the reduced form and the structural model, see McFadden (1999: 8). In addition,
we test whether the different estimates for γa and γc that can be calculated from
the reduced form model using coefficients attached to different instrumental-variables
do significantly differ. These tests can be regarded as quasi-tests for the validity of
over-identifying restrictions.
It is important to note that in a two-step approach not var(εa) but instead
var(γaυc + εa) = var(υa) is estimated as second-step regression variance. Clearly, this
analogously applies to var(εc). As long as we are exclusively interested in the model
coefficients this is of no relevance to our analysis. Yet, if we were interested in marginal
effects on actual, not latent, consumption an approach would be required that would
allow for the identification of var(εa), cf. Smith & Blundell (1986).
4TheData
4.1 Data Sources
This analysis uses data from the “Population Survey on the Consumption of Psy-
choactive Substances in Germany”14 collected by IFT15 Munich; see Kraus & Augustin
(2001) for a detailed description. The data is not a panel but consists of seven sepa-
rate cross sections at the level of individual consumers. Additionally, five significantly
smaller supplementary surveys were conducted primarily to address sampling issues.
The (regular) surveys were carried out by mail at irregular intervals in the years 1980,
1986, 1990, 1992, 1995, 1997, and 2000. The sample size varies significantly from
4455 in 1992 to 21632 in 1990. While the first two surveys concentrate solely on West
Germany the one carried out in 1992 exclusively deals with the former East German
GDR. All other waves cover Germany as a whole. Until 1992 only German citizens
but no immigrants without the German citizenship were interviewed. Later on, the
whole German speaking population was included in the survey. The data comprises
comprehensive information with respect to various legal as well as illicit drugs so also
on prevalence, frequency and intensity of consumption, consumption habits and age
at first use. Additionally, detailed information on socioeconomic characteristics is
provided along with attitudes towards several drug-related issues.
Unfortunately, the questionnaire and the study’s target population have changed
14Bundesstudie llReprasentativerhebung zum Gebrauch psychoaktiver Substanzen in Deutschland”
15Institute for Therapy Research (Institut für Therapieforschung)
11