How do we persuade biologists of the futility of significance testing between mean responses (ordinate values of y), at a series of dose-rates (abscissa values of x), along response curves? Statistical textbooks never recommend doing so, and they usually argue caution in the making of numerous comparisons even between unordered 'treatments', let alone between ordered ones. Yet the malpractice persists, at least in biological journals. Curves, or more usually jointed straight lines with vertical standard error bars calculated individually for each observed doserate, become decorated with a critical difference bar based on Student's t ('LSD') or on Tukey's Q, or may be littered with lower-case 'a'-'b'-'c'-labels indicating significance groupings based on Duncan's 'new multiple range test' of 1955. The sad consequence of this syndrome is that useful inferences are usually lost, obscured by a network of actual or implied interdependent tests. In some cases quite important lines of investigation have been dropped, because genuine trends were not detected. It may be helpful to bring out at least some of the reasons why biologists fail to use the appropriate analyses. I find such reasons fall into three groups:
No takes yet. Share an insight, caveat, or question.
H. C. Dawkins (1983) studied this question.