CADDIS Volume 3: Examples & Applications
Analytical techniques used:
Type of evidence supported:
Verified Prediction with Traits
In causal analysis we find that trait information is well suited to a type of evidence called verified prediction, where the knowledge of a cause's mode of action permits prediction and subsequent confirmation of previously unobserved effects. In this application, we would predict changes in the occurrence of different traits we would expect to occur if a particular stressor was present and causing biological effects. If we found that these traits do indeed occur at the impaired site, our prediction is verified and the causal hypothesis is supported by that evidence.
Analysis and results
Analytical approaches range from basic comparisons of measurements to more formal statistical tests (see page on establishing differences from expectations). Incorporating predictions of traits into causal analysis is an area of active research, and so we present a hypothetical example below.
Existing information about the relationship between a trait and environmental gradients can be used to predict how the occurrence of a trait will differ between the test site and reference expectations. The occurrence of a trait in a community from test site is compared with a community from a reference site. If the predicted occurance of a is supported, the result would support a claim of verified prediction.
We illustrate this with an example of clinger relative richness and sediment in streams across the eastern United States. Existing literature indicates that the relative richness of clingers decreases with increased bedded sediment (Figure 1, Pollard and Yuan 2010). Based on this existing relationship, we predict that if bedded sediments are a cause of impairment in a test stream, then the relative richness of clingers should be lower in the test stream than in comparable reference streams. Then, we compare the trait data from our reference site to the trait data from the test site. If the test site has fewer clingers than the reference site, the general prediction is confirmed.
If you have a set of regional reference sites you can incorporate these data into a more rigorous analysis of verified prediction. In this case one would determine the confidence intervals for the association between clinger richness over a sediment gradient for reference sites. Then, compare the test sites to the reference sites and ask whether the test sites fall outside of the confidence interval bounds of the reference conditions.
How do I score this evidence?
If the predicted pattern is observed (here, if the test site had fewer clingers than the reference site), the type of evidence "verified prediction" is scored as supported (+). If multiple predictions were verified or if the predictions were highly specific, the evidence may be convincing (+++).