Mh. Graham et Ms. Edwards, Statistical significance versus fit: estimating the importance of individual factors in ecological analysis of variance, OIKOS, 93(3), 2001, pp. 505-513
Although analysis of variance (ANOVA) is widely used by ecologists, the ful
l potential of ANOVA as a descriptive tool has not been realized in most ec
ological studies. As questions addressed by ecologists become more complex,
and experimental and sampling designs more complicated, it is necessary fo
r ecologists to estimate both statistical significance and fit when compari
ng the relative importance of individual factors in an explanatory model, e
specially when models are multi-factorial. Yet,,vith few exceptions, ecolog
ists are only presenting significance values with ANOVA results. Here we re
view methods for estimating statistical fit (magnitude of effect) for indiv
idual ANOVA factors based on variance components and provide examples of th
eir application to field data. Furthermore, we detail the potential occurre
nce of negative variance components when determining magnitude of effects i
n ANOVA and describe simple remediation procedures. The techniques we advoc
ate are efficient and will greatly enhance analyses of a wide variety of AN
OVA models used in ecological studies, Estimation of magnitude of effects w
ill particularly benefit the analysis of complex multi-factorial ANOVAs whe
re emphasis is on interpreting the relative importance of many individual f
actors.