57 



in a bivariate scatterplot generally decreases when one 

 stratifies on either variable. When the overall average tar, 

 based on the sales weighted average of all varieties tested by 

 the FTC was examined, both the WLS regression on time and the 

 Spearman rank correlation showed a statistically significant 

 decrease in tar over time. P-values for H,: There is no change 

 in tar level" over time are tabled as follows: 



