THE STATISTICAL POWER OF 2 BUTTERFLY MONITORING SCHEMES TO DETECT TRENDS

1. Monitoring schemes for butterflies in the United Kingdom and the Ne therlands are aimed at the detection of long-term trends. It is useful to examine the power of these schemes to detect trends in a given per iod of time. 2. The approach was based on an A NOVA-model, and the tre nds were compared with the random population changes from one year to another. Several assumptions were made for simplicity's sake: autocorr elation in the data was ignored and only linear trends in log(10) (N 1) transformed data were examined. 3. The relevant variance component s to examine were the year-to-year variances and year-by-site variance s. These were estimated from the time series of the British Butterfly Monitoring Scheme, Year-to-year variances appeared to be higher in nor thern Britain than in other regions. In addition, variance components were related to the voltinism of species. 4. Power assessment was base d on the estimates of variance components and on the number of samplin g sites. In the British scheme, for 37 out of 51 species studied a dec rease of 50% or less is detectable with a power of 80% within a 20-yea r period. In the Dutch scheme such a decrease is detectable for 29 out of 47 species. 5. Because the schemes lack power for a number of spec ies, several strategies are discussed to enhance power. For species pr esent at less than 25 sites, it is most effective to increase the numb er of sampling sites where they are present, if that is possible in pr actice. But for species that are present at more than 50 sites, a furt her increase hardly improves the power. For these species, it is more efficient to adjust the data for weather conditions than to increase t he number of sites, 6. The assumptions we made hardly affect the resul ts for common species. But for rare species the results are more or le ss questionable, To get better estimates of the power, methods to asse ss power for monitoring schemes need to be developed that treat count data as discrete random variables.