General

What is the equation?

Q. What is the "formula" for a given model and how can I determine this using the coefficients in the model list?

A. These nonparametric models do not have coefficients in an equation in the usual sense.  The "tolerances" (smoothing parameters) are used instead of the regular coefficients.  So a prediction is made not with an equation but by obtaining an estimate from the calibration data set, using a given set of tolerances.

See also "Model Specification" in the companion PDF document.  If you are interested in using the models in a predictive mode (predicting responses for new sample units).  You might also find helpful the section in the help system under Prediction | New Sites.

Repeated measures data

Q. Is it appropriate to use HyperNiche to analyze data sets that contain numerous samples obtained at different times from the same limited number of sites?  Each sample will have different values of most predictor and all response variables.   Time/date is used as a predictor variable.

A. This fine for descriptive statistics, but be careful about formal hypothesis tests.   If the sample units are not considered independent, you may bias a hypothesis test -- exactly the same problem as with parametric statistics.

Statistical significance

Q. How can I evaluate the statistical significance for a given model? For example, how can I interpret xR2 values for models with large and small numbers of data?  How can I determine the relative significance of different predictors?

A. You can evaluate whether the model is stronger than expected by chance by using the randomization test.  After selecting a particular model, be sure that you have the same file of predictors open, then run Fit Model | Evaluate Selected Model.   Then select Monte Carlo test on the output options tab.  This is very slow with large data sets.

In late 2004 we added a sensitivity analysis.  Use this to evaluate the relative importance of different predictors in a particular model.  This is automatically given now with Fit Model | Evaluate Selected Model.

Editing spreadsheets

Q. Why doesn't HyperNiche allow cell-by-cell editing of spreadsheets?  I have to go back and forth between HyperNiche and a spreadsheet program when I am cleaning my data.

A. HyperNiche doesn't attempt to duplicate the spreadsheet functions.  However the HyperNiche allows simple editing of the main and second matrices.  More complicated editing and sorting is best done in your spreadsheet software.  It is so easy to go back and forth between applications using Windows that there should not be a great inconvenience.  You should note, however, that some spreadsheet programs, such as Excel, keep the file open, unless explicitly closed.  This means that while a file is open in Excel it is not available to HyperNiche.

Convert output to tables

Q. How can I convert tables in HyperNiche’s output files into tables in my word processor?

A. Your word processor probably has functions for converting tabular text files into tables.  For example, using Microsoft Word, you can highlight a table in a HyperNiche output file, then select Table | Convert Text To Table, to convert the output into a standard tabular format.  Be sure to set the delimiter to a space character and adjust the desired number of columns.

Mac version?

Q. Are versions of HyperNiche available for the Macintosh?

A. No. There is only a Windows versions of PC-ORD.  However, PC-ORD will run on Virtual PC for Mac. It will also run on the Windows side of a partition on a Mac.

Mac emulation 3D crash

Q. I am using HyperNiche with Mac OS X running Windows XP via Parallels. Graphing 3D makes HyperNiche crash. Is there a solution?

A. Open XP Control Panels | Appearance and Themes | Display Properties | Settings | Advanced | Troubleshoot and set the Hardware acceleration slider to the second position from the left that says "Disable all but basic accelerations. Use this setting to correct more severe problems". Note this position also disables DirectDraw and Direct3D used by any other software.

HyperNiche CD will not install

Q. HyperNiche CD will not install on my system.  What is the solution?

A. Be sure to disable any antivirus program such as Norton before trying to install. If the CD still won't install, you may have the CD autorun feature disabled.  You can install manually from the Start Menu by selecting Run and entering x:Setup where x is the letter of your CD drive.  You also can click the Browse button and navigate to the Setup program.

OPENGL32.DLL not found

Q. I'm running Windows 95 operating system and when I start HyperNiche I get an error message saying, "A required .DLL file, OPENGL32.DLL, was not found.  What is the solution?

A. For Windows 95 you may need to download Opengl.exe (484K) to the Windows System folder or the HyperNiche folder and then execute the self-extracting program that includes OPENGL32.DLL.

AIC and BIC for model selection

Q. I would like to use the AIC (Akaike Information Criterion) and the BIC (Bayesian Information Criterion) for model selection.  Is this possible with NPMR in HyperNiche?

A. The AIC and BIC depend on the number of parameters in a model.  Because NPMR models do not have explicit parameters as such, these are not applicable to NPMR models.  Instead, use the controls on overfitting provided in HyperNiche (minimum average neighborhood size, minimum data:predictor ratio, and the improvement criterion).

File not found error

Q. When attempting to run NMS using starting configuration I provide in a "graphrow" file PC-ORD is unable to open or read the file.  The error message I get is "Error reading graph file. Graph file not found."  What is wrong?

A. Any analysis related file has to be stored in a folder that has a filename (including the path) less then 240 characters long.  Open Windows Explorer and navigate to the folder where the file is stored to get the complete path in the Explorer Address window.  If the path plus file name is longer than this limit move the files not found to a shorter path and/or use a shorter file name.

Won't work with Vista or Win 7

Q. I just installed HyperNiche and nothing seems to work.

A. The HyperNiche folder might not have write access.

1. Log in as Administrator
2. Right click the  HyperNiche icon or menu item that you run  HyperNiche
3. Select Properties
4. Select the Security tab
5. In the Permissions for SYSTEM box, check Allow for everything except Special Permissions
6. Click Apply and OK

Can't find saved files in Vista or Win 7

Q. Whether saving a matrix, graph, or results file, when I try to find the file, it is nowhere to be found.

A. Sounds like you are saving to the C:\Program Files\HyperNiche folder or a subfolder. (In 64-bit Windows: C:\Program Files(x86)\HyperNiche)  For security reasons Microsoft doesn't think you should write any files to the Program Folder or subfolders so it writes to a Virtual Store that you cannot see or get to.

Here are some solutions:

• Save your files to any folder other than a Program Files folder, for example, Documents.
  
  
• With Windows Explorer navigate to the Program Files\HyperNiche folder (with 64-bit Windows to Program Files (x86)\HyperNiche folder) and click Compatibility Files in the Windows Explorer bar.
  
  
• Run HyperNiche as Administrator.  Even if you are logged in as Administrator you still have to right click the HyperNiche icon on the Desktop or on the Start Menu and select Run as Administrator.
  
  
• One time, right click the HyperNiche icon and instead of selecting Run as Administrator, select Properties | Compatibility | Privilege Level and check Run this program as an administrator.

Model list and evaluation fits differ

Q. Why does the fit (xR2 or logB) differ between the value given in the "Models" window and what I get with Fit Model | Evaluate Selected Model?

A. You should get the same fit if the overfitting controls are the same.  One of the most common culprits for this is the "Minimum Neighborhood Size for Estimate" -- this can have a small but noticeable influence on the measure of fit.  This controls how small N* can be before it is considered a missing value.  Missing values are not included when calculating the fit.

Estimates do not match

Q. Why are there small discrepancies between estimated values for individual points, when compared between Fit Model | Evaluate Selected Model and Graph | Response Points?

A. When you run Evaluate Selected Model, the estimates are based with the leave-one-out estimation in force.  This is because the object is to simulate the prediction error rate, by using the leave-one-out cross validation.  When you run Response Points, however, the goal is to make as good an estimate as possible for each point, so the leave-one-out estimation is turned off.  The difference you see for a given point is the influence of the omitting the observation for that particular point.  Because the observation is the target point, it will receive maximum weight, and is thus an influential data point.  Another consequence of turning off the leave-one-out crossvalidation is that you should see slightly larger neighborhood sizes for individual points when you are running Response Points.

Results don't match analysis settings

Q. When I run an analysis, it appears to be using settings from another data set or a previous run.  What is wrong?

A. If there is more than one copy of the executable file (e.g. HyperNiche2.exe), unpredictable things can happen.  Usually this happens after updates were installed to a different folder than the one in which the software was originally installed.   All kinds of bizarre and unpredictable behavior arises when updates are put into the wrong folder on Win7 and Vista machines, particularly with 64 bit machines.  This is a result of misplaced "caching".

To search for multiple instances with Vista or Windows 7, type in HyperNiche2.exe into the Search box (upper right) when you open Windows Explorer.  You should select "advanced" and tell it to look in all of drive C.  You should check the box to include "non-indexed, hidden, and system files".

If you find multiple folders with HyperNiche2.exe, figure out which one was the original folder, then delete the contents of the other one.  Then reinstall the updates being sure to select the original folder.  If you have a 64-bit system, there should be "(x86)" in the folder name.

HyperNiche vs R

Q. How does NPMR in HyperNiche compare with R?

A. To our knowledge there is no public fully nonparametric NPMR in R, while HyperNiche offers this, smoothly integrated with overfitting controls, randomization tests, bootstrapping, and GIS support.  Nor are we aware of any nonparametric regression in R that automatically builds leave-one-out crossvalidation into the model selection phase.   Furthermore, HyperNiche offers the convenience of a point-and-click graphical user interface, the power and flexibility of interactive graphics, a built-in context-sensitive help system, and free dedicated technical support.  None of those is available in R.

Data Structure

Missing data

Q. What can I do about missing data (cells in my matrices for which I have no data)?

A. You must take steps to replace missing cells with numerical values, or you must delete the offending rows or columns.  If you ignore warnings about missing data, they are most likely to be interpreted as zeros, which may or may not be reasonable.   There are several approaches to substituting values for missing cells.  We recommend reading Tabachnik and Fidell (1989) for a good discussion of the pros and cons of various remedies for missing data.  HyperNiche expects you to deal with the missing value problem rather than having an automatic way of dealing with it.  The reason for this is that for most multivariate analyses there simply is no good universal way of handling missing values.

Error message about a cell outside data matrix

Q. When HyperNiche reads my data I get an error message about a cell that is outside of my data matrix.  Why?

A. There is something in your spreadsheet outside the matrix.  Even if the cell looks blank, there may be something there.  For example, if you go to a cell in a spreadsheet, hit the space bar, then move to another cell, the cell will look blank, but there are actually data written in that cell: first a code saying there is a label field, next the "blank" character.  When you put the cursor on the cell in question the spreadsheet will display one of the characters that signals a label field: usually ' " or ^.  This is displayed not in the cell itself, but in the separate display that shows the contents and any imbedded codes in the current cell.  You can get rid of these by erasing the cell in question or a whole block of cells.

Using numbers as row labels

Q. I wish to label my sample units with numbers, but HyperNiche insists on character data in the first column.  It seems that the only way to get number labels into HyperNiche is to either enter them in compact format and import, or reformat each cell by hand in Quattro or Lotus, or insert a character before the first number in Excel.

A. This is a problem with the Excel export. It's too smart for its own good, converting numbers defined as characters into numbers defined as numbers.  We could change it to allow numbers in that field, but at some risk of clarity -- some people will surely omit the row labels and the analyses will be based on one too few columns.

Bad entry in cell x. Expecting number

Q. I get the error: "Bad entry in cell x. Expecting number."  I looked at the spreadsheet and can find nothing wrong.  What is the problem?

A. HyperNiche is expecting a number but finding something else.  Sometimes the problem can be "invisible." Examples:
- the letter o instead of the number zero
- an imbedded space in a number (the spreadsheet automatically considers this character data)
- an equation instead of a number

If you can't see the problem, re-enter the data in the cell and explicitly format the cell to be a number.  This problem can also produce the error message: "Error reading main matrix. row x. Label encountered in spreadsheet when expecting number."

Another way to quickly remove all cell formatting that HyperNiche does not recognize is to save the file in Excel as a .CSV (comma delimited) file type rather than .wk1 and then import that file into HyperNiche.

Error reading main matrix. Data read error

Q. I've been having no success in importing a CSV file into HyperNiche. I get the message "Error reading main matrix. Data read error."  How can I correct the problem?

A. Check to be sure you have included the header rows.  Look at the text file "tempdat.tmp" in the HyperNiche folder. T his file contains info for the the last item at the time the import failed.  This file contains one data item per line.   You should be able to see the data for the row that failed or any other information out of place.

Number of rows

Q. I'm having some problems reading my data matrix (*.wk1) and I'm hoping you can help.   The matrix is 13228 rows X 12 columns.  When I try to open it, I get an error message saying that "13228 rows" doesn't match the actual number of rows (8188).   There are no missing data values and no apparent reason why it should stop at row 8188.  It always stops at that exact row, no matter what I try.

A. At least some versions of Excel export only the first 8192 rows (8188 not counting the 4 header rows) to the *.wk1 file.  This is true regardless of which of the three *.wk1 formats is selected in Excel.  You can get all 13228 rows (actually up to 32,000 rows) of your data into HyperNiche by saving the file from excel in CSV format.   Then, the first time you open it in HyperNiche, you will need to import it (File | Import | CSV).  Then save it as a *.wk1 file.  From then on you can use it as a normal data matrix in HyperNiche.  But if you reopen it and change it with Excel, you will need to go the CSV route again to get it to save the whole thing.

Excel won't save in .wk1 format

Q. Excel 2007 does not allow me to save a file in .wk1 format. Are you planning any changes to HyperNiche input/output formats related to that?

A. We have already fixed HyperNiche to allow direct access to *.xls and *.xlsx files. Those fixes have been posted, so you just need to download the latest modules. The way it now works is that now you can use either *.wk1, *.xls, or *.xlsx files. But *.wk1 is faster. Select File | Import | Excel to open an *.xls or *.xlsx file.

Registry policy blocks *.wk1 in Excel

Q. I get this error when I try to open a *.wk1 file with Excel: “You are attempting to open a file type that is blocked by your registry policy setting.” What's the problem and how can I get around it?

A. This is an issue that arises from Service Pack 3 for Office 2003. Microsoft is looking at reversing this change. Download a registry file from Microsoft to fix the issue. Run it on all computers that use these files.

Can not import Excel

Q. I get an "Invalid variant operation" when I try to import an Excel .xls or .xlsx file.  Is there a solution?

A. First download the latest modules and see if that solves the problem.

Microsoft Excel must be on the computer in order to import.  (Open Office will not work.)  During import, Excel is loaded into memory invisible and sent instructions to open the file, and copy the data to the clipboard in text format so it can be accessed.  On a few systems this does not work for unknown reasons.  If you continue to have problems, save the file from Excel in CSV format and then import as CSV.

Can not import CSV

Q. I received a data file from a colleague overseas and cannot import the file.   What is the problem?

A. It sounds like the problem is due to the difference in the list and decimal separators, one using comma and period respectively, and the other using semicolon and comma.  You can temporarily match your colleague's language platform by Start | Control Panel | Regional and Language | Regional Options.  Then import the file.

Error exporting to Excel spreadsheet

Q. When I try to export a matrix to *.xls or *.xls I get an "OLE error" if there are more than 256 columns.  This happens in both Excel 2007 and Excel 2010.   What is wrong?

A. You may have set the default format for saving to *.xls (Excel 2003).  This will cause an error in later versions of Excel if there are more than 256 columns.   To fix this:

In Excel 2007:

1. Click the Office button in the upper left corner
2. Click the Excel Options button at the bottom of the drop-down menu
3. Select Save | Save file in this format: Excel Workbook (*.xlsx).

In Excel 2010:

1. Select File tab
2. Options
3. Save | Save file in this format: Excel Workbook (*.xlsx).

Then in HyperNiche, export as .xlsx file type.

Graphs

View two graphs at once

Q. How can I view two graphs at once?

A. Yes. You can start two instances of HyperNiche, putting each in a window occupying half of your screen.  There is some potential for confusion if the two "sessions" are working from the same graph files.  Be sure to File | Save As | Graph and give the graph file a unique name for each before graphing.

Insert graphs into MS Word

Q. What is the best way to insert HyperNiche graphics into word documents?

A. Consider the following tips:

• Edit the graph completely in HyperNiche -- do not attempt to re-edit the picture in Word.
• Use "save black & white" option (Graph | Options | Preferences) to avoid gray scales for figures destined for black-and-white reports and manuscripts.
• Drag labels into clear spaces for legibility. (Not possible in 3-D graphics.)
• Edit title, axes, and legend in HyperNiche for clarity and informativeness
• When a figure is going to be reduced greatly in its final form, select larger font sizes and larger symbols in HyperNiche.
• Use options for Legend Symbol/Color to select easily distinguished symbols.
• Results may differ in sizing and other ways if you insert pictures from saved files vs. cut/paste using the clipboard. You may need to experiment with this.
• Resize the pictures in Word only by dragging the corners, not the sides, so that the proportions of the graphic are not altered.
• To include additional text, symbols, or graphics, do it _on_top_of_ the imported pictures, not within them.  This avoids editing the picture with Word, which will usually force unwanted changes on the graphic.

Vertical axis becomes horizontal in MS Word

Q. When I insert a graphic into Word, the text in the vertical axis label becomes horizontal.  How do I fix this?

A.
1. Right click the graphic and select Edit Picture from the popup menu
2. Select the text box that you want vertical
3. From the menu select Format | Text Direction

Categories with text labels

Q. Is a way in which categories can be given short text labels on ordination graphs, so that points are labeled with a letter indicating their category?

A. You can plot the row names from the second matrix, instead of getting those names from the graph file.  You can use this facility to assign ANY labels you wish, to be included in the ordination.  One application would be to give text names to groups, as you suggest.  First create a second matrix with row names that reflect group categories and open this matrix as the second matrix in HyperNiche.  Assuming you already have done the analysis and have a graph file, open the graph window.  Select preferences in the menu and then select Labels From Second Matrix.

Edit and save graphics

Q. How can I edit and save graphic images?

A. Create a Graph. Then select File | Save Graphic As. The graph is saved as a Windows metafile (*.wmf).  This file can then be inserted into many word processors and other programs where you can easily resize and edit it.

Convert graphics to PDF

Q. Sometimes figures do not convert well to PDF.  The left axis label consistently ends up horizontal. How can I fix this?

A. Two suggestions:

1. Use the enhanced windows metafile format (*.emf) if you are not doing so already.
2. Try Adobe Acrobat Distiller. Graphics are, in general, more reliably converted to a PDF by Distiller than by regular Adobe Acrobat.

3D graph label quality

Q. The labels on my 3D graphs are not crisp and have ragged edges.  Is there a solution?

A. This is a problem with Windows smoothing of fonts.

In Windows XP:

1. Select Start
2. Control Panel
3. Display
4. Appearance
5. Effects and uncheck Smooth edges of screen fonts
6. Click Apply and OK

In Windows Vista:

1. Select Start
2. Control Panel
3. Personalization
4. Window Color and Appearance
5. Click on Open classic appearance properties for more options
6. Click on Effects
7. Uncheck Use the following method to smooth edges of screen fonts
8. Click Apply and OK

In Windows 7:

1. Select Start
2. Control Panel
3. System and Security
4. System
5. Advanced system settings
6. Settings under Performance
7. (or Performance Information and Tools
8. Advanced tools
9. Adjust the appearance and performance of Windows)
10. Uncheck Smooth edges of screen fonts
11. Click Apply and OK

If Windows 7 Control Panel is Viewed by icons instead of Category then

1. Select Start
2. Control Panel
3. Performance Information and Tools
4. Advanced tools
5. Adjust the appearance and performance of Windows
6. Uncheck Smooth edges of screen fonts
7. Click Apply and OK

See also 3D graph label holes or rotation trails

3D graph label holes or rotation trails

Q. Labels that are behind my 3D graph show through and/or all labels leave trails when rotating the graph.  Is there a solution?

A. In the Graph | Preferences | Format tab check Hide labels that are behind graph.
Alternate solutions to try are:
Check Label one axis side so there are no labels behind the graph
Uncheck Adjustable labels

If you are Vista or Win 7 user and getting rotation trails, you need to take an additional step:

1. Right click on empty part of screen
2. Select Personalize
3. Select Window color and appearance
4. Select any of the Color schemes other than Windows Vista Standard or Windows Aero

Note, for Vista or 7 users, the problem only applies to what you see on the screen.   Labels will be properly hidden without the extra step of changing the screen appearance with Copy, Save, Print Preview, Print, and Save Animated GIF.

See also 3D graph label quality

Saving at high resolution line widths

Q. When I increase the width of lines in PC-ORD graphs and save the graph at high resolution (400% or more), then open the graph in other software (PhotoShop or Illustrator), the lines become hard to see. How can I fix this?

A. The higher you set the resolution (File | Resolution for Save and Copy), the more you have to compensate with larger fonts and heavier line weights, just because of the differences in the ways lines are rendered on the screen (which has a fixed relatively low resolution) and when writing a file. Also, do not use a higher resolution than necessary. You will almost always exceed publisher specs if you set the resolution to 300%. But even at this level you should compensate by with heavier line weights..

NPMR Models

NPMR equation

Q. Is there is a way to get an equation or a series of equations out for an NPMR model?   I find it difficult to evaluate the differences in behavior among models with the same numbers of variables but different tolerances or xR2.  So it feels very much like a black box.

A. The flexibility and power of NPMR actually comes from forfeiting a single equation.   But you can study the model behavior in a several ways:

1. Do 2D plots (Graph | Response Points) of various combinations of variables (including the residuals, estimated values, observed values, and various predictors).
2. Do 3D plots (Graphs | Response Curves | 3D) with the response variable against 2 predictors, then rotating the graph to observe the form of the relationship.
3. Use Graph | Response Points to force values of certain predictors while varying another predictor.

Choosing LM (local mean) versus LL (local linear)

Q. How does one choose between LM and LL models? In evaluating performance, the LL models consistently yield higher xR-squared. Is this a valid reason to choose LL over LM?

A. I usually choose LM vs. LL based on the nature of the data. LM models guarantee that the estimate is always within the range of observations. But they have stronger bias toward the central tendency of the responses than LL. This can be undesirable near the edges of the response curves -- LM curves will tend to flatten slightly at the edges while local linear models will run the local trend right to the edge of the response curve. I tend to use LM for species response data because I never want a prediction of less than zero abundance. I tend to use LL for more normally distributed responses to reduce the bias at the edges and give a little higher fit.

How to interpret N* (average neighborhood size)

Q. I am not sure how to make sense of the N* values. There is considerable variation in neighborhood size among sample units. If a sample unit has too small a neighborhood, their responses are not predicted. Does this mean these samples are outliers?

A. N* is just how much data is bearing on a particular estimate. So you can use that to restrict modeling and prediction to parts of the predictor space where you have what you consider to be an adequate sample. You can apply it with different values in the model fitting phase (as average N*) and the generation of response surfaces (as point-by-point N*). I tend to make it more conservative (require a larger N*) in the model fitting phase so that the resulting shape doesn't overfit the areas of weak data. But you can use a lower criterion N* when generating a response surface to help fill in some of the gaps in the surface.

Just because you have few data points in a particular area of the predictor space doesn't mean the response is an outlier there. Most of the time it is not, but IF IT IS an outlier, then that single data point has undue influence on the response surface.

N* is pretty much directly interpretable as local sample size, even though it is calculated as the sum of contributions from individual data points. So N* = 1 means you have the equivalent of one full data point to make an estimate.

Tolerance

Q. Tolerances seem like coefficients in a regression equation. What am I to make of their magnitudes relative to their ranges?

A. Tolerance are smoothing parameters. They control the breadth of local weighting functions used in the kernel smoother. In a LM model you can interpret the tolerance as something like a coefficient expressing the importance of a variable similar to a regression coefficient,(though smaller tolerance is more important). However that is not true with LL models. It's really better to use sensitivity analysis to evaluate the importance of individual predictors -- this works for any kind of model with quantitative predictors. See Sensitivity Analysis in the Help system.

The way that tolerances are expressed depends where you are looking in the output. Sometimes tolerances are reported as absolute values, sometimes as proportions of the range. In the Model List Window they are given as absolutes. In the case of a Gaussian weighting function, the tolerance is the span of a predictor covered by one standard deviation of the Gaussian function. Tolerances can be made more comparable among variables that have different ranges by dividing each tolerance by the range of that predictor. In the Evaluate Selected Model results they are given both ways, for example:

PREDICTORS

GIS Data

ArcView segmentation violations

Q. When exporting a GIS grid I'm having trouble with segmentation violations when I import the new grid to ArcView.

A. Be sure that you are following the naming conventions for ArcView files.  Do not include a space anywhere in the whole path name.  Certain other non-alphabetic and non-numeric characters are disallowed too.  It is safest to follow the 8+3 naming convention for file names with older software.  To do this, use 8 characters maximum in the first part of the file name, followed by a dot, followed by an extension of 3 characters at most (usually "asc" in this application).

ArcView FLOAT or INT error

Q. When exporting GIS grids to ArcView, it could only read the file if it contained binned data.  If I had it output as real numbers, I would get a FlOAT or INT error from arcview.

A. In the ArcView import dialogs you are asked what kind of data you have.  For example, after selecting File | Import Data Source | ASCII Raster, you select the *.asc file, then specify the output grid, then it asks if it should read "cell values as integers."  You should answer NO if you have real numbers, YES if you have binned estimates.

No GIS estimates

Q. I am trying to export GIS grids from Hyperniche.  Sometimes I get the error "No GIS estimates were calculated successfully".  I have checked the GIS grid file and it looks right and I have experimented with a variety of minimum neighborhood sizes.

A. The most likely cause of this is that the whole grid is being assigned the missing value.  This could happen in a number of ways, for example:

1. The values of one or more predictors are out of range of the calibration data set (for all parts of the grid).
2. One or more input grid is in different units than the corresponding variable in the calibration data.
3. The format of one or more input files is incorrect.
4. We have encountered one case of that, when a program (non-ESRI) wrote the text file in a slightly different format.  You can check this by comparing the first few lines of one of your input grids to the format given in the online help.