Farris, Parry and Ailawadi ( 1992; hereafter denoted FPA) demonstrate
that bias can arise in a regression involving a composite dependent va
riable where a subset of components of the dependent variable are used
as explanatory factors. They correctly observe that the Jacobson and
Aaker ( 1985; hereafter denoted JA) model has explanatory factors that
are also components of the ROI dependent variable and, as such, is su
bject to ''composite variable bias.'' FPA note that another way of vie
wing composite variable bias is that the coefficients in the model ref
lect not their impact on the dependent variable but rather their impac
t on the dependent variable less the elements of the components includ
ed as explanatory factors. As such, additional effects (analogous to i
ndirect effects) may be present to the extent strategic factors influe
nce the included components. FPA conclude that such bias explains the
low estimate of the market share effect reported in JA. However, FPA's
attempt to replicate our analysis and assess composite variable bias
is flawed by a mistake in their analysis, i.e., their disaggregate mod
els do not follow from the JA aggregate specification. The purpose of
this note is to correctly assess the extent to which the JA estimate o
f the market share effect is affected by composite variable bias and t
o suggest approaches for modeling a composite dependent variable in th
e presence of unobservable factors. In particular, we (i) show that th
e disaggregate specifications of FPA do not follow from JA, (ii) look
at specifications not subject to composite variable bias to investigat
e the magnitude of the composite variable bias in JA, and (iii) provid
e a disaggregate modeling framework that controls for unobservable eff
ects.