West-East Convergence in the Prevalence of Cannabis Use: Socioeconomics or Culture?

In contrast to West Germany, illicit drugs were virtually absent in East Germany until 1990. However, after the collapse of the former GDR, East Germany was expected to encounter a sharp increase in substance abuse. By analyzing individual data, we find that East Germany largely caught up with West Germany's ever-growing prevalence of cannabis use within a single decade. We decompose the westeast difference in prevalence rates into an explained and an unexplained part using a modified Blinder-Oaxaca procedure. This decomposition suggests that the observed convergence is only weakly related to socioeconomic characteristics and therefore remains mainly unexplained. That is, West and East Germans seem to have become more alike per se. We conclude that both parts of the country have converged in terms of the culture of cannabis consumption. --


Introduction
In the former East German GDR the consumption of illicit drugs, such as cannabis, widespread in western societies was virtually absent (Reißig, 1991). 1 This can easily be explained by the isolation of the country from its neighbors and an extremely high level of surveillance by security forces within the country. 2 However, when the Berlin Wall fell in 1989 and the former GDR was subsequently integrated into the West German Federal Republic in 1990, East Germany was expected to experience a sharp increase in the prevalence of illicit drugs (Reißig, 1991) that would ultimately result in the convergence of drug consumption patterns in East and West Germany. In fact, although the prevalence of illicit drugs is still smaller in eastern Germany, in relative terms this gap had closed substantially by the year 2000. 3 Gaining more insight into this process of convergence is interesting for at least two different reasons. Firstly, the case of East Germany as a country becoming suddenly exposed to the problem of illicit drug use is exceptional. Thus, it is worth examining how the prevalence of substance abuse has developed in East Germany compared to the western part of the country, which has experienced drug abuse for a much longer time. Moreover, it is worth looking at the factors that shape the process of convergence between both regions. If convergence is related to certain socioeconomic factors, one might potentially identify strategies for preventing East German prevalence rates reaching West German levels. However, if convergencethat goes along with an increase in East German drug use figures -cannot be related to such factors it might simply represent an inevitable consequence of the country's reunification.
Secondly, in Germany the question of whether the two formerly separated parts of the country are developing a joint "cultural identity" has been intensely debated since 1990. Substance abuse might be a particularly well suited issue to address the question of genuine "cultural convergence" since, in contrast to various areas of public and private life, no political interventions took place to align drug use patterns in East Germany with those in the western part of the country. Thus, figuring out what shapes the process of convergence in prevalence rates might help to answer the question of whether both parts of the country are developing a joint "cultural identity".
More specifically, our analysis aims at disentangling two different matters that might be reflected by convergence in prevalence rates. First, living conditions in East and West have become more equal. This applies foremost to the labor market. A large share of the East German population has already encountered unemployment and job loss by now, while employment was guaranteed to all citizens of the GDR prior to 1990. Failure and disappointment related to the individual labor market performance is found to increase the probability of drug abuse by numerous empirical studies, e.g. Pudney (2004) and Hüsler et al. (2004). In addition, western Germany and eastern Germany may have converged on other socioeconomic characteristics, too, such as the average level of educational attainment, average income, and the marriage rate, which often are found to be correlated with the consumption of psychoactive substances, and finally the availability of illicit drugs.
The second possibility is that East Germans and West Germans may have simply become more similar per se, i.e. the culture of drug consumption may be what has converged since 1990. Statistically it is possible to relate the first argument to the distribution of individual socioeconomic characteristics in both regions, first and foremost to variables related to the labor market, but not the second argument. In social sciences, "cultural differences" are often implicitly defined as differences that go beyond any hard and observable socioeconomic factors -but nevertheless are obviously present. 4 One, therefore, may use the term "culture" as a label for what cannot be explained by socioeconomic characteristics and interpret unexplained convergence in drug consumption as a facet of cultural convergence.
In our empirical application, we decompose the west-east difference in the prevalence of cannabis use into one part that is explained by socioeconomic factors and another part that remains unexplained and, therefore, represents cultural differences in drug consumption. By repeating this decomposition for several years, we can determine to what extent the convergence in cannabis consumption is due to socioeconomic convergence on the one hand and cultural convergence on the other. For this exercise, we employ a modification to the -commonly used -decomposition technique that was originally introduced by Blinder (1973) and Oaxaca (1973). This approach is similar to the one of Burda and Schmidt (1997), who decompose wages in order to determine whether socioeconomic characteristics or unobserved human capital endowments shape the west-east wage differential and the process of wage convergence in reunified Germany.

The Data 2.1 Data Sources
This analysis uses data from the "Population Survey on the Consumption of Psychoactive Substances in Germany" 5 collected by IFT (Institute for Therapy Research) Munich; see Kraus and Augustin (2001) for a detailed description. To the author's knowledge, this data represents the most comprehensive source of information on substance abuse among adults in Germany. The Population Survey on the Consumption of Psychoactive Substances in Germany is not a panel but consists of eight separate cross-sections at the level of individual consumers. The surveys were carried out by mail at irregular intervals in the years 1980, 1986, 1990, 1992, 1995, 1997, 2000 and 2003. The sample size varies significantly from 4 455 in 1992 to 21 632 in 1990. The data comprises comprehensive information with respect to the consumption of various legal as well as illicit drugs. Additionally, some information on socioeconomic characteristics and sampling weights are provided along with attitudes towards several drug-related issues.
The most recent survey is not yet available for public scientific use. The surveys prior to 1990 concentrate solely on West Germany, while the one carried out in 1992 exclusively deals with the former GDR. Therefore, our analysis only considers the surveys carried out in 1990, 1995, 1997, and 2000 that are both available to us and cover both parts of the country. We do not consider individuals that do not have German citizenship, since foreigners are not included in the 1990 survey. We must also exclude individuals living in Berlin, since some of the surveys do not distinguish the eastern part of the city from its western part. The design of the survey has substantially changed over time. One of these changes concerns the age groups that were interviewed. While teens and young adults aged 12 to 39 years were interviewed in 1990, the surveys after 1992 focussed on adults aged 18 to 59. In order to analyze an age group that is consistent across waves, we only consider respondents aged 18 to 39. To check for the sensitivity of our results to the rule for selecting the sample, we also try an alternative approach. That is, instead of focussing on a specific age group we consider a certain age pseudo-cohort, more specifically we include those individuals that were born in the years from 1951 to 1972; see Appendix B.2 for the corresponding results. That is, the selected sample remains unchanged for the year 1990. Yet, for the subsequent waves the selected samples -on average -consist of older individuals.
Unfortunately, not only was the target population substantially modified, but so were the questionnaires. In particular, the number of questions was considerably reduced in 1995, eliminating almost all concerned with the respondents' family background. Moreover, several questions were substantially rephrased. Therefore, it is not possible to ensure consistency across waves for any variable. However, consistent information on drug use participation and the frequency of consumption is available. Nonetheless, the reliability of self-reported data on substance use is always a matter of concern. Underreporting of consumption levels and drug use participation appears to be a likely and obvious behavior, as substance abuse is socially disapproved and may even represent a criminal offense. Several studies have investigated the nature of misreporting and in particular the role survey conditions play for this problem, e.g. Mensch and Kandel (1988) and Hoyt and Chaloupka (1994). Self-administered surveys -as the ones at hand -seem to be less vulnerable to underreporting bias than telephone or face-to-face interviews. Yet, surveys that have a special focus on drug related questions seem to yield lower reported prevalence rates and lower consumption levels compared to surveys with a general scope. In order to validate survey results, more recently, analytical chemists have tried to www.economics-ejournal.org calculate drug consumption figures from the concentrations of certain drug residues found in sewage water, cf. Becker (2005) andZuccato et al. (2008). For the cases of Milan, London, and Lugano Zuccato et al. (2008) conclude that by and large these figures are in line with survey based official statistics. Yet, for some German cities the concentration of benzoylecgonin -a cocaine degradation product -seems to indicate much higher consumption levels of cocaine than suggested by survey results (Becker, 2005). Thus, underreporting might be an issue in our data. Yet, it seems to be hard to precisely pin down the importance of the bias that originates from underreporting. Moreover, since our analysis has its focus on comparing drug use in West and East Germany, underreporting represents a minor problem as long as reporting behavior does not systematically vary across both regions.

Consumption of Illicit Drugs
The data considers various illicit drugs. These are cannabis, speed and other amphetamines, LSD, mescaline, heroin, methadone, polamidone, codeine, opium and cocaine. The more recent waves also consider crack-cocaine, ecstasy, and "magic mushrooms". In addition, the questionnaires address substances that are not explicitly mentioned through the use of open questions. Our empirical analysis focusses on the consumption of cannabis. Cannabis represents, by far, the most prevalent illicit drug. Compared to this drug, the prevalence rate for any other substance mentioned above is quite low, i.e. only a small number of individuals report having consumed illicit drugs other than cannabis. Thus, sufficient data for carrying out a detailed decomposition analysis is available only for cannabis consumption. In fact, the prevalence rates for cannabis use and for illicit drug use in general almost perfectly coincide. That is, for only a minuscule number of drug users, cannabis does not belong to the list of consumed substances. For cannabis, as well as any other substance, the data comprises several measures of consumption, such as the age at the time of first use, lifetime prevalence, twelve-month prevalence, one-month prevalence, lifetime frequency of use, as well as twelve-and one-month frequency of use. For this analysis, we consider the twelvemonth prevalence as the most appropriate measure. In particular, we prefer this measure to the lifetime and one-month prevalence for the following reasons. On the one hand, the lifetime prevalence does not seem to be an appropriate basis for comparing the current consumption of illicit drugs in West and East Germany. By this measure, even those individuals that might have smoked a single joint 20 years ago are classified as drug consumers. Moreover, since it was hardly possible to have experience with illicit drugs in the former GDR, using the lifetime prevalence is likely to bias any west-east comparison. On the other hand, the one-month prevalence misses many drug users that consume illicit drugs on an irregular basis, which seems to be the case for the majority of consumers in the sample. We now examine the ordinary empirical twelve-month prevalence of cannabis consumption stratified by region and year. It is quite clear that the prevalence of cannabis use rose in eastern Germany during the 1990s; see Table 1. In 1990, i.e. immediately after the collapse of the communist system in the former GDR, less than one percent of the East German population aged 18 to 39 years had consumed cannabis. By the mid 1990s, this number rose to more than three percent. Finally, in 2000, more than seven percent of East Germans stated having used marihuana or hashish in the last twelve months prior to taking the survey. Yet, somewhat surprisingly, a similar increase in cannabis abuse had also taken place in western Germany; see Table 1. While the twelve-month prevalence was lower than five percent in 1990, it reached almost nine percent by the mid 1990s and exceeded ten percent by the year 2000. In fact, the west-east gap in cannabis use seems to be rather stable during the 1990s and appears to be most distinct by the middle of the decade rather than at its beginning; see Table 2. In fact, none of the observed changes in the level of west-east difference is statistically significant. Correspondingly, Augustin and Kraus (2001) conclude that the prevalence of substance abuse -if at all -has only marginally converged.
If, however, ratios of prevalence rates are compared instead of differences, this impression no longer holds. In contrast to Augustin and Kraus (2001), Perkonigg et al. (1998) argue that -in relative terms -the increase in prevalence rates is much more pronounced in East Germany than in West Germany. According to our data in 1990 West Germans were almost eight times more likely than East Germans to have consumed cannabis during the twelve months prior to taking the survey. This figure drops to the range between two and three in 1995 and 1997. In 2000 West Germans were just 1.4 times more likely to take hashish or marihuana than their East German counterparts; see Table 2. Moreover, the gap in the prevalence of cannabis use has closed in a statistically significant way in terms of ratios. Taking the logs of ratios leads to differences in log-prevalence rates; see Table 2. As a monotonic transformation, changes in log-means mirror the changes in ratios. We base our further analysis on differences in log-prevalence rates. We believe that focussing on differences in absolute prevalence rates overlooks the distinct process of convergence that is revealed through considering ratios of prevalence rates.
As an alternative approach, we base the analysis on the twelve-months frequency of use rather than twelve-months prevalence. Thus, this alternative approach does not only use the information on hash-and marihuana-smoking participation but also information on the intensity of consumption. 6 In our data, the frequency of consumption is interval-coded. The different questionnaires (cf. Appendix A) allow for the construction of seven cross-wave-consistent categories of cannabis use during the last twelve months; in detail (i) no consumption, (ii) consumed once, (iii) two to five times, (iv) six to nine times, (v) ten to 19 times, (vi) 20 to 59 times, and (vii) 60 times and more. Table 3 displays the distribution of these categories among actual cannabis users for both regions and the four considered years. To some degree west-east convergence may be seen in the frequency of use too. For instance, in 1990 the share of non-occasional users (more than than 5 times a year) is more than three times larger in West Germany than in the eastern part of the country. By the year 2000 any West-East deviation in this share has virtually disappeared. In particular the share of notorious consumers (more than 59 times a year), which in 1990 were completely absent from the East German sample, seems to have converged. Yet, due to the rather small number of cannabis consumers in the East German subsample, these findings are less reliable from a statistical point of view than the corresponding results for the prevalence rates.

Socioeconomic Characteristics
The data comprises information about several individual socioeconomic characteristics that may be related to the consumption of illicit drugs. In particular, these variables are: gender, age, number of biological children, months of unemployment during the last five years prior to taking the survey, marital status, living arrangements, current educational arrangements, labor market status, highest educational attainment, type of current or most recent job, income measured as income strata, and, finally, city/town population. A precondition for relating any west-east convergence in drug consumption to the labor market performance of individuals is that some variables related to the labor market display different trends in both parts of the country. We, therefore, have a closer look at the answer to the question "How many months have you been registered as unemployed in the last five years?"; see Table 4. While the average time spent in unemployment in West Germany remained rather stable in the 1990s, this figure increased dramatically in East Germany. At the beginning of the decade, East German respondents had experienced unemployment to a much lower extent than those from the western part of the country. Yet, this pattern has already reversed by the middle of the decade. In 1995, East Germans had experienced 3.5 times as many months in unemployment on average than West Germans did. This figure remained stable until the year 2000. Given that disappointment related to the individual labor market performance is, in fact, closely related to the consumption of illicit drugs, the convergence of this prevalence may be explained to some extent by the sharp increase in unemployment in East Germany during the early 1990s. In addition to this retrospective variable, we look at the current labor-market status. In 1990, the share of employed individuals in the sample was 13% higher for eastern Germany than for the western part of the country. Yet, this gap in employment rates had entirely closed by the year 2000. The share of currently registered unemployed is three times higher among East Germans than among West Germans in the 1990 sample. By the year 2000, this ratio had even reached the value of four. This gives additional support for the hypothesis that changes in relative prevalence rates might be correlated with changes in relative labor-market conditions.
The analysis would certainly benefit from controlling for the supply side at the local market for illicit drugs and -closely related -local drug prices. 7 In the survey data used, the only available variable that captures the supply of illicit substances is the answer to the questions "How easily can you acquire cannabis, speed, LSD, etc. within 24 hours?". However, only drug users typically know how to acquire drugs in the short term, while non-users typically do not. For this reason, this variable is a rather imprecise and subjective measure for the actual supply of illicit drugs and fails to capture supply independently from demand. 8 We, therefore, do not include this variable in the list of right-hand-side variables.
As an alternative approach, we consider official criminal statistics on cannabis seizures as a rough measure for regional cannabis supply. 9 This data is available at the level of Federal States (Bundesländer). Yet, this approach exhibits certain limitations too. First, for the year 1990 statistics on drug seizures do not exist for the majority of East German States. Second, cannabis seizures do not necessarily mirror the local drug supply. Large quantities of illicit substances are confiscated in transit -e.g. Hamburg frequently reports exceptionally large seizures because of major (illegal) drug turnover at the city's harbor -which are not meant for sale at the local drug market. Finally, figures on cannabis seizures still reflect both supply and demand for illicit drugs. Being aware of these shortcomings, as a robustness check we report results for a model variant that includes regional cannabis seizures per head and year at the right-hand-side; see Appendix B.3. Yet, we do not consider this specification as our preferred model.

The Decomposition Rule
In order to answer the question of whether west-east convergence in the consumption of cannabis is associated with socioeconomic characteristics or represents an unexplained cultural phenomenon, we use a modified Blinder (1973) and Oaxaca(1973) decomposition technique. This technique allows the fractionalization of differences in conditional means into one part that can be explained by socioeconomic characteristics and another that originates from deviations in the model parameters. The second part, therefore, is unexplained and represents a cultural gap in the sense discussed above. The Blinder (1973) andOaxaca (1973) decomposition is based on separate estimates of the conditional mean of a dependent variable y for two distinct subpopulations. In our application, the dependent variable of interest is the dummy indicating that a respondent has consumed cannabis during the twelve months prior to taking the survey. The subpopulations are West Germans and East Germans.
If the decomposition rule is generalized to non-linear models, cf. Fairlie (1999 and2003) and Bauer and Sinning (2008), it can be written: set of individuals living in West Germany, and I east the corresponding set for East Germany. The vector x it consists of individual socioeconomic characteristics and β represents a vector of parameters. ∆ expl t captures the component of differences in conditional means that is explained by socioeconomic characteristics. In other words, by ∆ expl t we measure the counterfactual difference in expected prevalence rates that would arise if in East Germany the right-hand-side variables had the same joint pattern of association with cannabis consumption as they actually do have in West Germany. 10 ∆ unex t captures the component in conditional means that is not explained by socioeconomic characteristics and, therefore, captures cultural differences between both parts of the country. I.e. by ∆ unex t we estimate the counterfactual difference in expected prevalence rates that would still arise even if in West Germany the explanatory variables had the same distribution as they actually do have in East Germany. 11 whereas the coefficients' estimateβ west t is obtained from a probit 12 regression using the relevant subsample. Analogously, this applies to E(y it |x it , β east t ). Estimates for the expectations unconditional on Equation (1) still applies. Ultimately, our focus is on the changes in the unexplained part (∆ unex t − ∆ unex t−1 ). If these changes prove to be negative and significant, one can conclude that the cultural gap in drug consumption has in fact diminished during the 1990s, and that both parts of the country have in fact become culturally more akin.
In order to judge changes in ∆ unex t as statistically significant, standard errors are required. Unfortunately, standard errors are rarely reported for the Blinder-Oaxaca decomposition (Jann, 2005). Jann (2005) derives analytical standard errors for the basic linear case. Yet, these are not applicable in our case, since we apply a generalized non-linear decomposition rule. This is why we report bootstrapped standard errors rather than analytical ones. In the bootstrap sampling weights are accounted for by duplicating each observation as many times as indicated by its weight and subsequently drawing from the expanded sample. 14

The Regression Model
As a starting point, we estimate a pooled probit model using all valid observations. We include all available variables that may serve as explanatory ones, i.e. age, age squared, number of biological children, number of months of unemployment, gender and marital status, as well as groups of indicators indicating (i) living arrangements, (ii) current education arrangements, (iii) labor market status, (iv) highest educational attainment, (v) type of current or most recent job, (vi) income measured as income strata, (vii) city/town population. In addition, a dummy indicating living in East Germany and time-dummies are included. Any of these variables or groups of variables are statistically significant.
It is important to emphasize, that a significant relationship must not be interpreted in terms of causality. For many variables, for instance being unemployed and being single, the direction of causality is far from obvious: On the one hand, one may argue that being frustrated with both career and private life leads to the abuse of psychoactive substances. On the other hand, individuals who have problems with illicit drugs are less likely to find either a job or a spouse. We, therefore, interpret any relation of left-hand-side and right-hand-side variables in terms of correlation rather than causality. Correspondingly, coefficient estimates must not be interpreted as marginal effects. But still, decomposing differences in prevalence rates into one component that is associated with differences in socioeconomic characteristics and another that is not associated with them is meaningful, even if this association does not represent causality. Since the pooled model does not argue in favor of any exclusion restrictions, the straightforward approach is to estimate the full model separately for all eight subsamples defined by region and period of time. Yet, because of the relatively small sample size for East Germany and its relatively low prevalence rate, the full model cannot be estimated using only the East German subsamples; cf. Table 1. Two different approaches may be followed in order to impose more structure on the data and to circumvent this problem. In the first approach, the size of the model is reduced until it is estimable for all eight subsamples. In the second, the full model is not estimated separately for all eight cells, but subsamples are pooled either across time or across region. Yet, pooling comes with cost. If pooling is across time, i.e. two regressions are run (one for each region), changes in culture are captured only by time-dummies. That is, the association of cannabis prevalence with the right-handside variables is assumed to be constant over time. This certainly limits any analysis that targets cultural change. If pooling is across regions instead, i.e. four regressions are run (one for each period), cultural differences between both parts of the country are exclusively captured by the differences in constants (α west t − α east t ). 15 In fact, in the linear -but not the non-linear -case the unexplained component ∆ unex t simply reduces to (α west t − α east t ). This means that, in the case of regional pooling, the decomposition is only a tool that helps to interpret estimated regional constants α and their changes over time. Yet, even if a Blinder-Oaxaca decomposition that is based on a pooled regression appears as a degenerated decomposition-exercise, it is still beneficial in interpreting the estimation results. 16 Following the first approach leads to a regression model with only five right-handside variables, three dummies indicating gender, employment status and living in a city along with age and the number of months of unemployment in the five years prior to taking the survey. For the West German subsamples, all these variables are highly significant for any year. In contrast, only age turns out to be a significant predictor in the probit regression using the East German subsamples. All other variables -if at all -are only occasionally significant in one out of four years; some of them even show reversing signs in different periods. Therefore, although it is technically feasible to base a decomposition-exercise on these regressions, the decomposition results critically rely on estimates β east t that apparently do not contain any information that is statistically firm. The finding that the decomposition results are extremely sensitive to changing the region of reference corroborates the sceptism about this approach. Running separate regressions for all eight cells, therefore, does not appear to be a promising strategy, and pooling might be the preferable approach. However, pooling across periods does not substantially improve matters. In a probit regression that uses a pooled East German subsample and the full set of right-15 Technically, this difference is estimated as a single coefficient attached to one regional dummy variable.
16 As a compromise of (i) a pooled regression with just a regional-dummy and (ii) two separate ones for both regions, i.e. a regression that contains a full set of interaction terms with the regional indicator, one may think of using a selected set of interaction terms. Alternatively, one may impose even more restrictions on the model by combining regional pooling with pooling across periods; i.e. a specific constant term α is estimated for any of the eight subsamples defined by region and time, yet all other coefficients from the vectors β are not allowed to vary neither across regions nor across periods.
www.economics-ejournal.org hand-side variables, estimated standard errors are still very large. This may be explained by the small number of cannabis consumers observed in East Germany. Therefore, pooling over periods still presents the problem that any decomposition result critically depends on estimates β east that are barely reliable. In order to capture the unexplained part of west-east deviations in the prevalence of cannabis abuse using a measure that is more reliably estimated than ( β west t − β east t ), we prefer to pool across regions, although the resulting decomposition represents a somehow degenerated Blinder-Oaxaca approach. In fact, the indicator for East Germany is always highly significant; see Table 5. Moreover, almost all regressors or groups of regressors, respectively, are clearly significant in at least two out of four regressions. Imposing more structure on the data via west-east pooling, therefore, seems to improve the reliability of estimates, though it implies the restrictive assumption that cultural differences are entirely due to differences in conditional con- sumption levels and cannot be related to regionally differing patterns of association of cannabis consumption with socioeconomic characteristics. 17 4 Decomposition Results

Results for the Basic Model Specification
In this section, we present decomposition results that are based on the preferred specification 18 , for which estimation results are reported in the previous section; see Table 5. Table 6 displays estimates for the mean difference in log-conditional means ∆ t as well as its components ∆ expl t and ∆ unex t . For the latter two variants are displayed, one with West Germany serving as reference category, i.e. log E x [E(y it |x it , β west t )|i ∈ I east , t] is used as counterfactual log-mean, and another variant where the reference region is reversed, i.e. log E x [E(y it |x it , β east t )|i ∈ I west , t] enters the decomposition-formula. Both variants do just marginally differ. This does not come as a surprise. Due to pooling across regions, ∆ unex t rests on the deviation of constants α west t and α east t alone. Therefore, in the case of a linear model, both variants of the decomposition coincide. In the case considered here, the deviation of both variants is solely due to non-linearity, i.e. calculating normal probabilities and taking logarithms.
West-east differences in log-mean conditional prevalence rates are significant individually for any year as well as jointly. In any year, the unexplained component exceeds the explained one by far in absolute terms. This result is statistically con-17 All three variants of the model, i.e. (i) separate regressions for all eight subsamples using a small set of regressors, (ii) pooling across periods using the full set of regressors, and (iii) pooling across regions using the full set of regressors, impose certain restrictions on the general model that neither pools subsamples nor excludes right-hand-side variables. Yet, since the general model is not identified, it is not possible to test which one of the restricted specifications is preferable.
18 That is (i) pooling across regions, (ii) individuals aged 18 to 39, (iii) dichotomous dependent variable, (iv) no drug seizures measure among the right-hand-side variables.
www.economics-ejournal.org Table 7: Changes in Decomposition-Components Transition It is interesting to note that the explained part changes its sign from 1997 to 2000. Therefore, in the most recent survey year on the basis of socioeconomic characteristics, one should expect higher prevalence rates in East Germany than in West Germany. This result apparently aligns with worsening labor-market conditions in East Germany compared to West Germany during the 1990s. Yet, the explained part is more than compensated by the unexplained part, which strongly argues for higher prevalence rates in West Germany. This may be taken as a west-east difference in culture, i.e. West Germans seem to have a greater affinity for cannabis than East Germans.

Ref. Region West
We now turn to the changes in the difference of log-mean conditional prevalence rates and the changes in its components. The corresponding figures are displayed in Table 7. For the transition from 1990 to 1995 and the one from 1997 to 2000, the west-east difference in log-mean conditional prevalence rates decreases, as is the case for the empirical difference in logs, cf. Table 2. This does not hold for the transition from 1995 to 1997. Yet, in statistical terms the rather small increase in ∆ t does not insignificantly differ from zero. Jointly, the changes in ∆ t as well as the changes in its components ∆ expl t and ∆ unex t are clearly significant. That is, both socioeconomic factors and culture seem to contribute to the convergence of prevalence rates in both parts of the country. For the transition from 1990 to 1995 the change in ∆ t is dominated by its unexplained part. The explained component turns out to be rather small and even does not significantly differ from zero. Thus, in the early 1990s the convergence in prevalence rates represents almost entirely a cultural phenomenon. For the transition from 1995 to 1997, the changes in any components of ∆ t are statistically insignificant, as is the case for ∆ t itself. Finally both, the explained and the unexplained component, seem to contribute equally to the distinct decrease of ∆ t from 1997 to 2000. Yet, only the former significantly differs from zero. In order to quantify the contribution of socioeconomic factors and culture to the overall convergence during the 1990s, we directly compare the years 2000 and 1990, see Table 8. Although both the explained and the unexplained component contribute significantly, the unexplained component accounts for roughly three-fourths in the overall convergence in log-mean conditional prevalence rates. 19 That is, the process of convergence remains largely unexplained.
In sum, the prevalence of cannabis seems to have converged in West and East Germany during the 1990s. The decomposition results suggest that this convergence can be related to socioeconomic characteristics only to a minor degree and therefore is mainly unexplained. One may interpret this unexplained convergence as a cultural one. In other words, West Germans and East Germans have become more alike per se in terms of cannabis abuse.

Robustness Checks
This subsection discusses decomposition results for several variations on the basic model, which have been mentioned in the previous sections. Firstly, we look at the results based on interval regression, see Appendix B.1 for comprehensive summary tables. Generally, in terms of significance as well as magnitudes these decomposition results are very similar to those obtained from the basic model specification. As one major deviation, if interval regression is used for estimating the model the change in prevalence rates is dominated by the unexplained component even for the transition from 1997 to the year 2000. This leads to an even more important role cultural convergence plays for the over-all change during the 1990s. That is, for the interval regression variant of the model cultural convergence accounts for more than 80% of total convergence while the explained share is less than one fifth, see Table 12.
Secondly, we examine the results based on the age pseudo-cohort born between 1951 and 1972, see Appendix B.2 for comprehensive summary tables. To a certain extent these results still resemble those for the basic specification. Yet, estimated changes in west-east deviations in prevalence rates are substantially smaller and many of these even become statistically insignificant. The relative importance of the explained and the unexplained component, however, remains largely unaffected; see Tables 15 and 16. Thus, west-east differences in cannabis consumption appear to be more persistent during the 1990s when the focus is on age cohorts rather than age groups. This does not come as surprise, as the former approach intentionally ignores the effect of younger individuals entering the population that have spent an increasing share of adolescence in reunified Germany. Yet, as our focus is on general convergence in regional cannabis prevalence rates rather than behavioral convergence within specific age cohorts, we regard the results based on age groups as the more relevant.
Finally, we look at the results of the model that includes cannabis seizures as an additional right-hand-side variable in order to account for regional variations in the supply of illicit drugs, see Appendix B.3 for comprehensive summary tables. Due to missing information, this model can only consider the years 1995, 1997, and 2000. Yet, for these years the inclusion of cannabis seizures to the list of regressors exerts virtually no effects on the decomposition results; see Table 18. This corresponds to the regression results that -with the exception of the year 1997 -do not display any significant correlation of individual cannabis participation and regional cannabis seizures; see Table 17. One might explain this finding by the limitations of the regional drug seizures measure used; cf. Section 2.3. Nevertheless, this result provides some reassurance for the unexplained component capturing a cultural phenomenon.

Conclusions
Since the reunification of Germany in 1990 there has been an intense debate on about whether both parts of the country will soon develop a common "cultural identity" or whether cultural differences that have developed through forty years of separation are deeply rooted and are likely to persist for decades. This paper contributes to this discussion with a special focus on the issue of cannabis abuse. It has been shown that prevalence rates of cannabis consumption have, in fact, converged in West and East Germany. More importantly, decomposition results suggest that this convergence can only weakly be related to socioeconomic characteristics. This result proves to be robust to several variations on the model. Convergence, therefore, represents first of all a cultural phenomenon. That is, at least with respect to cannabis abuse, West and East Germans did become more similar per se during the 1990s.
Convergence in substance abuse that goes along with an increase in prevalence rates seems to be a rather undesirable manifestation of cultural convergence. Moreover, convergence in cannabis consumption represents a rather small facet of overall cultural convergence. Nevertheless, as the consumption of illicit drugs is strongly related to "youth culture" it may serve as an especially illuminative indicator for a general process of convergence that might continue for the future since the younger age cohorts are more likely to develop a joint "cultural identity" that is not conditioned by the two different political and social systems that existed in Germany prior to 1990. Zuccato, E., C. Chiabrando, S. Castiglioni, R. Bagnati and R. Fanelli (2008). Estimating Community Drug Abuse by Wastewater Analysis, Environmental Health Perspectives 116(8), 1027-1032.

Appendix A The Questionnaire
This appendix comprises those questions from the questionnaires which the dependent variables are calculated from. The original and complete German-language questionnaires are available from "Central Archive for Empirical Social Research, University Cologne"; http://www.gesis.org/en/za. The subsequent section displays an English-language translation. The relevant questions address a large number of substances in a uniform way, yet for the sake of brevity, here we only display cannabis.

A.1 Questionnaire for the Year 1990
Question 72: See the following list of drugs and pharmaceuticals, i.e. substances whose consumption may result in intoxication or a sense of well-being. Please tick those substances you have tried (not in the context of a medical treatment). To those who have ever tried illicit drugs or sniffing agents:

Marihuana
Question 66: When did you last consume the following substances? Please encircle the appropriate number or fill in the number of years.  To those who have ever tried illicit drugs or sniffing agents: This appendix displays results for the model variant that uses information on the frequency of cannabis use und therefore employs interval-instead of probit regression for estimating the coefficients β. These estimates, see Table 9, are not directly comparable to those for the basic model specification. In order to allow for this, raw coefficient estimates have to be scaled by the factorσ −1 . In contrast, decomposition results, see Tables 10 through 12, can directly be compared.          Notes: Regressions are weighted by inverse sampling probabilities. ** and * indicate significance at the 0.05-and the 0.1-level.