Intervention Analysis of Power Plant Impact on Fish Populations

Abstract

Alewife (Alosa pseudoharengus) and yellow perch (Perca flavescens) abundances, estimated from monthly gillnet and trawl catches at two transects, were monitored before (1973–74) and during (1975–82) operation of the D. C. Cook Nuclear Power Plant, southeastern Lake Michigan. Intervention analysis, a technique which accounts for autocorrelated observations, and analysis of variance (ANOVA) were applied to the monitoring data to assess any plant impact beginning in 1975. Both analyses disclosed no significant power plant impacts except for gillnetted yellow perch adults. The ANOVA indicated a significant decrease in abundance at the plant-discharge transect relative to the reference transect as plant operation began, which established a plant effect; intervention analysis showed no change. When April and May catches (months of low abundance) were deleted, this plant effect was insignificant. Monte Carlo simulation showed that as the first-order autoregressive coefficient increased positively, type I error of the ANOVA F-test increased. However, ANOVA was more powerful than intervention analysis when a first-order autoregressive component was included. Impact assessment based only on ANOVA can result in detection of impact when actually there was no effect (type I error) when observations are serially correlated (lack independence).