Time-series intervention analysis: Unreplicated large-scale experiments
Some important ecological questions, especially those operating on large or unique scales, defy replication and randomization (Schindler, 1987; Frost et al., 1988; Carpenter, 1990). For example, suppose one is interested in the consequences of lake acidification on rotifer populations. Experimental acidification of a single small lake is a major undertaking, so it may be possible to manipulate only one lake. Even with baseline data from before the application of the acid, such a study would have no replication, and thus could not be analyzed with classical statistical methods such as analysis of variance (Chapter 3). An alternative approach might be to use some biological!physical model of the system of interest that allows for replication. For example, one could construct small replicated enclosures in lakes, and acidify some of them. While this would (with proper execution) permit valid statistical analysis of differences among units the size of the enclosures, it is questionable whether this model allows valid ecological generalization to the lake ecosystem.· For large-scale phenomena, experiments on small-scale models may not be a trustworthy ecological substitute for large-scale studies (although they may provide valuable supplementary information). In this chapter, we examine how certain types of unreplicated studies can be analyzed with techniques developed for time series data. Time series are repeated measurements, or subsamples, taken on the same experimental unit through time. Time series analysis techniques include methods for determining whether nonrandom changes in the mean level of a series have occurred at prespecified times. The results of such analyses can help determine whether-other changes or manipulations occurring at those prespecified times may have caused changes in the observed series.
Chapman and Hall, Inc.