Alexandre Souza wrote: > Dear friends that work on spatial Ecology, > > I am proceeding with the analysis of a dataset on the spatial > structure of canopy openness in the southern Brazilian mixed > conifer-hardwood forests, and would like to ask your opinion on a > rather simple matter on which I have doubts. > > I have six one-hectare plots subdivided in 100 10 x 10 m plots each. > In the centre of each subplot we took a hemispherical photograh and > estimated canopy openness. In Legendre and Fortin (1989) it is said > that before examining each significant value in a correlogram, we > must first perfom a global test, since several tests are done at the > same time, for a given overall significance level. The global test is > made by checking whether the correlogram contains at least one value > which is significant after a Bonferroni correction. > I had a similar problem during my PhD, and it became an early introduction to the problems of p-values.
I think the Bonferroni correction is a bad idea in this context, because you expect that the auto-correlation will decrease with distance. Therefore, if a distance of 16 is not significant, then 17+ will not be either. So, suppose you start by testing for all distances up to 10m, and find that distances 1m and 2m are significant. Then, you decide to test for all distances up to 50m. Now the Bonferroni correction will hammer the critical p-value, so you could very well find that nothing is significant: not because the data is different, but simply because you are doing more tests, and test that a priori are not so interesting. My solution was to do the test sequentially: test distance 1, then 2, then 3 etc. You find you don't need a correction. The reference is this: O'Hara R.B., Brown J.K.M., 1997. Spatial aggregation of pathotypes of barley powdery mildew. Plant Pathology, 46: 969-977. Nowadays, I would probably try and find a more model-based approach to estimating aggregation, and wouldn't be so interested in p-values. I don't think they should be taken too seriously: they are a guide to what's going on rather than being the truth. For the damaged plot, I think I would still run the analysis, but in the knowledge that the results have a lower power, so the p-values will be higher than for a full plot. Hope this helps! Bob -- Bob O'Hara Dept. of Mathematics and Statistics P.O. Box 68 (Gustaf Hällströmin katu 2b) FIN-00014 University of Helsinki Finland Telephone: +358-9-191 51479 Mobile: +358 50 599 0540 Fax: +358-9-191 51400 WWW: http://www.RNI.Helsinki.FI/~boh/ Journal of Negative Results - EEB: http://www.jnr-eeb.org
