Troubleshooting
Problem
I ran the SPSS Linear Regression procedure with several predictors and requested partial plots from the Plots dialog for that procedure. One of my predictors was binary with values of 0 and 1. In the partial plot for this predictor, the predictor is on the horizontal axis and the dependent variable (DV) is on the vertical axis. I expected the points on the plot to form 2 columns at the values of 0 and 1 on the horizaontal axis, but they are spread across the graph with some predictor values less than 0 and some predictor values greater than 1. What does this partial plot represent and why do the points not match observed values on the variables?
Resolving The Problem
Partial regression plots for a dependent variable (DV) and predictor are scatterplots of the residuals from 2 regressions - regressing the DV on all of the other predictors, and regressing that particular predictor (as DV) on all of the other predictors.
The regression analyses that are run by the syntax commands below illustrate the meaning of a partial plot. The data for these regressions is in the file 'Employee data.sav', which is included with SPSS and can be found in the directory where SPSS was installed in your computer.
REGRESSION
/MISSING LISTWISE
/STATISTICS COEFF OUTS R ANOVA ZPP
/CRITERIA=PIN(.05) POUT(.10)
/NOORIGIN
/DEPENDENT jobcat
/METHOD=ENTER prevexp educ minority
/PARTIALPLOT ALL .
REGRESSION
/MISSING LISTWISE
/STATISTICS COEFF OUTS R ANOVA ZPP
/CRITERIA=PIN(.05) POUT(.10)
/NOORIGIN
/DEPENDENT jobcat
/METHOD=ENTER prevexp educ
/SAVE RESID (res_jobcat) .
REGRESSION
/MISSING LISTWISE
/STATISTICS COEFF OUTS R ANOVA ZPP
/CRITERIA=PIN(.05) POUT(.10)
/NOORIGIN
/DEPENDENT minority
/METHOD=ENTER prevexp educ
/SAVE RESID (res_minority).
CORRELATIONS
/VARIABLES=RES_jobcat RES_minority
/PRINT=TWOTAIL NOSIG
/MISSING=PAIRWISE .
The first regression is the main regression of interest. JOBCAT ('Employment category') is the dependent variable. (JOBCAT is actually ordinal, but is treated it as continuous for this example.) PREVEXP, EDUC, and MINORITY are the predictors. MINORITY is an indicator, or dummy variable, scored as 0 or 1. Partial plots were requested, as were the part and partial correlations of each predictor with JOBCAT.
In the Coefficients table for this first regression, note that the partial correlation for minority equals -0.12. This is the correlation of MINORITY and JOBCAT with the effects of the other predictors removed. In the partial plot for JOBCAT and MINORITY, you can see that the values don't form columns of points at values of 0 and 1 on the X-axis as they would if the X-axis represented the observed values of MINORITY. The x-axis represents residuals, i.e. errors in predicting MINORITY from EDUC and PREVEXP. The Y-axis represents residuals from predicting JOBCAT from EDUC and PREVEXP
The second regression has JOBCAT as DV but only EDUC and PREVEXP are used as predictors. The unstandardized residuals are saved to a new variable, RES_JOBCAT. The third regression uses MINORITY as the DV with EDUC and PREVEXP as the predictors. The residuals from this regression are saved to the new variable RES_MINORITY. Saving the residuals was the purpose of the second and third regressions. The names for the new variables were assigned by placing them after RESID on the /SAVE subcommands in the second and third regressions. You can run all of these regressions from the menu system, but the Regression dialogs don't give you an opportunity to name the saved variables. The residuals would then be saved to generic names like RES_1 and RES_2, which you can then change in the Variable View of the Data Editor.)
In the Correlations table after the third regression, you can see that the correlation between RES_JOBCAT and RES_MINORITY is -0.12, which was the partial correlation for JOBCAT and MINORITY as reported in the first regression. Finally, the Curvefit procedure is run with a linear fit line to predict RES_JOBCAT by RES_MINORITY. The scatterplot is just like the partial plot for minority from the first regression (except for the addition of the fit line). The b1 estimate in curvefit, which is the slope for predicting RES_JOBCAT from RES_MINORITY, is -.189, which was the regression coefficient for MINORITY in the first regression
Related Information
Historical Number
69152
Was this topic helpful?
Document Information
Modified date:
16 April 2020
UID
swg21478071