Normal probabilitiy plots
A normal linear model assumes that the model errors are normally distributed. A probability plot of residuals helps to detect non-normality.
Deer jaw length and weight
Warning
If the assumptions of linearity and constant variance are violated, or if there are outliers, the probability plot of residuals will often be curved, irrespective of the error distribution.
Only draw a probability plot if you are sure that the data are linear, have constant variance and have no outliers.
Outliers — Deer data
In the Deer data above, most of the probabilitiy plot is fairly linear. The long tails to the residual distribution could perhaps be caused by a small number of outliers in the data.
If possible, it would be worth checking that there were no measurement or transcription errors for the most extreme residuals. Perhaps these measurements were made by one particular hunter who incorrectly measured jaw length?
Nonlinearity — Antibiotic effectiveness
Nonlinearity in the relationship can have the opposite effect on the shape of a probability plot — the tails of the distribution of residuals may be shorter than those for a normal distribution.