Data become evidence when associations are made between types of data. General issues with analyzing associations are discussed here. More specific information is included under each type of evidence.

• Matching Data in Time and Space

  Associating measurements of stressor and effects are easiest when they are located together (co-located) in time and space. It becomes more difficult as stressors are dispersed over larger scales, occur intermittently, or cannot be measured. The evaluation of spatial associations must consider whether potentially affected organisms may have moved since exposure. In particular, it is helpful to consider the mobility of organisms relative to the extent of the observed exposed and unexposed reaches or areas. Clearly, fish are capable of swimming long distances and invertebrates may drift downstream or fly upstream. However, extensive experience with bioassessment of fish and invertebrate communities has demonstrated that the movements of these organisms are usually not so great as to prevent the observation of spatial associations. The movement of a few individual organisms from contaminated reaches to upstream reaches will diminish but generally not conceal the contrast or gradient among reaches. However, salmon and other species that regularly move long distances require special consideration when analyzing spatial associations. In such cases, consider the logic of the situation and possibly use GIS as a platform for modeling spatial relationships. The potential for time lags between exposure and effects should also be considered. For example, if a stressor, such as a diversion of water flow, prevents salmon from reaching the sea on their out-migration, the effect (i.e., destruction of the salmon run) may not be observed until 3 years later.

• Using Surrogate Data

  When measurements of the proximate stressor are not available, surrogates can be sought. Information on the location and attributes of possible sources can be useful surrogates. This information is particularly important for stressors that are intermittent in nature (e.g., high flow events), or degrade quickly (e.g., some pesticides). As sources become larger in scale and more diffuse, information on the sources becomes more difficult to use in site-specific causal evaluation. In addition, information on sources that produce many proximate stressors cannot be used to distinguish among those stressors. For example, increased impervious surfaces have been linked to proximate stressors such as increased flow extremes, temperature spikes, increased toxic substances, and decreased dissolved oxygen (Schueler et al. 1995).

• Quantifying Associations

  Whenever possible, associations should be quantified. For categorical data, construct a contingency table and calculate the frequencies of associations. For count or continuous data, use linear or nonlinear models. For example, the abundance of Ephemeroptera at a site may be regressed against the concentration of total sediment polycyclic aromatic hydrocarbons (PAHs). Select the analysis technique that best illuminates the association, based on the amounts and types of data available. If effects data are categorical or heterogeneous and exposure data are continuous, categorical regression may be used (Dourson et al. 1997). Some statistical descriptions of associations include correlation coefficients, confidence intervals, and p-values. However, avoid statistical hypothesis testing of the associations (tips on using statistics and statistical hypothesis testing for analyzing observational data in stressor identification are provided in a separate text box).

• Improving Associations by Identifying Confounding Factors

  Often associations between candidate causes and effects can be improved by identifying and isolating confounding factors in either the receptors or the environment. For example, a decline in fish species richness is a common measure of impairment, but the number of species present generally increases with increasing stream size (e.g., OEPA 1988a). Therefore, including a correction for stream size could strengthen the association between the degradation and species loss. Similarly, the frequency of hepatic neoplasms in fish is associated with both the age structure of the fish population and the concentration of PAHs in sediment (Baumann et al. 1996). Correction for age of fish increases the consistency and, potentially, the biological gradient in the relationship between hepatic neoplasm frequency and industrial contaminants.