Tobacco and Alcohol: Complements or Substitutes? - A Statistical Guinea Pig Approach

Note: “ovi-test” indicates p-values for testing over-identifying restrictions.

and females, in qualitative terms the results are similar to those obtained from the
pooled model for either gender. For both men and women ^_c is negative, yet - as in the
pooled model - over-identification tests reject the identifying assumptions. In contrast,
γa takes positive values for both genders and its identification is not fundamentally
questioned for the males’ or for the females’ model by the relevant test-statistics.

The main difference to the results obtained from the split model, therefore, is a
quantitative one. While for males γa is of a substantially larger magnitude than in
the pooled model, the parameter takes a much smaller value for females and even
becomes insignificant. Therefore, the positive effect from smoking to drinking that is
found in this analysis is a males’ phenomenon while for females it seems to be of much
smaller magnitude and might even be non-existent. Thus, drinking habits seem to
differ between males and females even beyond the mere level of alcohol consumption.
This result might reflect gender-specific variation in preferences as well as differing
social conventions.

7 Conclusions

In this paper a new approach for analyzing the interdependence in the consumption
of alcohol and tobacco was proposed and applied to German survey data. We use
an alternative measure of complementarity which - in qualitative terms - is shown
to be equivalent to conventional Hicksian cross-price derivatives, yet it is not based
on the estimation of cross-price effects. In fact, the proposed instrumental variable
approach mimics an experimental study and therefore - in contrast to the main body
of the existing literature - does not rely on high-quality price data which often may

and drinking are treated separately, pooled models are rejected for either drug.

21

<<   <   20   21   22   23   24   25   26   >   >>