1.2.2 Methods and confidence
Where long data series exist, the detection of trends or changes in system properties that are beyond natural variability has most commonly been made with regression, correlation and time-series analyses. When data exist from two (or more) discontinuous time periods, two-sample tests have frequently been employed. Testing is also done for abrupt changes and discontinuities in a data series. Regression and correlation methods are frequently used in the detection of a relationship of the observed trend with climate variables. Methods also involve studies of process-level understanding of the observed change in relation to a given regional climate change, and the examination of alternative explanations of the observed change, such as land-use change. The analysis sometimes involves comparisons of observations to climate-driven model simulations.
In many biological field studies, species within an area are not fully surveyed, nor is species selection typically based on systematic or random sampling. The selection of species is typically based on a determination of which species might provide information (e.g., on change with warming) in order to answer a particular question. The study areas, however, are often chosen at random from a particular suite of locations defined by the presence of the species being studied. This type of species selection does not provide a well-balanced means for analysing species showing no change. Exceptions are studies that rely on network data, meaning that species information is collected continuously on a large number of species over decades from the same areas; for example, change in spring green-up of a number of plants recorded in phenological botanical gardens across a continent (Menzel and Fabian, 1999). Analysis of change and no-change within network data provides a check on the accuracy of the use of the indicator for global warming and the ability to check for ‘false positives’, i.e., changes observed where no significant temperature change is measured. The latter can help to elucidate the role of non-climate drivers in the observed changes.
The analysis of evidence of no change is also related to the question of publication or assessment bias. Studies are more likely to be successfully submitted and published when a significant change is found and less likely to be successful when no changes are found, with the result that the ‘no change’ cases are underrepresented in the published literature. However, in contrast to single-species in single-location studies, multiple species in a single location and single or multiple species in larger-scale studies are less likely to focus only on species showing change. The latter studies often include sub-regions with no-change; for example, no change in the number of frost days in the south-eastern USA (Feng and Hu, 2004), little or no change in spring onset in continental eastern Europe (Ahas et al., 2002; Schleip et al., 2006), or sub-groups of species with no change (Butler, 2003; Strode, 2003).
An accurate percentage of sites exhibiting ‘no change’ can be assessed reliably by large-scale network studies (see, e.g., Section 1.4.1; Menzel et al., 2006b) for the locations defined by the network. For investigations of a suite of processes or species at numerous locations, the reported ratio of how many species are changing over the total number of species rests on the assumptions that all species in the defined area have been examined and that species showing no change do not have a higher likelihood of being overlooked. Both multi-species network data and studies on groups of species may be used to investigate the resilience of systems and possible time-lag effects. These are important processes in the analysis of evidence of no change.