Results

Table 3.2: Descriptive Statistics of Coffee Shop Municipalities in 2012

Note: ∗p <0.1,∗ ∗p <0.05,∗ ∗ ∗p <0.01.

Table 3.2 presents the descriptive statistics (mean, mean differences, and standard errors) for all-, treated, and control coffee shop municipalities in the post-treatment period of 2012.

Unlike the descriptive statistics presented in Table 3.1, the policy had an immediate impact on the number of registered soft drug crimes and the number of arrested suspects in connection with soft drug crimes. In comparison with the results presented in Table 3.1, the number of registered soft drug crimes increased by 11% and the arrest rate for soft drug crimes increased by 16% after the policy was implemented.

These findings corroborate earlier reports by Maalste and Hebbens (2012) and van Ooyen et al. (2014). Even though there was a insignificant increase in the number of hard drug crimes and arrests, the results reported by Table 3.2 provides an indication that the policy did not cause an increase after its implementation in 2012. Though there are some non-significant differences between the treated- and control coffee shop municipality characteristics, these are not reported as statistically different indicating that coffee shop municipalities do not differ substantially before and after the implementation of the policy.

Table 3.3: Effect of the Soft Drug Sales Restriction on the Drug Crime Rate of Southern coffee shop Municipalities (PSM-DID)

Note: ∗p <0.1,∗ ∗p <0.05,∗ ∗ ∗p <0.01.. Clustered standard errors are grouped at municipal level and reported in parenthesis. Treatment effect of the policy is captured by Equation 3.1.

Table 3.3 estimates the treatment effect of the policy on the municipal soft- and hard drug crimes using the difference-in-differences estimator. I included a time trend for purpose of controlling for exogenous increases in the outcome variable. Columns 1, 2, and 3 reports the municipal soft drug crime rate as the outcome variable and controlling for municipality-fixed effects, time-fixed effects, and the addition of a linear time trend respectively

Columns 1, 2, and 3 report that there is a statistical significant negative difference between Northern- and Southern coffee shop municipalities indicating that registered soft drugs crimes were lower in 2014 for Southern coffee shop municipalities. Therefore, in terms of the soft drug crime rate, the policy had a beneficial impact by reducing the number of soft drug crime.

Columns 4, 5, and 6 present the hard drug crime rate with municipality-fixed effects, time-fixed effects, and with a linear time trends respectively and reports no statistical significance indicating that the implementation of the drug policy had no significant effect on the number of drug crimes committed associated with hard drugs. Prior studies such as van Ooyen-Houben et al. (2014) who studied the effect of the policy on drug nuisance claimed that hard drugs within the Netherlands are not popular except for XTC. Other reasons cited were the availability of legal marijuana through coffee shops which reduces the inclination for soft drug users to visit illegal dealers, strict enforcement of the coffee shop policy to ensure that they do not sell hard drugs to the possible users, and strict enforcement of the police by minimizing the presence of illegal dealers. Therefore, the results presented in Table 3.3 indicates is that the policy did not change the number of hard drug crimes committed within Southern coffee shop municipalities when compared to Northern coffee shop municipalities.

Even though the initial results presented in Table 3.3 shows that the policy had a beneficial impact, it is important to examine whether the difference-in-differences estimator satisfies the parallel trend assumption as outlined earlier in this chapter. To check for the parallel trend, I apply Equation 3.2 from Autor (2003) through yearly interaction terms between the treatment (Di) and time dummy (dt). Table 3.4 presents the yearly treatment effect of the policy on the municipal drug crimes within coffee shop municipalities. The difference-in-differences estimator is unbiased when the pre-treatment years of 2010 and 2011 are reported as insignificant.

Table 3.4: Drug Crime Rate of coffee shop Municipalities Before and After the Implementation of the Sales Restriction (DID)

Note: ∗p <0.1,∗ ∗p <0.05,∗ ∗ ∗p <0.01.. Clustered standard errors are grouped at municipal level and reported in parenthesis. Treatment effect of the policy is captured by Equation 3.2

Columns 1, 2, and 3 presents the municipal soft drug crime rate as the outcome variable controlling for municipality fixed effects, time fixed effects, and a linear time trend. The results presented in Table 3.3 report significance in the pre-treatment period in the year 2010 and 2011, which biases the difference-in-differences estimator. Another observation is that the impact of the policy was not statistically significant during the year of implementation. This would indicate that the policy has a lagged time effect.

Columns 4, 5, and 6 presents the municipal hard drug crime rate as the outcome variable controlling for municipality fixed effects, time fixed effects, and a linear time trend. Unlike the pre-treatment significance reported for Columns 1, 2, and 3 for the municipal soft drug crime rate, the policy reports no significance for the hard drug crime rate in the pre-treatment period indicating that the results in Table 3.3 are credible for the municipal hard drug crime rate.

As seen for the municipal soft drug crime rate, there might have been a slight rise in the number of registered hard drug crime rate when comparing the change in the coefficient’s magnitude between 2011 and 2012. Prior studies mentioned that the closure of coffee shops to drug tourists had an immediate effect in creating an illicit drug market within the Southern coffee shop municipalities.

Therefore, as the previous sales channel for drug tourists was closed off, it is a possibility that dealers became more active than before the implementation of the policy as coffee shop customers turned to the illicit drug market.¹⁸¹⁹

The policy also resulted in the coefficient’s magnitude in 2014 to lower when compared to 2010 indicating that the policy had an effect though is not reported as statistically significant.

18According to prior research, usage of hard drugs within the Netherlands is relatively low due to the legal market of cannabis which separates the users from contact with hard drugs (MacCoun, 2011; van Laar et al., (2016)

19In research done by Maalste and Hebben (2012), coffee shop customers cited that the new regulations made it cumbersome to visit the coffee shops for marijuana and would instead turn to the illicit drug market due to ease of service, good quality, and a cheaper price than coffee shops.

The results presented in Table 3.4 indicates the necessity to include propensity score matching and match on the basis of observable municipality characteristics to solve the issue of the pre-treatment difference between Northern- and Southern coffee shop municipalities.

I use propensity score matching based on observable municipal characteristics to create a control group of Northern coffee shop municipalities that are similar in comparison with the treatment group of Southern coffee shop municipalities. The selected outcome variable is the number of registered soft drug crimes within municipalities.

I use the number of residents with employment, number of residents receiving poverty assistance, number of students in higher education, the median household income of households registered in the municipality, the total house value within the municipality, the density of hospitality firms, and the percentage change of suspects between 2009 and 2011 as independent variables for the propensity score specification.

Several matching algorithms such as pair matching with and without replacement, kernel match-ing, and nearest-neighbor matching up to 2, 4, and 6 neighbors with a caliper size of 0.04 and a bandwidth size of 0.06 for kernel matching were used to determine the robustness of the PSM-DID model. Nearest-neighbor matching up to 4 neighbors with a caliper of 0.04 provided the best matching result and are thus presented in this section.

Table 3.5: Propensity score balancing test of coffee shop municipalities

Note:∗p <0.1,∗∗p <0.05,∗∗∗p <0.01.Registered soft drug crimes is used as the dependable variable in the propensity score specification. Nearest-neighbor matching algorithm up to 4 neighbors with a caliper size of 0.04.

Matching results showed similar results indicating robustness. Table 3.5 shows the matching result of the nearest-neighbor matching with up to 4 neighbors of which the estimation results are based on. Results indicate that the matching procedure produced a control group that satisfies the requirements as the majority of listed variables are within the threshold of 5%. In addition, the t-tests on the bias difference as seen in Columnp >|t|reports no significant difference between the treatment- and control group. The matched sample for the PSM-DID model contains 16 treated coffee shop municipalities of the original 20 and 32 control coffee shop municipalities of the original 83.

Table 3.6: Rosenbaum Bounds Test - Nearest-Neighbor Matching (4)

Note: Soft drug crimes is the dependent variable. Nearest-neighbor matching algorithm up to 4 neighbors with a caliper size of 0.04.

Recent procedure that has become more popular with the PSM model is to check the sensitivity of the propensity score specification towards possible unobserved influences through the Rosenbaum bounds test (Diprete and Gangl, 2004). The Rosenbaum bounds procedure tests for the sensitivity by introducing an unobserved variable into the specification to check the whether the probability of treatment assignment can altered.

One of the assumptions for the PSM method is that observations would have the same proba-bility of receiving treatment based on the set of observed characteristics. However, in the case of a strong unobservable influence, two observations with similar observed characteristics might receive two difference probabilities of receiving treatment which would undermine the selection process and thus potentially create a biased counter-factual group.

As the value of Γ increases in the first column by 0.1, the impact of the unobserved variable becomes stronger. Once the Γ reaches a value where either the significance on the upper- or lower bound becomes insignificant (5% significance level), the propensity score specification no longer satisfies the assumption that observations with similar observed characteristics receive an equal probability of receiving treatment. Existing literature on the Rosenbaum bounds argue that a Γ value of 2.0 is considered to significantly insensitive towards the influences of an unobserved factor influencing the propensity score specification.

Table 3.6 shows the sensitivity of the propensity score specification to unobserved influences through the Rosenbaum Bounds Test. As explained in Section 3.3, a Γ value of 1.9 (.047) indicates that the specification can be considered to be insensitive to the possibility of unobserved influences.

Having passes the initial balancing test on observed characteristics and tested for bias sensitivity, the PSM procedure can be considered successful.

Table 3.7: Treatment effect of the 2012 Drug Policy on the Municipal Drug Crime Rate within Southern coffee shop Municipalities (PSM-DID)

Note: ∗p <0.1,∗ ∗p <0.05,∗ ∗ ∗p <0.01.. Clustered standard errors are grouped at municipal level and reported in parenthesis.

Table 3.7 presents the estimates the treatment effect of the policy on the municipal drug crime rates on coffee shop municipalities using the propensity score matching with difference-in-differences. As with Table 3.3, I divide the municipal drug crime rate into soft- and hard drug crime rate.

Columns 1, 2, and 3 presents the municipal soft drug crime rate of coffee shop municipalities as the outcome variable controlling for municipality fixed effects, time fixed effects, and a linear time trend. Results indicate that the policy did not have a statistical significant effect on the municipal soft drug crime rate. Comparing the results between Table 3.3 and Table 3.7 shows that the difference in the coefficient’s magnitude concerning the number of registered soft drug crimes is significantly larger but that the results are no longer considered to be statistically significant.

The results presented in Columns 1, 2, and 3 indicate that the policy did not significant impact on the soft drug crime rate after its implementation. One issue that could explain its insignificance is due to having only two post-treatment years. Additionally, prior research indicates that the response to restricting the sales of marijuana to municipal residents only had an adverse effect on the number of committed drug crimes. Van-Ooyen-Houben et al. (2014) acknowledges that the number of soft drug crimes went up in 2012 as a direct response.

The policy did achieve one of its primary objectives by deterring drug tourists away from Southern coffee shops. However, prior qualitative studies cite that the increased bureaucracy for previous coffee shop clientele has caused a portion of them to divert from the legal drug market in the form of coffee shops to the illicit drug market such as street dealers citing benefits such as ease of service and equal or better price-quality ratio (Maalste and Hebben, 2012).

Columns 4, 5, and 6 presents the municipal hard drug crime rate as the outcome variable controlling for municipality fixed effects, time fixed effects, and a linear time trend. Even after propensity score matching, results indicate that the policy had no significant effect on the number of registered hard drug crimes committed within Southern coffee shop municipalities.

The results reported in Columns 4, 5, and 6 can be explained due to two potential reasons. First is the fact that hard drugs are not popular within the Netherlands except for the drug referred to as XTC, a popular among certain youths within the Netherlands. Existing literature establishes that is true because the Netherlands offers legal recreational marijuana to isolate soft- and hard drug users by separating the legal and illicit drug market as part of the harm reduction model. While there was a large influx of previous coffee shop customers that substituted marijuana from coffee shops with those from the illicit drug market, it is unlikely that hard drugs would become a more attractive product within a short time period. In addition, unlike the decriminalized approach to soft drugs, hard drugs are still illegal and the consequences for the ownership and consumption of hard drugs can be severe.

Second could be attributed to geographical location. As the Southern coffee shop municipalities are located within a border region, it might that the profile of the stereotypical drug tourist has little to no interest in hard drugs. Therefore, it is possible that most of the hard drugs can be found

within the tourist municipalities within the Northern provinces such as Amsterdam, Rotterdam, and Den Haag as these municipalities are tourist hotspots. However, without proper data on this issue, this is merely an assertion.

In order for the results from Table 3.7 to be considered credible, the propensity score matching with difference-in-differences estimator has to satisfy the parallel trend assumption in the pre-treatment period. Table 3.8 presents the yearly interaction terms for the municipal soft- and hard drug crime rate using Equation 3.2 to check for possible significance in the years prior to the drug policy.

Table 3.8: Drug Crime Rate of coffee shop Municipalities Before and After the Implementation of the Sales Restriction (PSM-DID)

Note:∗p <0.1,∗∗p <0.05,∗∗∗p <0.01.. Clustered standard errors are grouped at municipal level and reported in parenthesis. Outcome captured by the interaction term (Dixdt) is presented in Equation 3.2.

Table 3.8 presents the yearly treatment effect of the 2012 drug policy on the municipal drug crime rate of Southern coffee shop municipalities using difference-in-differences with propensity score matching. Columns 1, 2, and 3 presents the municipal soft drug crime rate as the outcome variable controlling for municipality fixed effects, time fixed effects, and a linear time trend.

In Table 3.4, the results from the difference-in-differences estimator were not credible due to the significant difference in the pre-treatment period. According to the yearly interaction term of 2010 and 2011, the propensity score matching procedure was able to capture the observable characteristics that caused the significant difference between the Northern- and Southern coffee shop municipalities. Therefore, the results for Table 3.8 and 3.8 can be considered reliable to passing the parallel trend assumption.

The results presented in Columns 1, 2, and 3 are interesting. First, prior evidence does it indeed shows that the absolute number of soft drug crimes must have increased shortly after the implementation of the sales restriction policy within the Southern coffee shop municipalities when comparing the magnitude of the coefficients between 2011 and 2012. Even though it is reported as insignificant, this does corroborate that there was small uptick in the number of soft drug crimes

committed. However, it is most likely due to strict police enforcement that has reduced illicit drug market from expanding within 2012.

Second, the policy had a lagged effect when comparing the coefficients for 2012 and 2013. The policy resulted in a statistical significance decrease in the number of soft drug crimes that were committed within Southern coffee shop municipalities. This effect widened when comparing the coefficient of 2014. Thus, it can be concluded that the policy was beneficial for Southern coffee shop municipalities in combating the issue of drug tourism and soft drug crimes.

Columns 4, 5, and 6 present the municipal hard drug crime rate as the outcome variable by including municipality fixed effects, time fixed effects, and a linear time trend. The findings in Table 3.8 report no significance in the pre-treatment period indicating the validation of the parallel trend assumption. Though no significant difference is reported between the Northern- and Southern coffee shop municipalities about the municipal hard drug crime rate, the change in the magnitude of the coefficient between 2011 and 2012 indicates that there was a slight increase in the hard drug crime rate. This result is most likely correlated with the fact that with the increase of street dealing in soft drugs would also result in a spillover in street dealing for hard drugs as there is no longer a separation of soft drug- and hard drug users. Though the policy does not show significance up to 2014, there is a increasing downward trend for hard drug usage when comparing the change in the coefficient’s magnitude since 2012.