expenditures were deflated by a media cost index (2004 = 100) computed from
annual changes in promotion and advertising costs by media type provided by Dairy
Management Inc., and (iii) bottled water was added to the soft drinks group to reflect
its current place of the second-largest non-alcoholic beverage category by volume.7
Some years of bottled-water data - consumption prior to 1976, price prior to 1984,
and advertising prior to 1985 - were not available (note its consumption per capita
was very low at 1.6 gallons in 1976 and its advertising expenditures were only $12
million in 1985, compared with its counterparts of 23.2 gallons and $116 million in
2004) and therefore interpolated linearly using data from their most adjacent years.
Equations were estimated using the PROC MODEL procedure in SAS
version 9.13, and as a system using Seemingly Unrelated Regressions to account for
contemporaneous correlation among individual equation errors (Griffiths, Hill, and
Judge 1993, p. 551).8 In the case of the Rotterdam and AIDS models one equation
(juices) was dropped to avoid singularity in the variance-covariance matrix. As
indicated, the Rotterdam and AIDS models were estimated with homogeneity and
symmetry imposed on both prices and advertising expenditures (Selvanathan 1989),
and adding-up was used to recover the coefficients from the omitted equation.
Results and Marketing Implications
Table 3 reports the parameters estimates for models F - J. Estimation results are
satisfactory in the sense that the adjusted R2s range from 0.83 to 0.99 in the AIDS
model to between 0.38 and 0.53 in the Rotterdam model. The majority of the
Durbin-Watson statistics center around two with some falling into the inconclusive
13