54 



Analyses Performed: 



1. Analyses of the Trend of Nicotine Level on Year: 



The nicotine data provided by the FTC consisted of four 

 vectors: 1) Nic(i,j) - mean nicotine level of cigarettes tested 

 in year i, teir category j; 2) SDN(i,j) = std. dev. of Nicotine 

 in year i, tar category j; 3) N(i,j) - number of cigarettes 

 tested in year i, tar category j ; 4) S(i,j) = number of 

 cigarettes sold in year i, tar category j. In this data, the 

 mean and the std. dev. were computed, according to Joe 

 Mulholland, by taking sales weighted means and sd's of all of the 

 varieties tested within a given year and tar category. 



We did six analyses of the trend of Nicotine Level on Year, 

 from 1982 through 1991. These were as follows: 



1) weighted least squares regression of nicotine, Nic(i,j) 

 on year i, stratified by tar category j - (high, low, and 

 ultra-low) . The weights were proportional to reciprocals of 

 the variances of the sample means = l/VarN(i,j) = 

 N(i,j)/SDN(i,j)-2. 



2) weighted least squares regression of average nicotine on 

 year, ignoring tar category. In this regression, the 

 dependent variable was the sales weighted average of 

 nicotine of all varieties tested in year i. This was 

 computed from the FTC data as Nic(i) - 2{S(i, j) *Nic(i, j) ) 

 /i:S(i,j) where the sums were over j = ultra-low to high. 

 The weights in this regression were again proportional to 

 reciprocals of variances of the Nic(i) . Specifically, the 

 weights were w(i) « I{S(i,j)''2) / Z{VarN(i, j) *S(i, j) ^2 ) = 

 Z{S(i,j)-2) / E{S(i,j)^2 * SDN(i,j)-2 / N(i,j)). In all 

 cases, the suns are over j -ultra to high. 



3) ordinary (unweighted) least squares regression of 

 nicotine on year, stratified by tar category. (This was the 

 same as analysis 1 except the weights were now ■ l/io.) 



4) ordinary least squares regression of average nicotine on 

 year, ignoring tar category. (This was the same as 

 analysis 2 except the weights were ■ 1/10.) 



