Hi,

Iâ??m carrying out some analysis on the variability of two continuous variables which Iâ??d expect to be correlated. Iâ??d therefore expect that in areas where one of my variables varies considerably the other one would follow the same pattern.

My variables are also not expressed in the same unit of measurement. I thought of using the Slope function in Spatial Analyst to compute the rate of change in Z-values of my two variables within each cell of my raster dataset (note: both are non-elevation related variables, which makes me questioning my approach). This would give me a first idea of the cell-by-cell variability of each of my variable. Iâ??d then compute a zonal statistics analysis (using a fishnet to define my zone areas), deriving the standard deviation from the slope values of my variables. Iâ??d expect that areas (in my fishnet) with high values of the standard deviation for one variable would correspond to area with high value of the standard deviation for the other one. Any departure from this expected pattern would indicate the presence of an anomaly which I may want to investigate further.

Any feedback on the approach just described would be very much appreciated.

P

Iâ??m carrying out some analysis on the variability of two continuous variables which Iâ??d expect to be correlated. Iâ??d therefore expect that in areas where one of my variables varies considerably the other one would follow the same pattern.

My variables are also not expressed in the same unit of measurement. I thought of using the Slope function in Spatial Analyst to compute the rate of change in Z-values of my two variables within each cell of my raster dataset (note: both are non-elevation related variables, which makes me questioning my approach). This would give me a first idea of the cell-by-cell variability of each of my variable. Iâ??d then compute a zonal statistics analysis (using a fishnet to define my zone areas), deriving the standard deviation from the slope values of my variables. Iâ??d expect that areas (in my fishnet) with high values of the standard deviation for one variable would correspond to area with high value of the standard deviation for the other one. Any departure from this expected pattern would indicate the presence of an anomaly which I may want to investigate further.

Any feedback on the approach just described would be very much appreciated.

P

Thanks for this you gave me some elements to clarify my thinking. In the end I reverted to carry out an analysis of the correlation between two variables in the raster by computing the correlation coefficient. I followed what suggested in one of the older thread (http://forums.esri.com/thread.asp?c=93&f=1740&t=288054).

I have a question about the workflow in there, and I hope very much that you could clarify it. In my case the two grids have no NoData values, so I only need to compute the Mask in order to do a cell count (which I believe is what the �??n�?� parameter is in the formula used to derive r).

Well, my correlation coefficient raster provides weird values indeed. I�??d have expected the values to vary in the -1/+1 range. I shouldn't need to normalise my grids prior to the calculation of the correlation coefficient, as dividing the covariance (numerator of the equation) by the product of the standard deviations of variable X and Y would already ensure that r values would range as expected. So the problem may be somewhere else. Any suggestion?

Thanks in advance

Paola