Large observational data sets can provide valuable information regarding causal relationships between changes in environmental factors and biological responses. However, controlling for background, confounding factors in these data can be difficult, which reduces our confidence when attempting to infer causal relationships. Propensity scores, a technique developed by epidemiologists, summarize the effects of all known confounding factors as a single index. This single index allows one to ascertain whether treatment and control groups are drawn from similar background characteristics (e.g., stream types), and allows one to more confidently estimate the size of the treatment effect. Propensity scores provide a more robust approach for controlling for confounding factors than commonly-used multiple regression techniques, and when used together with multiple regression, can provide for more reliable causal inferences. We apply propensity scores to estimate the effects of increased nutrient concentrations and increased atrazine concentrations on macroinvertebrate assemblage composition using two national databases.