The importance of influence diagnostics: examples from Snake River chinook salmon spawner-recruit models
The rapid decline of some salmonid populations in the Columbia River Basin led investigators to analyze spawner-recruit (SR) data in order to understand the potential gains of improving main-stem passage conditions and quantify the effectiveness of the juvenile transportation program. Direct measurements of passage survival and transportation were not always available, so instead, the researchers attempted to tease out the passage or transportation effects by using trends in production estimated from SR models. Small subsets of data, or even single observations, highly influenced the estimates of passage survival and transportation effectiveness derived from these models. For stream-type chinook salmon, deleting 1 of 13 stocks changed the estimate of main-stem passage survival from 11 to 34%. For ocean-type chinook salmon, the conclusion that transportation should be immediately halted hinged on a single observation. The Snake River salmon SR models starkly illustrate the importance of using influence diagnostics to temper inferences.