Gunter, Bert (1994, ASQ)
Normality is best viewed as a benchmark for comparison, rather than an immutable standard, since there are situations in which normal distribution could represent abnormal behavior. When there is no stable underlying population from which to draw inferences, the sources of instability must first be identified and removed before the inferences have meaning. An understanding of how the shape of the distribution changes over time and space can provide valuable clues. Quantile-quantile (Q-Q) plots are an effective way to compare two distributions. The most typical use of Q-Q plots is to benchmark against normality by plotting sample quantiles on the y axis against theoretical quantiles of a standard normal distribution on the x axis. On some occasions it may be more appropriate to use other theoretical distributions besides the normal distribution as a benchmark. With the proper software, Q-Q plots can be modified to handle such situations. Because of their versatility and ease of construction, the Q-Q plot is a powerful graphical tool in any data analysis toolbox.