Regression analysis of log-transformed data: Statistical bias and its correction

Abstract

Abstract Power and exponential models are used frequently in environmental chemistry and toxicology. Such models can generate biased predictions if derived with least-squares, linear regression of log-transformed variables. An easily calculated but seldom used estimate of bias can enhance the accuracy of subsequent predictions. This prediction bias and means of correcting it are presented, along with several examples.