The Anderson-Darling (adjusted) statistics are goodness-of-fit measures associated with the probability plots for standardized and Cox-Snell residuals. Anderson-Darling (adjusted) statistics measure the area between the fitted line (based on the chosen model) and the nonparametric step function (based on the plot points). Precisely, the Anderson-Darling (adjusted) statistic is a squared distance that is weighted heavier in the tails of the distribution.

Compare the Anderson-Darling (adjusted) values for several different models to assess the appropriateness of the different models. Anderson-Darling (adjusted) values provide a relative measure of fit, by comparing the values from several models, rather than an absolute measure of fit.

Smaller Anderson-Darling (adjusted) values indicate that the fit of the probability plot is better. In other words, the assumption(s) the probability plot tests are more valid for the model with the lower Anderson-Darling (adjusted) value.

For the compressor data, only one set of Anderson-Darling (adjusted) values is provided; therefore goodness-of-fit cannot be assessed. To assess goodness-of-fit you would need to compare Anderson-Darling (adjusted) values for two or more models.