Independence of Residuals (cont.): the same residual vs. predicted plot that you used for "homoscedasticity" and "correct relationship between DV and IV" can be a rough way of testing this. But another (I think better) way would be to make a boxplot similar to Figure 4.4.6 (C) on p. 135 of the text.
To do that:
1) create a residual variable.
a) in your linear regression window, click "Save"
b) check the box for "Standardized" under "Residuals."
c) run the analysis
2) go to Graph --> Chart Builder --> Boxplot --> Clustered Boxplot.
Now you can put the "Standardized Residual" on the Y axis, a categorical variable in the X axis clustered by another categorical variable. The medians should be similar if there is independence of residuals.
Technically, you could also transform height into a categorical variable to cluster around, but you guys are probably getting exhausted already. So don't worry about it.
You wouldn't have to worry about serial dependency because this is cross-sectional data.
Please ask if you have more questions.
Good luck!
1 comment:
I should add that for the independence of residuals, you can also make a plot like Figure 4.4.6 (A). To do that, go to Analyze --> Forecasting --> Sequence Chart. Set your residual (that you created) as your variable and click OK.
Post a Comment