Re: [Gretl-devel] Q-Q plot, again

Sven, et al, Here is how Hamilton and Cleveland frame their use of the qq plots. Lawrence Hamilton's book stresses analytical graphics. In the intro he states, "Regression summarizes (or models)complex data in a compact way..." and then, with respect to graphics, he says, "...Graphs are not compact: if they are 'worth a thousand words,' we cannot easily describe or compare them. Graphs show data rather than summarize them." Hamilton's first chapter presents the exploratory tools and graphical tools, mean+variance -> normal distribution -> median+interquartile -> boxplot -> symmetry plots -> quantile plots -> quantile-quantile -> quantile normal Lawrence's theme is a restatement of Tukey's exploratory data analysis. For the quantile plot Lawrence places the fraction of data, 0 to 1, on the x axis, and quartiles on the y-axis. For the qq plots he plots observations on both the x-axis and y-axis. He notes that qq plots are for comparing two empirical distributions or for comparing an empirical distribution against a theoretical distribution. For qn, quantile-normal plots (aka quantile probability plots), he plots quantiles on the vertical axis "against the corresponding quantiles of a theoretical Gaussian (normal) distribution with the same mean and standard deviation). Cleveland refers to the goal of comparing to distributions, "usually to rank the categories according to how much each has of the variable being measured." His example plots male verbal SAT on one axis, female verbal SAT scores on the other. Next to that qq plot he also provides a Tukey mean- difference plot showing mean on the horizontal and difference on the vertical. Peter Sven Schreiber wrote: