Abstract:Modified jackknife test was applied to examine the robustness of a topology-polarity correction model for estimating the sorption coefficient(Koc) of organic compounds. The test was performed by four methods:1) random deletion of a single chemical from the data set (100 times);2) deletion of a chemical with abnormal value (difference between predicted and measured values is larger than 0.9 log-units,27 outliers);3) deletion of 50 chemicals randomly selected from the data set (30 times);and 4) deletion of a set of chemicals belonging to the same class (17 classes).The results indicated that similar trends were observed with various jackknife tests.The coefficient of determination (R2) of the multivariate regression model showed relatively high robustness compared to other regression coefficients. The sixth-order chain molecular connectivity index (6Xch) was the least robust parameter in the model.Among the 17 polarity factors, the robustness of SO2, O, F, NH2 and SO were lower than the others.