variance reduces as the response values increase. With so few data points such an
interpretation of this actual plot would not be sensible.
The plot of residuals vs. case ID in this example has no meaning because the data were
collected at the same point in time. The adequacy of model fit is shown
Figure 8.7c: Fitted least squares
regression line for prediction of
standardized maths ability score from
teacher estimate of maths ability with
95 per cent confidence level for
individual predicted values
in Figure 8.5 where residuals are plotted against the explanatory variable. Again the
suggestion of a pattern to the data, a funnel shape, narrowing as MATHS score increases,
is indicative of a non-linear relationship, i.e., lack of model fit. It will be clear by now
that interpretation of residual diagnostic plots is somewhat subjective and if you have any
doubts about the model fit then try different variables or combinations of variables and
Statistical analysis for education and psychology researchers 278