Statistical Testing Method for Continuous Predictors

Processes | Predictive Modeling | Statistical Testing Method for Continuous Predictors

Statistical Testing Method for Continuous Predictors

Use the drop-down menu to select the statistical method to be used for filtering the continuous predictors .

Statistical options for reducing predictor continuous variables are listed in the following table:

Statistical Test ¹

Description

Pearson Correlation

Choose this option to compute a parametric measure of association for each pair of variables .

See Pearson Product-Moment Correlation for more information.

Hoeffding Correlation

Choose this option to compute a a nonparametric measure of association that detects more general departures from independence.

See Hoeffding Dependence Coefficient for more information.

Kendall Correlation

Choose this option to compute a nonparametric measure of association based on the number of concordances and discordances in paired observations .

See Kendall’s Tau-b Correlation Coefficient for more information.

Spearman Correlation

Choose this option to compute a nonparametric measure of association based on the ranks of the data values.

See Spearman Rank-Order Correlation for more information.

T-Test

Choose this method to conduct a t-test on the group means .

Refer to the SAS PROC MULTEST documentation for more details.

Shrunken T-Test

Choose this method to use empirical Bayes methods to shrink the variance across tests.

Refer to the SAS PROC MIXED documentation for more details.

Unequal Variance T-Test

Choose this method to fit a separate variance for each group as in a Welch's t- Test.

Refer to the SAS PROC MIXED documentation for more details.

Asymmetric Loss T-Test ²

Choose this method to assign different weights to observations with residuals that are positive versus those that are negative.

When you select this option, you must also specify a value, denoted as f , in the Asymmetric Loss Fitting Proportion text box . When a residual is positive, a weight of f is used in the T-test, and when it is negative a weight of 1 - f is used.

Quantile Regression Test

Choose this method to fit a quantile regression model to each group.

Refer to the SAS QUANTREG documentation for more details.

Quadratic Trend Test

Choose this method to fit a quadratic regression model to each group.

Refer to the SAS PROC REG documentation for more details.

Radial Basis Test

Choose this method to fit a radial basis kernel function to each group.

Refer to the SAS PROC GLIMIX documentation for more details.

Rank Sum Test

Choose this method to use a Wilcoxon test .

Refer to the SAS PROC NPAR1WAY documentation for more details.

Median-Score Test

Choose this method to use a median score test.

Refer to the SAS PROC NPAR1WAY documentation for more details.

Robust Chi-Square Test

Choose this method to fit a robust regression model that reduces the effect of the outliers.

Refer to the SAS PROC ROBUSTREG documentation for more details.

1

Note : Available tests may vary depending on the analytical procedure you are using.

2

Efron, B. (1991). Regression percentiles using asymetric squared error loss. Statistica Sinica 1: 93-125.

To Specify a Statistical Test:

Check the Use statistical testing to filter predictors box.

Select the desired statistic using the drop-down menu.

Statistical Test ¹	Description
Pearson Correlation	Choose this option to compute a parametric measure of association for each pair of variables . See Pearson Product-Moment Correlation for more information.
Hoeffding Correlation	Choose this option to compute a a nonparametric measure of association that detects more general departures from independence. See Hoeffding Dependence Coefficient for more information.
Kendall Correlation	Choose this option to compute a nonparametric measure of association based on the number of concordances and discordances in paired observations . See Kendall’s Tau-b Correlation Coefficient for more information.
Spearman Correlation	Choose this option to compute a nonparametric measure of association based on the ranks of the data values. See Spearman Rank-Order Correlation for more information.
T-Test	Choose this method to conduct a t-test on the group means . Refer to the SAS PROC MULTEST documentation for more details.
Shrunken T-Test	Choose this method to use empirical Bayes methods to shrink the variance across tests. Refer to the SAS PROC MIXED documentation for more details.
Unequal Variance T-Test	Choose this method to fit a separate variance for each group as in a Welch's t- Test. Refer to the SAS PROC MIXED documentation for more details.
Asymmetric Loss T-Test ²	Choose this method to assign different weights to observations with residuals that are positive versus those that are negative. When you select this option, you must also specify a value, denoted as f , in the Asymmetric Loss Fitting Proportion text box . When a residual is positive, a weight of f is used in the T-test, and when it is negative a weight of 1 - f is used.
Quantile Regression Test	Choose this method to fit a quantile regression model to each group. Refer to the SAS QUANTREG documentation for more details.
Quadratic Trend Test	Choose this method to fit a quadratic regression model to each group. Refer to the SAS PROC REG documentation for more details.
Radial Basis Test	Choose this method to fit a radial basis kernel function to each group. Refer to the SAS PROC GLIMIX documentation for more details.
Rank Sum Test	Choose this method to use a Wilcoxon test . Refer to the SAS PROC NPAR1WAY documentation for more details.
Median-Score Test	Choose this method to use a median score test. Refer to the SAS PROC NPAR1WAY documentation for more details.
Robust Chi-Square Test	Choose this method to fit a robust regression model that reduces the effect of the outliers. Refer to the SAS PROC ROBUSTREG documentation for more details.