Causal Analysis/Diagnosis Decision Information System (CADDIS)

Step 4: Evaluate Data from Elsewhere

• 4.1. Overview
• 4.2. In-Depth Look
• 4.3. Results and Next Steps

4.2.1. Stressor-Response Relationships from Other Field Studies

Concept
At the impaired sites, the cause must be at levels sufficient to cause similar biological effects in other field studies.

Figure 4-1. Stressor-Response Relationships from Other Field Studies. Field studies from the region that associate measurements of the candidate cause (orange dots) with the intensity of the effect (proportion of dead fish) can be used to generate an exposure-response model. If concentrations in site water at the time of the fish kill (the impairment) are in concentration range A, the candidate cause is weakened. If they are in range B it is strengthened. (General explanation of symbols)

Examples
Consider increased levels of deposited fine sediments as a candidate cause of decreased mayfly (Ephemeroptera) taxonomic richness. What findings support or weaken the case for increased fine sediments as the cause, based on stressor-response relationships from other field studies?

• Supporting evidence - Monitoring data from sites throughout the state show that mayfly taxonomic richness declines steadily once the area of stream bottom covered by more than 1 mm of fine silt exceeds 10%; in the case, 15% of the stream bottom is covered with a 1 mm thick layer of fine silt.
• Weakening evidence - Monitoring data from sites throughout the state show that mayfly taxonomic richness steadily declines as the area of stream bottom covered by more than 1 mm of fine silt exceeds 10%; in the case, only 5% of the stream bottom is covered with a 1 mm thick layer of fine silt.

How do I analyze the data?
The objective of analyzing stressor-response relationships is to provide evidence that organisms at impaired sites are exposed to the candidate cause at quantities, durations or frequencies sufficient to induce observed biological effects. Although these relationships most frequently have been used to evaluate chemicals, a similar approach can be used for other agents, such as sediment, water, and temperature (example stressor-response relationships).

Stressor-response relationships from other field studies are most compelling when they are based on many studies providing stressor-response curves. Preferably these studies would be from similar ecosystems and the effects data would include the specific taxa showing impairment within the case. However, dichotomous relationships based on presence-absence data are also used to determine the frequency of occurrence of a stressor with an effect as a stressor-response relationship.

Once stressor-response relationships are developed from other field studies, they must be related to comparable measurements of stressors and effects from the case. For example, matched data from measurements of early-morning dissolved oxygen levels and mayfly taxonomic richness may be used to generate a stressor-response relationship for the region in which the case is located. That relationship has qualitative properties such as linearity and sign of the slope as well as quantitative properties such as the value of the slope or the concentration at which mayfly taxonomic richness is reduced by a specific percentage. The relationship can be compared with data from the case in several ways, depending on the type of evidence that is available:

• If data from the case provide a Stressor-Response Relationship from the Field, that relationship may be compared qualitatively and quantitatively to the stressor-response relationship from other field studies. For example, if both relationships are linear with positive signs and a threshold, that supports the candidate cause qualitatively.
• If data from the case provides only Spatial/Temporal Co-occurrence, then the level of effect at each impaired location can be compared to the level of effect for the corresponding level of the candidate cause in the stressor-response relationship for other field studies. For example, consider an impaired location where the DO concentration is X mg/L and the number of mayfly taxa is reduced 60% of the number of mayfly taxa at a local reference site. After examaining the stressor-response relationship for DO and mayfly taxa for the region, you find that at X mg/L the number of mayfly taxa were reduced 75% on average, relative to the maximum. This outcome would provide quantitative support, because it shows that the low DO was sufficient to account for the impairment.
• Finally, if a candidate cause is expressed as a dichotomous variable, such as whether a stream reach is or is not impounded, then the frequency of association (e.g., 80%) of that variable (impoundment) with an effect (e.g., reduced mayfly taxa) in other streams could be related to the reduction in mayfly taxa in the case.

Confidence in the assessment is greatest when stressor and biological effect measurements are matched, that is, data at each site are collected and analyzed at the same time, using similar methods. Stressor-response relationships usually are based on measurements of the candidate causal agent itself, but measurements of one or more surrogates for the candidate cause also can be used. Keep in mind, however, that one measure may be a surrogate for more than one stressor.

Field stressor-response relationships are commonly evaluated by regression analysis. Quantile regression also may be a particularly useful method, but as of this writing it has not yet been applied to causal analysis (Cade and Noon, 2003). Large slope values increase confidence that the stressor-response relationship is real. Statistical tests of these relationships should be interpreted cautiously, because they are very sensitive to sample size, and stressor levels are not randomly assigned.

Co-occurring stressors can complicate analyses of stressor-response relationships from other field studies. For example, a strong relationship between a stressor and the observed effect might be observed if that stressor is correlated with another stressor, which is the true cause. For this reason, it is helpful to make use of both field-derived associations and controlled laboratory studies when evaluating a case (Stressor-Response Relationships from Laboratory Studies). In addition, examining correlations among a suite of stressor variables can provide useful insights. Multivariate techniques such as principal components analysis can be used to divide stressors into groups that increase or decrease together. This type of grouping or classification can be based on any of a number of variables, including waterbody type, known point sources, stream gradient, climate, or known land uses.

What evidence would support or weaken the case for a candidate cause?
Supports

• Data showing that a direct stressor-response relationship (i.e., increasing effect with increasing stressor) observed in the case is similar to the stressor-response relationship observed in other field studies
• Data showing that the candidate cause occurs in the case at levels frequently associated with observed biological effects in other field studies

Weakens

• Data showing that the stressor-response relationships at other sites are dissimilar to the stressor-response relationship found at the impaired sites
• Data showing that the candidate cause occurs in the case at levels too low to result in observed biological effects in other field studies

How do I score the evidence?

Helpful tips

• Both the direction and the magnitude of the stressor-response relationship factor into the scoring.
• Confidence is increased when this type of evidence is supported by Stressor-Response Relationships from Laboratory Studies.
• Note that, although this line of evidence uses site data in the form of Stressor-Response Relationships from the Field or Spatial/Temporal Co-Occurrence, this is not a case of double counting. This type of evidence addresses the consistency of the relationships with other field data.
• Note that it is not possible to refute the case for a candidate cause using evidence based on stressor-response relationships from other field studies.