and Picone et al. (2004).4 Irrespective of the level of aggregation and the country
considered, most of these studies find cross-price effects being negative and therefore
conclude that alcohol and tobacco are complements. As the only exception, Goel &
Morey (1995) find positive and significant cross-price elasticities.
Despite its rigorous derivation from micro-economic theory, the well-established
cross-price approach, however, suffers from severe shortcomings when applied to survey
data at the level of individual consumers. More specifically, for the identification
of cross-price effects this approach critically relies on high-quality price data that
is both exogenous and displays sufficient variation. Unfortunately, prices in general
are not consumer specific. That is, the same price should apply to any individual
consumer in the same market. Therefore, such analyses typically have to rely solely on
price-variation across periods (e.g. Jimenez & Labeaga (1994)) and/or across regions
(e.g. Zhao & Harris (2004) or Picone et al. (2004)), though price-variation often seems
to be rather limited.
Yet, consumption patterns are likely to vary across periods and regions for other rea-
sons than different prices alone. Even worse, price-differences might even reflect vary-
ing consumption patterns rather than the other way round. Quite regularly the lack
of individual specific price data does not allow for or, at least, seriously hampers con-
trolling for period- and region-specific heterogeneity.5 Hence, it seems to be strongly
disputable whether the corresponding price-coefficients capture true price effects but
not time or regional effects. Jimenez & Labeaga (1994: 235) quite tellingly exemplify
this problem by characterizing the price variables employed in their analysis “as a
group of quarterly dummies with restriction”. Others (e.g. Picone et al. 2004: 1068)
complain about prices and time effects being highly collinear, almost equally well re-
vealing the problem of price data being barely sufficient for identifying substitutability
or complementarity.
4Chaloupka & Laixuthai (1997), DiNardo & Lemieux (2001), and Williams et al. (2001), address
the interdependency of the consumption of alcohol and drugs others than tobacco, for instance,
marijuana. Moreover, several related papers do not use prices as explanatory variables and, therefore,
are concerned with correlation of drinking and smoking rather than interdependency, e.g. Su & Yen
(2000), Lee & Abdel-Gahny (2004), and Yen (2005).
5If price variation is either exclusively across time or exclusively across regions, corresponding
sets of period- or region-specific dummies obviously are not identified. If, instead, prices vary across
time and regions, in theory two sets of dummies could be included along with prices. In practice,
however, this will often not be feasible because of severe collinearity. Moreover, random effects is just
a feigned alternative to time- or regional dummies, since the implicit assumption that period- and
region-specific effects are uncorrelated with period- and region-specific prices is rather implausible.