#
Exercise: Multiple regression of muscle strength on age and height, 3

Question 3: Carry out graphical checks of the assumptions that the residuals follow a Normal
distribution with uniform variance.

##
Suggested answer

SPSS will provide a number of plots to check the assumptions.
Alternatively, we can save the predicted values and residuals and use
the __Graphs__ commands.
We did this in Question 6 in Week 9.

Click __Analyze__, __Regression__, __Linear__.
Select Quadriceps strength into __Dependent__ and
Height and Age into __Independents(s)__.
Now click __Plots__.

I did not actually do these plots in the lecture (my mistake),
but we did them for the simple linear regression in
Questions 5 and 6 in Week 9.
The only difference is that we cannot plot the residuals against the predictor,
because there is more than one.
We plot against the value predicted by the regression equation instead.

To plot residuals against predicted values we need to put one of
the residuals variables into Y and one of the predicted variables into X.
ZPRED and ZRESID are standardised, i.e. made to have mean zero
and standard deviation one.
ADJPRED and DRESID are in the original units, the natural predicted strength
in newtons and the difference of the observed and predicted strength in newtons.

I chose the natural units, so put __*ADJPRED__ into __X:__
and __*DRESID__ into __Y:__.
I also clicked __Histogram__ and __Normal probability plot__.
Click __Continue__.
Click __OK__.

You will get five tables of output, then three graphs.
I have edited these to make them easier to read on screen.

The histogram has too many intervals for only 41 observations,
so we can make the intervals larger:

The distribution is a bit skew to the left, but not much.
The Normal plot also has a bit of a curve in it.
The Normal plot is of the P-P type, rather than the Q-Q type, but the interpretation is the same.
A curve towards less steep indicates negative skewness.

The plot of residual against predicted shows a possible relationship,
with residuals being smaller at the ends and bigger in the middle.
The model does not appear to be a very good fit to the data.

Back to Exercise: Multiple regression of muscle strength on age and height.

To Applied Biostatistics index.

To Martin Bland's M.Sc. index.

To Martin Bland's home page.

This page maintained by Martin Bland.

Last updated: 19 December, 2006.

Back to top.