Use the probability plot to compare the fits of common distributions in order to choose the best fitting distribution. If the data points fall along a relatively straight line on the probability plot, then you can conclude that it is reasonable to model your data using that distribution. Therefore, the best fitting distribution is the one in which the points most closely follow the fitted line.

The points on the plot are the estimated percentiles for the failures based on a nonparametric method. When you hold the pointer over a data point, Minitab displays the observed failure time and estimated cumulative probability.

The line is based on the fitted distribution. In this example, Weibull, lognormal, exponential, and normal are the fitted distributions. When you hold the pointer over the fitted line, Minitab displays a table of percentiles for various percents.

For the maximum likelihood estimation method (MLE), Minitab displays the Anderson-Darling (adj) statistic to assess the fit of each distribution.

When you use the least squares (LSXY) estimation method, Minitab displays the Pearson correlation coefficient, which is a positive number that can be no greater than 1. Higher correlation coefficient values indicate that the distribution provides a better fit.

###### Note

For more information on interpreting the Anderson-Darling (adj) statistic, go to "Goodness-of-fit".

## Interpretation

For the engine windings data, the probability plots show that the Weibull distribution and the lognormal distribution fit the data much better than the exponential distribution or the normal distribution. The lognormal distribution appears to fit these data best.