution assumption method. An error made by assuming the data are from a normal population, when in fact the population is lognormal, causes problems primarily when estimating tail percentiles. When estimating upper percentiles, the error of assuming lognormality, when the data are really normal, has more serious consequences than does the error of assuming normality, when the data are lognormal. When estimaing lower tail percentiles, the error of failing to perform a log-transformation has more serious consequences than does the error of transforming. When estimating percentiles that are between the 10th and 80th, the consequence of either type of error is not great if the coefficient of variation is less than 0.27. (ERA citation 05:031632)