Using ANOVA when dose-response is better

Distinctions between biological sub-disciplines are blurring. Many biologists have multiple areas of expertise, and initiate new projects stretching the boundaries of their previous experience. This is a good thing, but lack of experience can crop up at the highest levels of experimental design and analysis.

A few years ago I noticed some ecological studies that examined the toxicity of Bt corn on stream ecosystems. Couched as stream ecology and conducted using a field component, the work blurred the line between ecology and ecotoxicology. They described the effects of Bt corn leaves on stream macroinvertebrates. Since Bt corn was implicated in causing “collateral damage” to non-target species, like butterflies, it stood to reason that the toxins in Bt corn could harm aquatic insects as well. The topic is insightful, but they used an ANOVA experimental design. I recently saw a poster in the hall of a university that descibed work looking at the effects of UV-radiation on zooplankton. It also used an ANOVA design.

Comparing a control group to a treatment group in the presence of Bt corn material or UV radiation certainly is a legitimate experimental design, but it may not be the best choice. That’s because the question is really one of toxicity, and that’s toxicology. In other words, were a toxicologist to approach this problem, he might take a wholly different approach.

He’d use a Dose-response design. In dose-response, treatment groups are exposed to different doses of the stressor, so as to characterize the response of the organism, the receptor. Whereas the differences in the mean number of surviving organisms is compared between treatment and control groups with ANOVA, the shape of the mortality curve is evaluated in dose-response. These curves are typically logistic, so a logistic curve is fitted to the data.

Dose-response design is the standard method of evaluating toxicity. It allows one to assign “Lethal-dose” values, or the dose required to kill, typically, 50%, 90%, or 99% of the organisms (e.g. LD50, LD90, LD99). Given a value for the LD99, you can get an idea of how toxic a stressor is, allowing direct comparison between stressors. Thus if the concentration needed to achieve the LD99 is lower for one stressor compared to another, then its more toxic compared to that other. You can also spply this to effects rather than lethality.

I suspect that ANOVA is being used where Dose-response would be better because ecologists are not generally trained in toxicology. I, for example, never heard of it until I did my own toxicological work at NOAA. This supports the notion that exposure to other fields of study can lead to great advances, by applying ideas or techniques that have been developed elsewhere.

There will be limitations if dose-response is taken out of a laboratory setting, especially if there is a maximum dose that not lethal to 100% of the recptors. Still, ecologists ought to be aware of, and use, dose-response design.