Estimating statistical power to evaluate ongoing waterfowl population monitoring

The probability of failing to detect a trend when 1 exists has been considered only rarely in the interpretation of monitoring studies. Retrospective power analysis accomplishes this assessment. me apply retrospective power analysis to evaluate both population trends and survey design in the waterfowl surveys conducted by the Canadian Wildlife Service and others around Riske Creek, British Columbia. Eleven of 18 species showed long-term (17 yr) and short-term (10 yr) trends. For the remaining 7 species, the long-term analysis had sufficient power (0.8) to have detected at least a 5% annual change, had 1 existed, which supported the conclusion that little change occurred. However, statistical power and detectable effects varied considerably among species, with a range of 3-14 years of data needed to be able to detect a 5% annual trend. When we used the shorter-term dataset, power was reduced below acceptable levels for 4 of the 7 species failing to show a trend. It would be a mistake to conclude that the numbers of these 4 species were not changing. Statistical power was highest for the species for which the surveys were originally designed, Barrow's goldeneye (Bucephala islandica) and mallards (Anas platyrhynchos), which had narrow confidence intervals and relatively small minimum detectable trends. In contrast, blue-winged teal (Anas discors), gadwall (Anas strepera), green-winged teal (Anas crecca), northern pintail (Anas acute), and northern shoveler (Anas clypeata) had relatively large minimum detectable trends and nide confidence intervals. Much of the power of these surveys was due to repeated surveying within seasons. For most species, power increased substantially by including up to 4 surveys as replicate observations within a year, but power increased little when data from a fifth or sixth survey were included.