In the light of this rather general problem we suggest a different approach that is not
based on prices6 but directly estimates effects from the consumption of one drug to the
consumption of the other. In fact, explaining the consumption of two drugs mutually
by each other is not completely new to the literature. One recent analysis (Bask &
Melkersson 2004), which is based on aggregate data from Sweden, uses consumption
levels of tobacco as an explanatory variable for contemporaneous alcohol consumption
and vice versa. Nevertheless, Bask & Melkersson (2004) still critically rely on price data
that serve as instrumental variables and the analysis still ultimately aims at estimating
cross-price effects. In contrast, the approach presented by the following section does
not rely on - presumably insufficient - price data altogether. Moreover, our work
contributes to the existing literature as, to our knowledge, it is the first econometric
application where German micro-data is used to analyze the interdependency in the
consumption of alcohol and tobacco. In Germany, prices of alcohol and tobacco display
only minimal variation over time as well as across regions. Therefore our approach
seems to be particularly well suited to the German case.
3 The Econometric Framework
3.1 A Statistical Guinea Pig approach
If we were biologists rather than economists and if we were interested in the behavior
of guinea pigs rather than human beings we would typically not consider estimating
demand functions with prices serving as explanatory variables.7 Nevertheless, we still
could examine whether our test animals consumed two kinds of feed as complements
or substitutes. In order to do this, we would treat them with certain doses of one
feed and measure the intake of the other. As a matter of course, in the context of
drug (ab)use it is not possible or, at least, it is highly questionable in ethical terms to
carry out such an experiment with human beings.8 Nevertheless, we can mimic such
an experiment by applying statistical procedures to survey data. Hence, we express
6Picone et al. (2004) do not exclusively rely on price data, since they employ a “smoking ban
index” as additional explanatory variable for both tobacco and alcohol consumption.
7Yet, one may consider using implicit prices - such as the effort that is necessary to collect certain
kinds of feed - instead of obviously inexistent market prices.
8 A clinical study which tried to address the research question of this paper at least had to exclude
non-smokers from the experiment, in order to avoid exposing them to a substance that is known to
be addictive and noxious.