The next set of gateway-related policy experiments examines the effects of changes in the costs of drug consumption in the form of excise taxes on cigarettes and beer, the legality of marijuana consump- tion for medical purposes, and the likelihood that illegal drug consumption results in an arrest. These simulations are more realistic than the scenarios presented above, since policymakers have some control over these types of mechanisms. The first three experiments investigating taxes and the legalization of 6 Predicted rates for cigarette smoking remain much higher than predictions from a model ignoring unobservables, which
predicts a reduction in heavy smoking rates to 21.13 percent; heavy drinking is predicted to be 24.28 percent when ignoring unobservables.
medical marijuana are similar to how some prior studies of gateway theory approach the subject.7 The final simulation exercise examines differences in the likelihood a drug user who is committing a crime by using a particular drug is arrested.
Figure 6.6 depicts predicted outcomes from a doubling cigarette taxes in the jointly estimated model.8Higher cigarette taxes have minimal effects on predicted drinking incidence and heavy drinking rates. I forecast that drinking incidence decreases slightly. At the end of the age range, the estimated decline becomes marginally statistically significant. Estimates of heavy drinking increase slightly after age 18, though the point estimates are not statistically different from one another. Predicted marijuana use is slightly higher with the imposition of the tax increase, but only barely. Predicted own price ef- fects are also minimal but not in the expected direction since estimated cigarette smoking rates increase slightly under this tax increase scenario. Thus, while the experiment provides some evidence that higher cigarette taxes discourage drinking incidence, the forecasted impacts on heavy drinking and, especially, marijuana consumption, are slightly positive but small.
Figure 6.7 shows the results forecast from a doubling of beer taxes across all time periods, a potential policy to prevent progression into marijuana use under the assumption that drinking is a gateway to marijuana (LaChance, 1988).9 A doubling of beer taxes slightly decreases the predicted rate of age 24 marijuana use, which is 15.26 percent under the baseline case and 15.05 percent after the theoretical tax increase. Ignoring unobservables, however, the predicted consumption rate increases to 15.55 percent. Also, while use rates decline slightly, but not statistically significantly, over the middle of the age range, these declines disappear after age 22. Forecasts for cigarette smoking are largely unaffected. Forecasted heavy smoking rates are slightly higher but lie inside the baseline case confidence interval.10 Finally, impacts on forecasts of arrests in previous simulations have not been yet discussed because the earlier 7 See Thies and Register (1993) and Pacula (1998b) for examples.
8 The average state-level cigarette tax per pack of 20 cigarettes across all time periods for respondents in the survey is $0.49,
so doubling the tax would take the average to $0.98 a pack. The median tax per pack is lower ($0.41), such that doubling the tax would increase the median to $0.82 per pack.
9 The average state-level beer tax per case of 24 cans or bottles across all time periods for respondents in the survey is $0.28.
Doubling the tax would take the average to $0.56 a case. The median tax per case is lower ($0.21), such that doubling the tax would increase the median to $0.42 per case.
10Unlike the previous experiment, the predicted own price effects of higher beer taxes are in the expected direction, par-
ticularly for heavy drinking. Similar to the results for cigarette taxes, ignoring unobservables results in higher predicted drinking rates under this experiment than when unobservables are accounted for in the model, again indicating a relationship between unobservable factors and the price elasticity of drug demand.
Figure 6.7: Beer Tax Doubled
experiments either explicitly change a legally prohibited behavior or, with higher cigarette taxes, do not produce significant impacts. Here, I predict that a doubling of beer taxes produces statistically significant reductions in arrest rates until age 23. This result is consistent with evidence that alcohol is a contributing factor for many arrests not directly tied to alcohol consumption (Greenfield, 1998).
The next experiment I conduct is timely, given ongoing debates about marijuana legalization and a recent increase in the number of states legalizing marijuana for medical and even recreational uses. Without price data for marijuana with which to conduct a simulation similar to the previous two ex- periments, this simulation makes use of the variable for whether or not a state has legalized medical marijuana. I examine how the model predicts behavior to change if marijuana is legalized for medical use in all states (i.e., lowering the price of marijuana), versus the counterfactual case where medical marijuana is illegal in all states.11 Figure 6.8 presents the results of this experiment. The predictions indicate that legalized medical marijuana results in statistically significant reductions in age 24 smoking rates (from 38.43 to 33.72 percent for smoking incidence and from 30.69 to 25.25 percent for heavy smoking). These reductions are substantially overstated ignoring unobservables, with results from a model estimating the equations independently predicting cigarette use rates of 30.56 percent and heavy smoking rates of 20.89 percent. Heavy drinking rates remain almost entirely unchanged; age 24 drink- ing incidence increases by 1.88 percentage points. Predicted marijuana consumption rates increase by a statistically meaningful 3.37 percentage points at age 24 under this simulation exercise (ignoring un- observables, the predicted rate is 3.43 percentage points higher). While such an increase is not trivial, these results potentially indicate that legalization of marijuana for medical purposes might not lead to skyrocketing marijuana consumption rates, and are consistent with the findings of a recent study of use rates in states where medical marijuana has been legalized (Harper, Strumpf, and Kaufman, 2012).
My concluding experiment examines differential arrest likelihoods with respect to illegal drug con- sumption. Figure 6.9 depicts a hypothetical test of how drug consumption responds when no one is arrested for any reason versus the case where any illegal drug use results in an arrest.12 This scenario 11While one might presume that the use of medical marijuana is highly regulated, since possession requires a prescription
and thereby makes it theoretically difficult to obtain for recreational use through state-regulated programs, the experience of states that have legalized medical marijuana suggests that such laws tend to make marijuana more accessible generally (Salomonsen-Sautel et al., 2012).
12In practical terms, because of how arrests are defined in the model, defining all drug use to be legal simply means that no
one is ever arrested for any crime. This definition does not mean that cigarettes and alcohol become illicit drugs, but that any underage consumption of those goods results in an arrest.
Figure 6.8: Medical Marijuana - Legal Versus Illegal
presents another useful thought experiment that, while not realistic, potentially offers insight regarding the presence of gateway effects. Estimates of drinking incidence at age 24 decrease significantly, from 71.15 to 63.62 percent, as do forecasts for heavy drinking (from 28.30 to 24.78 percent).13 Predicted age 24 marijuana use declines slightly, from 15.60 percent to 15.19 percent. Ignoring unobservables, however, the predicted marijuana use rate jumps to 18.30 percent at age 24, and the predicted heavy drinking rate increases to 29.80 percent, again demonstrating the importance of controlling for latent factors.14