[Skip Global Navigation]

SPSS Tech Tips

SPSS Direct Home

Using the output of a PCA/Factor analysis within a Regression model

This is a complimentary tip available only to SPSS Direct subscribers.

Question

I am using Clementine Client and would like to specify predictor fields to reduce and simplify the amount of data used in my analysis. How do I use the output of a PCA/Factor analysis within a Regression model?

Answer

In Clementine® 9.0 you can use the output of a generated PCA/Factor model by following the steps listed below. Please note that this process is simplified by the introduction of the Feature Selection node in Clementine 10.0.

To create a new PCA/Factor model, place a Source node for your data onto the canvas. Then connect a Type node to the Source node. As shown below, configure the Type node so that the direction of your dependent variable is set to “out” and the potential factors are set to “in.” Connect a PCA/Factor Modeling node by utilizing the Type node settings rather than designating new settings in the Modeling node. Then, you must execute the stream to generate the PCA/Factor model for use within your Regression model.


Figure 1: Type node configured for use in Factor analysis.
(Click to enlarge)

Look under the “Models” tab in the upper right hand corner, find the model you’ve generated, and drag it to the stream canvas. Connect your model to the stream from the data source. It should connect at the same spot as a Modeling node.

Next, connect a Type node to the generated model, and specify the direction of the target and factor variables. As illustrated below, the target variable should be set to “out,” while the factor variables should be set to “in.”


Figure 2: : Target variable type configured for use in Regression analysis.
(Click to enlarge)

Finally, connect the Regression node to the Type node. Since we already specified the direction of “Esti_annual_spend” as “out,” we can use the default setting of “Use Type node settings” within the properties of the Regression node. You can now alter the Regression node and make additions to your stream as needed. The stream is ready for execution.

As mentioned earlier, the process of data reduction can be simplified in Clementine 10.0 by using the Feature Selection node. This node screens predictor fields for removal by using a set of criteria and ranks predictors relative to a specified target.