Homework

• Acquire SmartPLS. Now is the time to make sure you acquire and correctly install the SmartPLS software used in Week 14 for coverage of partial least squares.. Although free, you must register for it, then follow the download instructions. To register, go to http://www.smartpls.de, click on "Register now!" and follow the instructions. You will get back two emails, one with the username and password you have chosen and one with your activation code. Go back the the pls website and click on "Download" in the upper right to download the program, the example files, and the handbook. The instructions mention you will have to have the Java Runtime Environment, but it is probably already loaded on your machine (if not, you will have to go to http://www.sun.com to download it). When you first run the program, which you should do now, it will ask for the activation code you were emailed. The program will be active for three months, though there is an option to extend it through additional activation codes obtained on the website.

• Partial Correlation. Print out the output as described on p. 216 of Chapter 11 and be prepared to comment on it, then turn it in. In addition, create a version of Figure 11-1 with the data points labeled, as explained a the end of the chapter. Note that when you are doing Figure 11-3, when you close the Chart Editor, the reference line will appear in the SPSS Output Viewer. Skip Figure 11-6. For Figures 11-8 and 11-9, note you must have two variables selected to enter them into the Paired Variables list; you cannot enter them individually one at a time. Skip Figure 11-10. For Figure 11-11, add abdomcirc to the varlist as well as those Norusis lists on p. 212.

  Printing output: To export to Word and print, highlight Output at the top of the outline pane on the left, then right-click and select Export. Export as type HTML file (the instructor found on his setup that this worked better than using type Word/RTF). After saving, load the file into Word, select View, Print Layout; then Insert, Page Numbers, add your name, and print to bring to class.

• Regression. Print out the regression output as described in Norusis I, pp. 273-274 and be prepared to comment on it, then turn it in.

  For the WLS regression part, use the file wls_draper.sav, which you may obtain by going to http://www2.chass.ncsu.edu/garson/pa765/wls_draper.sav and clicking the "Save" button when prompted. This file is an approximation of the data shown in Norusis 2, Figure 15-1. Since it is an approximation, the coefficients will not be exactly the same as in the textbook. Once you have downloaded the file, load it into the SPSS data editor and do the following:

  1. Run a regression of y on x, accepting the default options, but click the Save button to save unstandardized residuals and unstandardized predicted values which SPSS will label RES_1 and PRE_1 respectively.

  2. Plot residuals on Y vs. predicted on x, using Graph, Legacy Dialogs, Scatter/Dot, Simple Scatter, Define.

  3. Do weight estimation (Analyze, Regression, Weight Estimation) as described in Norusis 2, p. 329. Be careful as instructed to use the Options button to save weights as a new variable, which SPSS will label WGT_1.

  4. The the regression as before (Analyze, Regression, Linear), but this time putting in WGT_1 as the WLS weight variable. SPSS will save PRE_2 and RES_2.

  5. If you try to plot directly from Regression, you will get this warning: "No plots are produced for Weighted Least Squares regression. You can SAVE the appropriate variables and use other procedures (e.g., EXAMINE and PLOT) to produce the requested plots. To plot weighted versions of the residuals and predicted values, use COMPUTE before plotting: COMPUTE RESID = SQRT(REGWGTvar) * RESID COMPUTE PRED = SQRT(REGWGTvar) * PRED."

     Therefore, create weighted predicted and weighted residual values by using Transform, Computer Variable to compute WtdPred=SQRT(WGT_1)*PRE_2, and WtdResid=SQRT(WGT_1)*RES_2.

  6. Plot WtdResid on the y axis vs. WtdPred on x, using Graph, Legacy Dialogs, Scatter/Dot, Simple Scatter, Define.

Explore

PARTIAL CORRELATION

• Partial Correlation Slide Show (Ansorge)
• Partial Correlation (VassarStats)
• Partial Correlation Calculator. Enter correlations, see computed partial correlation.
• Partial and Semi-Partial Correlation Slideshow (1 MB file)
  REGRESSION
  
• Movie Tutorials on Regression
• Introduction to Regression, Analytic Tech
• Introduction to Regression, with SPSS Output, University of Exeter
• Regression Tutorial Using Spreadsheets, University of Illinois - Urbana-Champaign
  WLS REGRESSION
  
• Weighted Least Squares Regression (.doc format)