by Kim Iles

There are those who move plot locations.  In the US it is mostly because of ignorance, but it also happens in other parts of the world where it is to the purchasers advantage to get the wrong answer (because they pay less with an incorrect inventory).  In some professions, ecology in particular, you sometimes are required to choose a “typical” plot to measure, even though this biases the result.  In very small sample sizes, this might be an advantage, but that is seldom the case. 

When plots are used to “identify” a stand in some way, for instance, you might find that you have fallen on the 10% of the area which is a Cedar patch inside a mainly Hemlock stand.  What do you do?  If you want the stand label to show “mainly Hemlock” and need some data for a rough description, you might move the plot into the Hemlock and take the plot there.  Better for that stand you might correctly think, but in the long run your data indicates 100% Hemlock in “mainly Hemlock” stands, which is not true.  The right long run answer is “mainly Hemlock, with about 10% Cedar”.  Moving plots or changing strata when you do the field work is a dicey business.  In general it is a bad idea, sometimes it is fraud.  Like golf, the best practice is to “play it where it lies”. 

In some cases, people feel that plots should be moved further into the stand rather than be taken near the “untypical” edge.  They might have wondered, since these areas are so untypical “why they keep falling into them” – but bad ideas are hard to overcome by mere logic.  If 20% of the area of the stand is in “the edge”, then 20% of the plots ought to be there too – especially when it is different than the main part of the stand – again, unless the only purpose is to give the stand a “mainly X” label, and any data does not need to be valid when computing an average. 

Some things show up when you check the original plots.  The data might have been changed after the cruiser made their measurements, but that should show up with an examination of the original plot data and a simple check of the original plot.  Which trees were dropped as being “of no interest” should also show up on a check cruise.  “Hidden defect” deductions and other disagreements that affect volume will also be spotted.  Check cruises should examine the compiled Gross and Net volume data differences as well as the dollar value difference of the original plot data. 

What do you do, however, when you suspect that the plot has been moved in order to bias the result, and the plot data was perfectly correct when you checked the cruisers (relocated) plot in the field? 

On rare occasions, fraud or serious procedural errors need to be detected, and perhaps corrected.  It can be done, but it should be done carefully.  Suppose you want to prove that plots have been moved intentionally.  The British Columbia forestry community has evolved such a process [1].  It is like a “resample” of the area, but rather than cover the whole area with a new set of plots, the original plots (or a valid actual sample of them) are surrounded by 4 comparison plots at some fixed distance from the original plot.  The plots should be far enough apart not to have most of the bias that moving the original plot created.  20 feet or so should be enough in most forest types. 

In terms of  “the best distance between comparison plots”, there are ways to discover when a plot starts to be independent of a nearby plot.  This is done by calculating the correlation between the original and matching plot volumes as they are moved apart a little at a time.  This is obviously a research problem, perhaps using actual stem maps of typical stands.  The results are graphed as distance increases.  When the correlation is essentially zero, that would indicate the best distance to use for comparison plots.  Lacking such research, you can use your own opinion about how far the comparison plots must be from the originals in your own forest type.  If you do not get them far enough apart, it will only be less sensitive to whatever bias that moving the original plot produced.  Getting them too far apart would not be a statistical problem, just a field hassle. 

The bias is most likely to be found in differences by species in the comparison plots, or the total value of the plot.  It is easier to bias the tree count in the smaller trees, and this is especially true when the average tree count is low – so look at the results on smaller vs. larger trees.  These comparisons can be done using a statistical “paired t-test” between the original plots and the average of the 4 comparison plots.  If the results are quite close, of course, then either the plots have not been moved or they are not been moved enough to affect the results, and no statistical test is required.