Abstract from Research Report No. 260

D.E. Chamberlain, S.N. Freeman, G.M. Siriwardena & J.A. Vickery (2001)

The Effects of GMHT Crops on Summer Bird and Mammal Occurrence – A Power Analysis ISBN 978-1-906204-17-4

1. EXECUTIVE SUMMARY

1. The effects of genetically modified herbicide-tolerant (GMHT) crops on farmland biodiversity are being assessed by comparing the abundance of various taxa between conventional and GMHT halves of experimental fields in a large-scale experiment termed the Farm-Scale Evaluation (FSE).
2. A bird and mammal survey was carried out in spring and summer 2000 in order to assess the feasibility of including birds and mammals as target groups in the FSE. Specifically, this project was designed to assess the power of the FSE experimental design to detect differences in the utilisation of conventional and GMHT crops. This was achieved through a power analysis of various measures of bird and mammal occurrence for selected species, including measures of foraging activity in the crop.
3. There were 24 sugar beet sites, 12 maize sites and 12 rape sites surveyed five times between April and August, but many species were absent from several sites. Three different survey approaches were used: a territory mapping of all birds in the experimental and surrounding field boundaries; point counts of birds in fields (to record foraging events); and foraging observations of aerial feeders. Observers had no prior knowledge of how each half of a field had been treated.
4. The power analysis was based on Poisson and binomial models incorporating site, seasonal and treatment effects. Two different comparisons were made: GMHT crop before herbicide application vs. conventional crop in the same time period (i.e. early in the season) and GMHT crop after herbicide application vs. conventional crop in the same time period (i.e. late in the season).
5. Ten bird species were selected for analysis based on their abundance and detectability given the survey methods used and also to represent a range of dietary preferences. The species were Red-Legged Partridge Alectoris rufa, Skylark Alauda arvensis, Wren Troglodytes troglodytes, Dunnock Prunella modularis, Whitethroat Sylvia communis, Robin Erithacus rubecula, Blackbird Turdus merula, Song Thrush T. philomelos, Chaffinch Fringella coelebs and Yellowhammer Emberiza citrinella. For aerial feeders four species were considered: Swallow Hirundo rustica, Sand Martin Riparia riparia, House Martin Delichon urbica and Swift Apus apus.
6. Most analyses had low power to detect significant differences on a realistic scale in the data collected. The most powerful analyses involved a comparison of GMHT treated (i.e. sprayed) crop with conventionally treated crop later in the breeding season. Point counts collecting foraging data and information on bird registrations in open fields were most likely to yield analyses of high power in this data set.
7. Results for sugar beet were of relatively low power in comparison with results from maize and oilseed rape. This in part may have been because many sugar beet sites were in intensive arable areas with few, if any, trees or hedges. The geographical location of additional sites as well whether they are established as whole fields or sub-plots within fields (lacking boundaries) is therefore likely to have a big impact on future statistical power of the FSE for birds.
8. Simulated data for additional sites showed surveys of Red-Legged Partridge, Skylark, Dunnock, Whitethroat, Chaffinch and Yellowhammer in GMHT (treated) and conventional crops were predicted to achieve 95% power given the combined sample sizes for 2000 and those to be part of the FSE in 2001 - a total of 50 sugar beet, 39 maize and 37 rape sites. If it is assumed that the same number of sites is to be surveyed in 2002 as is planned for 2001 then the analyses predict that two further species, Robin and Blackbird, will also achieve 95% power.
9. In the case of mammals, only Hare Lepus europeaus and Rabbit Oryctolagus cuniculus were recorded in sufficient numbers for analysis in the mammal survey and these were of lower power relative to the bird analyses. As for birds, the most powerful analyses involved comparison of GMHT (treated) versus conventional crop treatments. Hare abundance on maize crops was the most powerful analyses and 95% power was predicted with a total of 46 sites.
10. The assumptions of all of these analyses were based on estimates derived from small sample sizes. An examination of the full set of results for all bird species combined predicted that to achieve a statistical comparison with 95% power, a difference of approximately 75% in numbers between treatments would be required for sample sizes of 40-100 sites for Poisson models. For binomial models, the odds ratio would need to be approximately three to four times greater on one treatment for 40-80 sites and 2.5 times greater for 80-100 sites to achieve 95% power.
11. In summary, the power analyses presented here indicate that a high probability of detecting significant differences in bird abundance or occurrence between treatments would be possible for six species with the additional sites proposed for 2001. Point counts recording foraging individuals on maize and rape crops in the later half of the breeding season, (i.e. after herbicides application to GMHT crops) are likely to yield the most powerful analyses. Detection of significant differences in mammal abundance was less likely.
12. There are a number of important caveats relating to these results. Crucially the findings are dependent on acceptance of the assumption that derived parameter estimates are precisely estimated and likely to remain similar when further sites are surveyed. Caution is therefore needed when interpreting these power analyses due to the small sample sizes involved. We have used a stringent level of power of 95%, but if a lower level of power were considered acceptable, the sample sizes and the magnitude of differences required to achieve this level would obviously be lower. For example, a further 26 species/crop/variable specific data sets could provide analyses of sufficient power if we considered a 75% level of power to be acceptable. However, the analysis presented in this paper provides a good baseline to direct future studies.
    

Back to Research Reports 248 onwards