Latest Contributions by EricKrause

You definitely want to take a subset of your data.  The Trend Analysis tool is not really designed to be used with so many points; it will just look like a huge blob, and it will be very difficult to actually see any trend.  Also keep in mind that when you rotate the graph, it's trying to dynamically move hundreds of thousands of points; you will definitely get lags and delays.  In geostatistics, several hundred points is more than enough for building the interpolation model.  Giving so many points is unnecessary when looking at any of the ESDA tools. ... View more

I just sent the workflow. ... View more

Ted, areal interpolation is fundamentally a kriging method.  We assume that the starting polygons are the result of averaging some underlying Gaussian process, and we model this underlying process using the data in the polygons (this is often called the "change of support" problem).  This modeling creates a smooth prediction surface for the variable of interest.  This surface is then averaged (or integrated) in the new polygons to get the new predictions. The big difference between this approach and proportional intersection methods is that the latter assumes the density of the variable is constant across each polygon, which isn't a realistic assumption.  Our approach also provides standard errors for the predictions. You can read a short paper about our method here.  We're also working on making a longer version available.  We'll also have a workflow topic up for Beta 2.  I can send you a copy early if you want to try to see it now. ... View more

http://help.arcgis.com/en/arcgisdesktop/10.0/help/index.html#/Median_Center/005p00000019000000/ The tool is called "Median Center," and it's in the Spatial Statistics toolbox under the "Measuring Geographic Distributions" toolset. You must have just missed it.  Happens to the best of us. ... View more

Hi Lily, Yes, the variance is taken into account in the back transformation.  If you'd like to learn more about exactly what is happening behind the scenes, our method is described in the following paper: Cressie, N. 2006.  "Block Kriging for Lognormal Spatial Processes." Mathematical Geology 38: 413-43 Good luck, and thanks for the interest in geostatistics. -Eric ... View more

We support cokriging with raster and point input, so I'm not sure why the method is failing to initialize.  Most likely, it has to do with your specific data. If you can send the data to ekrause@esri.com, I'll try to figure out why the method isn't working for your data. ... View more

This is a very strange looking crosscovariance cloud.  It actually looks like the correlation increases up to about 180 meters, then it decreases from there.  This means that points that are 180 meters apart are more highly correlated than points that are 10 meters apart.  This is problematic because it appears to violate the idea that points that are closer are more similar. It may be possible to use cokriging, then turn the "Shift" option to True.  This will attempt to correct for asymmetric cross-covariance (which you appear to have here).  If the Shift correction doesn't help, this just may not be a good dataset for interpolation. I'm also concerned about the Crosscovariance Surface on the bottom-left of the graphic.  Maybe you just need to refresh your screen, but you should be able to see a surface there. If you're able to send the data to ekrause@esri.com, I'll take a look at it and try to give a more specific recommendation. ... View more

I should have noticed this when I first looked at the pdf, but the problem is that the areas of no prediction are surrounded by points that lie on the same line.  For second-order trend removal, you need to have at least three distinct x and three distinct y coordinates in the neighborhood around each prediction location (among other restrictions).  So, if all the neighbors fall on the same two lines, there is instability in the predictions, and the prediction can�??t calculate.  As Gail discovered, if you increase the maximum neighbors, you�??ll force more points in that don�??t fall on the same two lines. ... View more

Is it possible to send the data to ekrause@esri.com?  I'll need the point feature class, and I'll need the xml file for the kriging layer.  You can get the xml by right-clicking the layer in ArcMap and choosing Method Properties.  This will open the Wizard, then click Finish.  The Method Summary window pops up, and clicking Save... will allow you to save the xml file. ... View more

When you click Finish in the Wizard, the Method Report window pops up, and you have the option to save an xml file that contains all the parameters. Use the Create Geostatistical Layer tool with the xml as the model source.  Then specify the new dataset (the one that includes the old locations and the new locations with mock values), and the tool will interpolate on the new dataset with the old model parameters. Once you have the GA layer, use the GA Layer to Points tool to predict back to the original data (without the new locations added).  Don't worry about the field to validate on.  Open the attribute table on the point feature class it creates, and the last column will be Standard Error.  You can calculate the mean of the Standard Errors by right-clicking the column and choosing "Statistics."  You can also use the Summary Statistics tool. To compare against the old standard errors, do the same process with the GA layer created only using the old locations. ... View more

Hi Steve, Could you please a brief explanation about "optimize model "  in Arcgis 10 geostatistical analysis.  Is it sufficient to choose this option to provide best map to the user. In kriging, the optimize button uses weighted least-squares.  The algorithm iterates through the range and fits the nugget and sill using weight least-squares for each iteration.  It then calculates the combination with the minimum root-mean-square.  Because the covariance matrix between the three parameters is not estimated, this process will not necessarily find the global root-mean-square minimum.  If you play with the parameters enough, you may find a better root-mean-square. Looking at your cross-validation statistics, you can justify either model.  Your model has a better root-mean-square, but it has a worse mean prediction error and rmse-standardized. ... View more

If the other transformations aren't working, try a Normal Score transformation.  It's the default option for simple kriging in the Geostatistical Wizard.  http://help.arcgis.com/en/arcgisdesktop/10.0/help/index.html#//00310000000v000000.htm http://help.arcgis.com/en/arcgisdesktop/10.0/help/index.html#/Using_normal_score_transformations/00310000000w000000/ If you're able to send your data to ekrause@esri.com, I can take a look to see if kriging is appropriate here. ... View more

All interpolation methods assume there is spatial autocorrelation.  Some methods explicitly model it (kriging, for example), and others just make assumptions about it (IDW, for example). With only 27 points (and a possible outlier), I wouldn't try to use anything other than IDW.  If you're concerned about the logarithmic scale, you can make a new field and calculate the antilog, then take the logarithm of the results. I wouldn't attempt to do kriging with so few points and an outlier. ... View more

I would suggest you look into Geographically Weighted Regression, or maybe even Ordinary Least-Squares.  You can use geostatistics to interpolate how prices, but if you have extra information (like square feet), you can incorporate it with GWR or OLS. Both tools are in the Spatial Statistics toolbox.  After reading the documentation, if you still have questions about to use the tools, the Spatial Statistics forum is the best place to ask: http://forums.arcgis.com/forums/110-Spatial-Statistics ... View more

The scale of the x-axis depends on the Lag Size, Number of Lags, and the size of the window. You can always export the points in the semivariogram as a dbase file and open it in Excel, then you can scale it however you want.  For version 10, as shown in the attached picture, you can export the binned values (red dots), the averaged values (blue crosses), and the model (coordinates of the blue semivariogram curve).  For 9.3.1, you can right-click the semivariogram and save the values (red dots) and the model (blue semivariogram curve). If you absolutely need the screenshot from the Geostatistical Wizard, you can manipulate it to give round numbers, but keep in mind that this manipulation will change the model, so be careful.  To do this for ArcGIS 10, set the Lag Size to a round number (like 10000), and set the Number of Lags to 11.  Then play with the window size until the numbers are round.  For 9.3.1, use a round number for the Lag Size, and set Number of Lags to either 8 or 16, and the scale should be round. ... View more