Type 1 error and power of the linear trend test in proportions under the National Toxicology Program's modified...

This paper investigates the relationship between the probability of detecting a statistically significant linear trend in the proportion of animals observed to have tumors for the modified pathology protocol currently being employed by The National Toxicology Program. We considered over 300 possible dose-response relationships and tested at 5% and 1% significance levels. The results indicate that a small loss in the probability of detecting a statistically significant result will occur when the modified protocol is used instead of the previous pathology protocol. This loss varies as a function of the design of the study in terms of dose and animal allocation and as a function of the steepness of the dose-response relationship. The absolute difference between the probability of a statistically significant result for the original pathology protocol and that for the modified pathology protocol never exceeded 0.03. This difference was always less than 8% of the probability of a statistically significant result for the original pathology protocol.