CAJ | 학술논문

一个"好"的试验,统计检验的显著差异可证明处理效应存在;而一个"坏"的试验,统计检验的显著差异本身并不能证明处理效应的真实存在.试验单位的不独立就可能使干扰因素偏倚地影响试验结果,是多种形式的伪重复的根本原因.如果不统筹地考虑试验方法和数据分析方法,环境要素的时空格局就可能被错误地当成处理效应.以案例分析的形式探讨了中国生态学试验设计与解释中常见的3个问题:(1)简单伪重复问题,这是Hurlbert[2]早已指出的伪重复情形中的一种,通常是把取样的重复当成了处理的重复;(2)把反复测量结果当成重复的问题,即对同一个对象的反应变量前后进行多次观测,却把这些观测值视为重复而进行统计检验所造成的问题;(3)混淆时空效应与处理效应的问题,由于取样方法(破坏性取样)或者研究对象(例如流动的水体)性质的特殊性等原因,数据中所体现出来的格局有可能由于时空效应而并非处理效应所造成.在这3种情况下,数据的产生方式与所用统计方法的前提相违背.
일개"호"적시험,통계검험적현저차이가증명처리효응존재;이일개"배"적시험,통계검험적현저차이본신병불능증명처리효응적진실존재.시험단위적불독립취가능사간우인소편의지영향시험결과,시다충형식적위중복적근본원인.여과불통주지고필시험방법화수거분석방법,배경요소적시공격국취가능피착오지당성처리효응.이안례분석적형식탐토료중국생태학시험설계여해석중상견적3개문제:(1)간단위중복문제,저시Hurlbert[2]조이지출적위중복정형중적일충,통상시파취양적중복당성료처리적중복;(2)파반복측량결과당성중복적문제,즉대동일개대상적반응변량전후진행다차관측,각파저사관측치시위중복이진행통계검험소조성적문제;(3)혼효시공효응여처리효응적문제,유우취양방법(파배성취양)혹자연구대상(례여류동적수체)성질적특수성등원인,수거중소체현출래적격국유가능유우시공효응이병비처리효응소조성.재저3충정황하,수거적산생방식여소용통계방법적전제상위배.
The logic of experiments is discussed, and three common mistakes in designing and interpreting experiments among ecologists are analyzed. Researchers sometimes overlook logic problems, while paying too much attention to statistical techniques. A statistically significant difference can demonstrate treatment effect for a well-designed experiment, but it might not for a "bad" experiment. Any extraneous variables that change systematically along with the treatment threaten the internal validity of an experiment. The bias in some cases arises from interdependence between experimental units, which is the case with pseudo-replication, a term coined by S. Hurlbert in 1984. In this paper we review experiments published in one issue each of two of the most influential ecology journals, Journal of Plant Ecology and Acta Ecologica Sinica. Some cases are selected and analyzed to exemplify three mistakes which are commonly made by Chinese ecologists when designing an experiment or interpreting data. They are: (1) simple pseudo-replication, (2) confusing repeated measurements as replicates or temporal pseudo-replication in terms of Hurlbert, and (3) wrongly treating a spatial or temporal pattern as a treatment effect. Among the experiments reviewed, 17.9%-42.9% in Journal of Plant Ecology and 14.3%-42.9% in Acta Ecologica Sinica involved at least one type of the errors mentioned above.