HyperNiche Use Overview


1. Prepare Your Data

  1. Put your data in a spreadsheet (e.g. Excel) with the sites as rows and variables as columns.  Assign a name to each row and column.

  2. Partition your variables into two worksheets, one with species response variables (presence, abundance, or other measure of performance), and the other with predictors (habitat variables).  Copy the row names into both worksheets.

  3. Insert header rows declaring the contents of your response matrix, and whether each variable is quantitative (Q) or categorical (C).

    ResponseExample.gif (23983 bytes)
    PredictorExample.gif (15630 bytes)
  4. Save each of these two worksheets as a separate spreadsheet in *.wk1 format.

2. Open HyperNiche and Your Data Files

Start HyperNiche, then open your response matrix, the file containing your response variables:
FileOpenResponse.gif (19219 bytes)
Open your predictor matrix in the same way.


3. Fit Models to Your Data

  1. Use Fit Model | Free Search with your method of choice.  For general purposes we recommend nonparametric multiplicative regression based on a local mean and Gaussian weighting function (LM-NPMR).

    FreeSearch.gif (7148 bytes)
    FreeSearchSetup.gif (16636 bytes)

  2. HyperNiche examines a large number of models (2008 in this case), then lists the results for those models in the Model List window in the upper right corner of your screen.

    UnsavedModelList.gif (20083 bytes)

  3. Filter and save the best models for each response variable for each number of predictors.  Use Edit | Delete All But Best For N Predictors.

    DeleteAllButBest.gif (8107 bytes)

    This results in 12 models retained. For each of the four response variables, the best 1-, 2-, and 3-predictor models are saved.  The model list is then saved by choosing File | Save As | Unsaved Model List, then supplying the file name, in this case CraneExample.spx.

  4. Choose the number of predictors by evaluating the diminishing returns of adding predictors.  Consider the fourth response variable, abundance of species Isomyo.   Adding a third predictor "Live", resulted in little improvement in fit (measured by xR2, the cross-validated R2).  So we chose to pursue the model with two predictors, LogDia and Height.

4. Explore Your Models Graphically

You can explore your models with:

  • 2D and 3D response surfaces
  • Estimated vs. observed values
  • Residual plots
  • Partial models

For example, generate a 3D response surface as follows.  Select the model you wish to graph by clicking on the model in the model list window.  In this case we chose model number 1540, relating Isomyo to the two predictors LogDia and Height.

ModelListCut.gif (28056 bytes)

We get the following graphic:

IsoMyoResponseWireAve3.gif (8864 bytes)

It is a bit difficult to see the full response shape, so let's rotate the graph for a different view:

IsoMyoResponseWireAve3B.gif (9882 bytes)

The response surface is broken in the areas where there was insufficient data, as set by the minimum average neighborhood size.  This parameter was set during the model fitting phase.  To achieve a smoother, continuous curve we can increase that parameter from 3 to 10.

The following graphic shows the best 2D model with this stronger smoothing:

IsoMyoResponseWireAve10.gif (7725 bytes)