The conclusion made here is not to say that the founding of this particular threshold applies to all situations.  However the founding from this project suggests that when making inference to real world phenomena using statistical measures, the researches really need to take into consideration to the accuracy of making such inference, base on the fact that different conclusion can be made with different resolution of data if raster grid is used.  In fact, Odland (1988) has introduced in his book that resolution of data will affect the measure of spatial autocorrelation, although he refers this to vector polygon coverage and that he does not go further into this issue.  Therefore, choosing the right cell size for the raster grid is very important, however this is not always easy to do.  Of course the best model will be the one that has the smallest possible cell size, but this is not always preferable since this will create huge dataset if a large area if chosen.  This is actually illustrated by the grid coverage having cell size of 10m resulting 62000 cells and the S-Plus software does not take it.  A balance between accuracy and the huge dataset problem is always needed for the researchers to decide before any analysis and inference are carried out.
 

GO TO Reference -->

<-- Back to Index