14.03.2014 Views

Modeling and Multivariate Methods - SAS

Modeling and Multivariate Methods - SAS

Modeling and Multivariate Methods - SAS

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

Chapter 13 Recursively Partitioning Data 341<br />

Example<br />

Categorical<br />

If the Missing Value Categories checkbox (see Figure 13.2) is not selected, missing rows are not included in<br />

the analysis. If the Missing Value Categories checkbox is selected, the missing rows are entered into the<br />

analysis as another level of the variable.<br />

Continuous<br />

Rows with missing values are not included in the analysis.<br />

Missing Predictors<br />

The h<strong>and</strong>ling of missing values for predictors depends on if it is categorical or continuous.<br />

Categorical<br />

If the variable is used as a splitting variable, <strong>and</strong> if the Missing Value Categories checkbox (see Figure 13.2)<br />

is not selected, then each missing row gets r<strong>and</strong>omly assigned to one of the two sides of the split. When this<br />

happens using the Decision Tree method, the Imputes message appears showing how many times this has<br />

happened. See Figure 13.16.<br />

If the Missing Value Categories checkbox is selected, the missing rows are entered into the analysis as<br />

another level of the variable.<br />

Continuous<br />

If the variable is used as a splitting variable, then each missing row gets r<strong>and</strong>omly assigned to one of the two<br />

sides of the split. When this happens using the Decision Tree method, the Imputes message appears<br />

showing how many times this has happened. See Figure 13.16.<br />

Figure 13.16 Impute Message<br />

Ten observations were<br />

missing <strong>and</strong> r<strong>and</strong>omly<br />

assigned to a split<br />

Example<br />

The examples in this section use the Boston Housing.jmp data table. Suppose you are interested in creating<br />

a model to predict the median home value as a function of several demographic characteristics.<br />

The Partition platform can be used to quickly assess if there is any relationship between the response <strong>and</strong> the<br />

potential predictor variables. Build a tree using all three partitioning methods, <strong>and</strong> compare the results.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!