Influence of the Regression Technique on the Extrapolated Retention Values of Single-Point Adsorption Compounds on Silica Gel

Summary

The retention (R_M) values of nine one-point adsorption model compounds: diphenylamine, indol, 2-naphtol, 1-naphtol, 1-naphtylamine, 4-toluidine, carbazole, 4-chloraniline, and thymol were investigated on silica gel using six modifiers: acetone, dioxane, hexane, isopropanol, methylethylketone, ethyl acetate, and tetrahydrofurane (in hexane). These compounds showed small but visible curvilinearity of dependence of R_M vs. modifier concentration. This curvilinearity is very similar among the investigated compounds, so relative differences of extrapolated R_M are almost the same (strictly intercorrelated) regardless of the regression technique used. We have compared several robust and weighted regression methods and investigated their impact on extrapolated values. It can be concluded that one should primarily consider weighted regression with 1/x weights during retention extrapolation. It seems to be a better alternative than classical regression (better extrapolation) and also better than polynomial approaches (better stability).