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

Table 3: Results for non-linear model extensions

additional instruments; see Table 3 for the estimation results concerning γ_a1 , γ_a2 , γ_c1 ,
and γc2 .

These results indicate that the additional insights gained from estimating non-linear
extensions to the basic model are rather limited. Tests for over-identifying restrictions
always reject the assumptions which are necessary to identify the non-linear model or,
at least, do not allow for accepting them. Moreover, the coefficients γa2 and γ_c2 often
turn out to be insignificant. Therefore, we stick to the original linear model.

6.2 Separate Models for Males and Females

Our analysis reveals pronounced gender-effects on the consumption of tobacco as well
as the consumption of alcohol, see Tables 1 and 2. In order to analyze whether gender
does matter not just for the level of consumption but also for the interdependence
in consumption, the model is separately estimated for males and females. Table 4
displays estimates for the structural coefficients γ ; see Tables 9 to 12 in Appendix B
for a comprehensive list of estimation results.

Though the LR-statistic²² strongly argues in favor of separate models for males

²²The comprehensive χ²(58)-statistic for the reduced form takes a value as high as 2978. If smoking

20

<<   <   19   20   21   22   23   24   25   >   >>