Carroll Steven S, Pearson David L
Department of Biology, Arizona State University, Tempe, AZ 85287-1501, U.S.A.
Conserv Biol. 2000 Dec 18;14(6):1893-1897. doi: 10.1111/j.1523-1739.2000.99432.x.
Due to the structuring forces and large-scale physical processes that shape our biosphere, we often find that environmental and ecological data are either spatially or temporally-or both spatially and temporally-dependent. When these data are analyzed, statistical techniques and models are frequently applied that were developed for independent data. We describe some of the detrimental consequences, such as inefficient parameter estimators, biased hypothesis test results, and inaccurate predictions, of ignoring spatial and temporal data dependencies, and we cite an example of adverse statistical results occurring when spatial dependencies were disregarded. We also discuss and recommend available techniques used to detect and model spatial and temporal dependence, including variograms, covariograms, autocorrelation and partial autocorrelation plots, geostatistical techniques, Gaussian autoregressive models, K functions, and ARIMA models, in environmental and ecological research to avoid the aforementioned difficulties.
由于塑造我们生物圈的结构化力量和大规模物理过程，我们经常发现环境和生态数据在空间上或时间上——或者在空间和时间上——是相关的。在分析这些数据时，经常应用为独立数据开发的统计技术和模型。我们描述了忽略空间和时间数据相关性的一些有害后果，如低效的参数估计、有偏差的假设检验结果和不准确的预测，并列举了一个因忽视空间相关性而出现不良统计结果的例子。我们还讨论并推荐了在环境和生态研究中用于检测和建模空间和时间相关性的可用技术，包括变差函数、协变差函数、自相关和偏自相关图、地质统计技术、高斯自回归模型、K函数和自回归积分移动平均模型，以避免上述困难。
