NEAR IR SURFACTANT ANALYSIS 451 Table V Linear Regression Data for NlRA Prediction of SLS Constituent nm Ko .. Kn r SEC F Active Detergent 197.70 0.95 0.29 164 1100 1715.24 1646 -2756.03 222O 8O9.O4 Solids - 269.35 0.99 0.17 1465 1184 1162.66 1450 - 110.73 Benzoic Acid - 33.18 0.90 0.03 61 1366 -75.38 1632 -538.22 1674 561.68 2248 58.68 pH 175.60 0.87 0.09 46 1366 73O.34 1394 - 2O4.18 1842 -748.45 1912 67.31 cosity, and pH for the ammonium lauryl sulfate and sodium lauryl sulfate samples (2 replications per sample) used in this study are found in Table III. The multiple linear regression statistics at selected wavelengths are listed in Tables IV and V. Figures 3 and 4 are examples of the linearity achievable by NIRA in constituent predic- tion versus the actual or primary analyses values as calculated using the multiple linear 30.00' 29.50' 29.00' i,-i m 28.50' 28.00- + 4. 4. + 4. 4. 4. + 4. 4. 4. 4. 4. 4. 4. •, + + ,I,+ 4. 27.50- 27.00 27.00 27.50 28.00 28.50 29.00 29.50 ACTUAL Figure 3. Scatter diagram •restimation ofactive detergentcontentin ALS by NIRA. :30. O0

452 JOURNAL OF THE SOCIETY OF COSMETIC CHEMISTS 33.00' 32.50' 32.00' o 31 50' M * o 31.00, 30.50, 30.00, 29.50' 29.00' 28.5C • 28.50 29.00 29.50 30.00 30.50 31.00 31•50 32.00 32.50 ACTUAL Figure 4. Scatter diagram •restimation ofsolidscontentin SLS by NLRA. ß ß ß ß ß ß ß 33. O0 equations listed in Tables IV and V. The results indicate that for SLS all constituents investigated are more than adequately predicted by near infrared reflectance analysis. The standard error of calibration for active detergent, solids, benzoic acid, and pH is at least as good as or better than the standard deviation obtained by primary methods. The lesser success with ALS when comparing correlation coefficients for constituents (active detergent, solids, benzoic acid, viscosity, and pH) is suspected to result from several sources. The Technicon liquid-sampling drawer is not adequate for use with highly viscous samples such as ALS, although the high water content of this type of sample requires the temperature and pressure stability available only through this de- vice. Until modifications are made, this problem cannot be solved. The narrow range of samples tested yields equations that cannot adequately predict outliers. More rejected samples or altered samples dried by microwave as was done with SLS must be included in the next data set pending availability of a viscous sampling cell. The pH range was much too narrow for acceptable calibration. Ideally, the standard deviation of the range should be at least 5 times greater than the acceptable standard error of calibration. The viscosity calibration set for ALS consisted of an adequate range, yet resulted in statisti- cally inadequate prediction equations. It is not certain whether the viscosity taken at 80øF over the eight-month collection period correlates linearly with the 45øC operating temperature of the near infrared liquid drawer. There was insufficient sample remaining to remeasure the viscosity under more controlled conditions. The wavelengths listed in Tables IV and V are the result of the multiple linear regres- sion with no constraints. Ideally, the wavelengths chosen by the computer can be as- signed to covalent bonds of the constituent being quantitated. A comparison was there- fore made of the near infrared spectrum of the neat individual constituent. Figures 5