On Tue, 18 May 2004 09:13:44 +0200, "Rafal" <[EMAIL PROTECTED]> wrote:
[snip, cited comment] > > Plots are in forested area. On each plot 6 trees had its level of > defoliation assesed. The 'treatments' are different soil conditions and > tree's age. The number of trees was selected for economical reasons. I > suspect that more trees on plot e.g. 20 could make smaller differences in > averagaes between plots. This is where the idea of creating clusters of 3 or > 4 plots and treating it as one one plot with 18 - 24 trees comes from. > > > If the two plots had the same treatment and yet differ that > > way, then it has to indicate that the locations make a > > difference, and the interaction has to serve as the error > > term for testing. > > I agree. My assumption is (it comes from analogy to other forest research) > that when you have more trees on plot the differences between trees on one > plot caused by the factors which we cannot maeasure are reduced and make > more visible differences caused by more general natural conditions on plots > which we can measure. Ouch! What? Do you say, non-random selection of the 6 trees? Or are you merely commenting?: that one acre of uniform forest will have a homogeneous set of trees, by type and de-foliage; whereas a mixed environment will have mixed choices. I hope that there were careful instructions about how to select the trees to be part of the sample, and that relevant other factors were recorded. > > > From the little you have implied about the purpose, > > I don't see any justification for averaging or smoothing. > > I hope that above I wrote more about my problem. The purpose is that I would > like to use the point data for interpolation e.g. kriging > I have not done kriging. From the little I have read, I think I would want to try it with the raw numbers, first. I suppose that a sort of blurring *might* help - if your 'plots' are small compared to the features that are to be delineated. Filtering does not have to be as heavy as you describe, though. I mean, you can weight the original plot 1.0, and less for the ones around it. On the other hand, I thought that kriging used algorithms rather like that. I don't know if I added anything useful this time or not. -- Rich Ulrich, [EMAIL PROTECTED] http://www.pitt.edu/~wpilib/index.html . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
