搜尋此網誌

2017年8月11日 星期五

Residual Analysis

To get SPSS to highlight any outliers, we click the Statistics button in the regression window, and tick the box for Casewise Diagnostics.

Note that the default value of standardised residual SPSS thinks of as "unusual" is 3, but edit this to 2 to get more information. Any unusual observations will be displayed in a table in the output.
To study the Normality (or otherwise) of the residuals, click the Plots button in the regression window, and select both the Normal Probability Plot and the related Histogram. Remember that if Normality is present in the residuals, we would expect the points in the Probability Plot to fall on the straight line.
From the Plots button menu we can also obtain a plot of the standardised residuals against (standardised) predicted values (of the dependent variable). Select ZRESID and ZPRED from the menu (as y-axis and x-axis respectively). Remember we are looking for random scatter in this plot. In particular, look out for "funnelling" illustrating that the variance is not constant, and a functional relationship, illustrating a deficiency in our model form.
Finally, we want to plot the standardised residuals against (some of?) the independent variables. To do this, we need to use the Save button/menu to store the standardised residuals in a variable. Once this is done, and the regression OK button clicked, you can use the SPSS plotting facilities to create the relevant scatterplots. It probably only makes sense to make these plots if an independent variable is interval scaled (or continuous). Do you have the desired random scatter?