Tobacco Documents Online

BROWN & WILLIAMSON TOBACCO CORPORATION RESEARCH, DEVELOPMENT & ENGINEERING FILE NOTE TITLE: INTERIM REPORT - HOW MUCH DEVIATION FROM A MENTHOL TARGET CAN A PRODUCT HAVE BEFORE IT IS NOTICEABLE TO SMOKERS?/413 AUTHOR: P. L. Chen/W. H. Deines DATE: September 25, 1986 CC: M. L. Reynolds, T. F. Riehl, C. N. Kounnas, D. L. Gordon, A. McMurtrie, D. V. Cantrell, D. S. Roth, B. A. Reasor, J. H. Lauterbach SUMMARY Probit analyses were applied to Dupont sensory data from six paired attribute comparisons. The reference sample (current KOOL KS at 0.41% menthol) was paired with each of three higher menthol samples and three lower menthol samples. Among the six sensory attributes in this study, smokers more readily noticed changes in irritation and impact than they noticed changes in menthol taste, cooling effect, tobacco taste, or aftertaste. Ten percent of smokers* noticed a difference in irritation when % menthol was deviated from the reference cigarette at +0.06%/-0.02%; for impact, 10% of smokers noticed a difference when the deviation was +0.05%/-0.04% and the figures were +0.08%/-0.07% for menthol taste. For cooling effect (+0.10%), tobacco taste (+0.08%), and aftertaste (+0.07%), differences due to negative deviations were not well defined. These results for one month old product are graphically depicted in Figure 1. These comparative smoking test results appear to support our current menthol specification of target ± 10% relative (0.40% ± 0.04% for KOOL KS). The same holds for Dupont monadic smoking test results(1) (by 14 expert PRISM panelists) for these cigarettes when viewed in the usual manner for 95% confidence levels of difference; however, probit-type analysis of these monadic results, which will be reported later, suggests that for KOOL KS the menthol specification tolerance of target ± 10% might be too lenient. Therefore in addition to repeating the above comparative (60 KOOL KS smokers) and PRISM monadic smoking on these same cigarettes at three months age, we will include the same KOOL smokers doing PRISM type monadic rating. * Strictly speaking, it is 10% of the times smokers noticed the difference. That is because the N= 250 in our data-base representing 63 smokers x 4 reps, not 250 smokers. 620697171

INTRODUCTION Probit analysis was proposed as a statistical tool to aid in setting product specifications based on consumer sensitivity.[31 To "try-out" this approach, seven KOOL KS samples were made with different menthol levels: 0.60%, 0.52%, 0.48%, 0.41% (reference), 0.35%, 0.29%, and 0.23%. Six paired attribute comparisons were carried out at Dupont.t In each paired comparison, 63 KOOL KS smokers were asked to judge which cigarette had more of each of six attributes (irritation, impact, menthol taste, cooling effect, tobacco taste, and aftertaste). Each comparison was replicated four times. Sensory testing was conducted four weeks after sample manufacture. RESULTS OF PROBIT ANALYSIS For each attribute, the probit model was fitted separately to the three data points representing positive deviations in % menthol and to the three data points representing negative deviations. The necessity for separating the model fitting (one for positive deviations and one for negative deviations) was explained in a previous file note.[21 Data used for probit analyses are summarized and given in Appendix A. The results of probit analyses are reported in terms of Critical Difference (CD), and tabulated in Figure 2. A CDlo =+0.05%/-0.06%, for example, means that it takes a positive deviation of 0.05% (in menthol) or a negative deviation of 0.06% from the reference sample before 10% of smokers in this testing noticed the difference. DISCUSSION Test results merit the following discussion points. • In the comparative smoking test, changes in % menthol were noticed by smokers mainly through their perception of changes in irritation and impact, not necessarily through their perception of changes in menthol taste or cooling effect. • The results of these probit analyses are consistent with the same/different test results and paired attribute comparison results reported by B. A. Reasor.1l1 They are also consistent with monadic attribute rating results[l] in terms of degree of sample differentiation. However, the monadic results of 14 FF-M PRISM panelists showed sharper and more accurate perception on menthol taste and cooling effect than those 63 panelists in the paired attribute comparisons. This underlies the need for careful choice of attributes and demonstrates how choice of test method (comparative.vs. monadic) can influence the specifics of results. Same/different and monadic attribute rating tests were also conducted on these seven samples. All the test results have been reported by B. A. Reasor[i) (attached). - 2 - 620697172

• If not for some practical constraints, more samples (data points) could have been made and used in the probit analysis. The practical constraints in the case of % menthol are: (1) The difficulty of achieving cigarette samples which differ small increments of % menthol. in (2) At menthol levels appreciably higher than the "appropriate level" (e.g., 0.40% for KOOL KS), menthol perception becomes "overpowering"; this causes menthol carryover (lingering effects) which in comparative testing can cause perceptual confusion (e.g., samples being judged different in impact and irritation but not in menthol attributes). • Our other menthol brands should be examined similarly as time permits. Also, it must be noted that although two separate forms of panel testing (comparative and monadic) support the existing menthol guideline of target ;10%, full time "living" with the product might support a different guideline. • It is of interest to compare the results we obtained in this study with the results obtained in a RJR paper.[41 For the FF-M products, the RJR paper estimated that 10% of smokers noticed a difference in "menthol cooling" when % menthol was deviated from the reference cigarette at +0.27%/-0.27%. This study, on the other hand, estimated that 10% of smokers noticed a difference in "menthol taste" when the deviation was +0.08%/-0.07%, or a difference in "cooling effect" when the deviation was +0.10%/--. The values estimated in the RJR paper were about three times larger than the values estimated in this study. There were several differences between the two studies: 1. Experimentwise, the RJR study had its reference sample at 0.49%; this study had its reference sample at 0.41%. The RJR study had 15 panelists tested twice for each paired comparison, while this study had 63 panelists tested four times for each paired comparison. Also, the RJR study used "menthol cooling" as its only sensory attribute while this study used six sensory attributes. 2. For statistical analysis, both studies used probit analysis. However, the probit analysis was not properly carried out in the RJR study. A detailed discussion about this was given in a previous file note.[31 This is part of the reason that we used "Critical Difference" (CD) as our notation instead of the "Just Noticeable Difference" (JND) used in the RJR paper. -3- 620697173

Overall, this study has larger sample size and more sound statistical analysis; and more importantly, it has a better foresight by including irritation, impact, etc., in addition to menthol taste and cooling effect. Consequently, we believe that the results obtained in this study are more reliable and give better insight. • One final remark, the RJR study also encountered the "overpowering" phenomenon we discussed earlier. The attached plot (Figure 3) from the RJR paper is its probit model fitting for FF-M products. It can be seen (the four stars in the upper right portion) that the percentage of smokers who noticed the difference in cooling actually decreased when the deviation from the reference sample increased. fo, .e c.(, .., &-L P. L. C. W. H. D. PLC/klc 0520o REFERENCES 1. Reasor, B. A., "Dupont Results for Menthol Perception Testing", Internal Correspondence to W. H. Deines, Project 311, (8/14/86). 2. Chen, P. L., "Our Current Process Capability for In-Cigarette Menthol", B&W File Note, Project 413, (6/27/86). 3. Chen, P. L.,_"Probit Analysis - A Statistical Tool to Set Product Specifications Based on Consumer Sensitivity", B&W File Note, Project 413, (12/30/85). 4. Perfetti, T. A. and Gordin, H. H., "Just Noticeable Difference Studies of Mentholated Cigarette Products", Tobacco International (21): 57-66, (5/85). - 4 - 620697174

Figure 1 Deviations Noticeable to 10% of Smokers Deviation in % Menthol from the Ref. Sample +0.10% +0.08% • +0.06% +0.04% +0.02% ------- ------ Irrita tion------ Impa ct------ Menthol------ Cooling------ Tobacco------ Afte rtaste -0.02% Ta I I ste Effect Taste -0.04% - -0.06% - -0.08% - -0.10% -

Figure 2 Irritation Impact Menthol Taste Cooling Effect Tobacco Taste Aftertaste Critical Differences in Menthol CDlo CDzo CD ao +0.06% / -0.02% +0.11% / -0.05% +0.22% / -0.22% +0.05:C / -0.04% +0.09% / -0.08% +0.16% / -0.24% +0.08% / -0.07% --------------- --------------- +0.10% / ------ --------------- --------------- +0.08% / --------------- --------------- +0.07% / ------ +0.10% / ----- --------------- Remarks: 1. The CD's for some negative deviations are not reported since there was no clear effect pattern. 2. For menthol taste (both positive and negative deviations), cooling effect (positive only), tobacco taste (positive only), and aftertaste (positive only), probit model/common-sense expectation worked well when the menthol deviations were within ±0.12%. Beyond that, the percentage of smokers who noticed the difference actually decreased or leveled off. (See the last column in Figure A3,.A4, A5, and A6.) Because of this, the data points with the largest % menthol deviations (i.e., +0.19% or -0.18%) were excluded from the model fitting in these cases. Also, extrapolations to CD20/CD40 were avoided. 6 - 62069'71~I6

Figure 3 A Plot from the RJR Paper to Show the "Overpowering" Phenomenon hoe.o+ny co pM*~e - CO-" Flpvn 5. JHD for Pack MMttiol of FF-M Producys. .9-1 .6---~ .5~ .4-f .3-~ .2--a 0 -2 -1 0 1 2 3 Pack Mentt" (%) l - 7 - 620697177

APPENDIX A 620697178

Appendix A Summary of Data Used in Probit Analyses Figure Al - More Irritation % of % of Samples Deviation N Observation* Correct Responses True Detectiont 0.41% vs. 0.48% +0.07% 247 nl = 107 n2 = 140 56.7% 13.4% 0.41% vs. 0.52% +0.11% 248 nl = 104 n2 = 144 58.1% 16.1% 0.41% vs. 0.60% +0.19% 249 nl = 78 n2 = 171 68.7% 37.3% 0.35% vs. 0.41% -0.06% 246 nl = 94 n2 = 152 61.8% 23.6% 0.29% vs. 0.41% -0.12% 250 nl = 94 n2 = 156 62.4% 24.8% 0.23% vs. 0.41% -0.18% 249 nl = 74 n2 = 175 70.3% 40.6% * The higher the % menthol, the more irritation. t 2(% of correct responses) - 100%. Figure A2 - More Impact Samples Deviation N Observation* % of - Correct Responses % of True Detectiont 0.41% vs. 0.48% +0.07% 247 nl = 103 n2 = 144 58.3% 16.6% 0.41% vs. 0.52% +0.11% 248 nl = 94 n2 = 154 62.1% 24.2% 0.41% vs. 0.60% +0.19% 249 nl = 64 n2 = 185 74.3% 48.6% 0.35% vs. 0.41% -0.06% 246 nl = 104 n2 = 142 57.7% 15.4% 0.29% vs. 0.41% -0.12% 250 nl = 91 n2 = 159 63.6% 27.2% 0.23% vs. 0.41% -0.18% 249 nl = 83 n2 = 166 66.7% 33.3% * The higher the % menthol, the more impact. t 2(% of correct responses) - 100%. 0520o 62069'71'79

Appendix A Summary of Data Used in Probit Analyses (Continued) Figure A3 - More Menthol Taste Samples Deviation N Observation* % of Correct Responses % of True Detectiont 0.41% vs. 0.48% +0.07% 247 nl = 114 n2 = 133 53.8% 7.7% 0.41% vs. 0.52% +0.11% 248 nl = 106 n2 = 142 57.3% 14.5% 0.41% vs. 0.60% +0.19% 249 nl = 111 n2 = 138 55.4% 10.8% 0.35% vs. 0.41% -0.06% 246 nl = 111 n2 = 135 54.9% 9.8% 0.29% vs. 0.41% -0.12% 250 nl = 110 n2 = 140 56.0% 12.0% 0.23% vs. 0.41% -0.18% 249 nl = 115 n2 = 134 53.8% 7.6% * The higher the t 2(% of correct % menthol (within ±0.12%), the more menthol taste. responses) - 100%. Figure A4 - More Cooling Effect Samples Deviatio n N Observation* % of Correct Responses % of True Detectiont 0.41% vs. 0.48% +0.07% 247 nl = 120 n2 = 127 51.4% 2.8% 0.41% vs. 0.52% +0.11% 248 nl = 108 n2 = 140 56.5% 12.9% 0.41% vs. 0.60% +0.19% 249 nl = 129 n2 = 120 48.2% -3.6% 0.35% vs. 0.41% -0.06% 246 n1 = 126 n2 = 120 48.8% No 0.29% vs. 0.41% -0.12% 250 nl = 121 n2 = 129 51.6% Clear 0.23% vs. 0.41% -0.18% 249 nl = 136 n2 = 113 45.4% Effect * The higher the t 2(% of correct 05200 % menthol (up to +0.11%), the more cooling effect. responses) - 100%. 620697180