An alternative to traditional goodness-of-fit tests for discretely measured continuous data
Traditional goodness-of-fit tests such as the Kolmogorov-Smirnov and x2 tests are easily applied to data of the continuous or discrete type, respectively. Occasionally, however, the case arises when continuous data are recorded into discrete categories due to an imprecise measurement system. In this instance, the traditional goodness-of-fit tests may not be wholly applicable because of an unmanageable number of ties in the data, sparse contingency tables, or both; therefore, a flexible alternative to goodness-of-fit tests for discretely measured continuous data is presented. The proposed methodology bootstraps confidence intervals for the difference between selected percentiles of the empirical distribution functions of two samples. Application of the approach is illustrated with a comparison of loblolly pine (Pinus taeda L.) tree crown density distributions at the 10th, 25th, 50th, 75th, and 90th percentiles simultaneously.