SKIN ROUGHNESS ASSESSMENT 291 the value of use indicates the percentage change the product has achieved for each parameter." Values above 100% indicate increased skin roughness, while values below 100% relate to smoothed skin. The methylene blue roughness (MBR) values are plotted versus those for image analysis roughness (IAR) in Figure 1. This figure shows that there is an outlying point that on the one hand is well contained in the data range for the MB method, but on the other hand is well separated from the range of the IA results. The correlation of the data is negative and not significant. Though the outlying point is contained in the original data of Schrader et al. (2), they leave it out without comment for their regression analysis. Accepting, for the time being, the extreme value as an outlier, a "cleaned" set of data is obtained, summarized in Figure 2, which is equivalent to Figure 1 in reference 2. The parameters for the linear regression line (solid line in Figure 2) are in agreement with reference 2' slope: b = 1.70 (0.705) y-axis intercept: a = - 66.1 (10.27) correlation coefficient: r = 0.484 number of data points: n = 21 The values in brackets are the standard deviations of the parameters. Though the data points relate to 20 individual results and are actually mean roughness values, they will be treated as single values in what follows, since there is no information about, for example, their standard errors. The regression analysis is based on the assumption that the image analysis method provides an independent variable that exhibits an accuracy by 130 120 110 100 •0 80 ß ß ß ß ß ß 70 50 60 70 8'0 9'0 1(50 110 IAR, % Figure 1. Methylene blue roughness (MBR) vs image analysis roughness test results (IAR) for all data in Table I. Linear regression line ( ).

292 JOURNAL OF THE SOCIETY OF COSMETIC CHEMISTS 130 110 90-1 70 •0 i i i i i 90 92 % 96 98 100 102 IAR, % Figure 2. MBR vs IAR results for the "cleaned" data set. Linear regression line ( limits for the prediction of single MBR values (---). ) and 95% confidence far superior compared to the MB method. This assumption appears reasonable in view of the added digit in the IAR values as compared to the MBR values in reference 2 (see Table I). To test the significance of the regression, it is checked whether the true slope of the regression line [3 is significantly different from zero. This may be done by applying a t-test (4). For the test the statistical significance is set to the usual 95% level. A parameter t is calculated from the data given above according to t = b/s b = 2.41 (1) where s b is the standard deviation of the slope. t is checked against the relevant t-value of the Student distribution for a double-sided test and for DF = n - 2 = 19 degrees of freedom, which is t95%(2) ' 19 -- 2. 093. Since t is larger than this value, the hypothesis [3 = 0 has to be rejected on the 95% level. However, it is important to note that already an increase to a 98% level (t = 2. 539) leads to the acceptance of the hypothesis [3 = 0 and hence to a rejection of the assumption of correlation between MBR and IAR results. The correlation can hence be considered as being only just significant. This has severe consequences with respect to two types of important conclusions that may be drawn from the correlation. First, it must be asked, on the basis of the data in Figure 2, within which range an MBR result may be expected when conducting a test with a product for which the IAR result is known.