# The SPP Procedure

### Example 105.2 Testing Covariate Dependence of a Point Pattern

In most spatial analysis applications, you are likely to have one or more covariates in addition to the point pattern data set. Hence, you can test for a possible dependency between the observed point pattern and the covariates by using covariate dependency tests that compute empirical distribution function (EDF) statistics. These tests are nonparametric, and the selected EDF statistic indicates whether the covariate values interpolated at the point locations are independent of the transformed covariate and a known model of point pattern intensity such as CSR. Covariate dependency testing serve multiple objectives:

• They tell you whether the empirical distribution of the covariates at the point locations and the empirical distribution function weighted and transformed according to the underlying intensity model (or the predicted distribution) are similar.

• In cases of dissimilarity between the empirical distribution and the predicted distribution, the interpretation is that the covariate gives evidence against the intensity model (CSR in this case).

To request a covariate dependency test that is based on an EDF statistic, you use the COVTEST statement, in which you specify the point process and the covariates that need to be tested. The following statements perform a covariate dependency test that is based on an EDF statistic:

```proc spp data=sashelp.bei;
process trees = (x,y /area=(0,0,1000,500) event=Trees);
trend elev = field(x,y,Elevation);
run;
```

When you do not request any specific EDF test statistic, by default the SPP procedure produces an EDF test that uses the Kolmogorov-Smirnov statistic for each covariate that is specified in the COVTEST statement, as shown in Output 105.2.1.

Output 105.2.1: Weighted EDF Test Statistic

The SPP Procedure

Kolmogorov-Smirnov Weighted EDF
Test for Covariate Values
Source D Statistic p Value
Output 105.2.1 reports the value of the Kolmogorov-Smirnov D test statistic and a p-value. The p-value indicates that the null hypothesis of independence on the covariate is rejected. In addition to the test statistic and the p-value, PROC SPP produces a plot of the empirical and transformed distribution function for each covariate. Output 105.2.2 and Output 105.2.3 show the plots of the Kolmogorov-Smirnov statistic for the covariates `Gradient` and `Elevation`, respectively. In each figure, the plot on the left includes the empirical density function (EDF) of the variable (solid blue line) and the weighted EDF (transformed EDF or transformed line). The plot on the right is a PP plot that plots the Empirical probability (EDF) versus the Transformed probability( Transformed) against the reference standard uniform (red line). The same plot also shows the largest vertical difference between the normal and uniform lines, which is the Kolmogorov-Smirnov statistic D. From the right plots in Output 105.2.2 and Output 105.2.3, it is quite apparent that the EDF by Transformed line overlaps only at the ends for the `Gradient` covariate and crosses the uniform line once for the `Elevation` covariate. Thus, you can infer that the `Gradient` covariate (in addition to having a higher D statistic value) deviates considerably from the uniform line.
Output 105.2.2: Kolmogorov-Smirnov EDF Test Plot for `Gradient`
Output 105.2.3: Kolmogorov-Smirnov EDF Test Plot for `Elevation`