Endogenous alcohol prohibition and drunk driving.

I. Introduction

Public concern about alcohol abuse has stimulated a considerable amount of federal, state, and local legislation aimed at reducing alcohol consumption and alcohol-related behavior. An extensive empirical literature focuses on the impact of these policies on motor vehicle accidents, the leading cause of alcohol-related deaths. For example, studies suggest that increases in federal or state excise taxes on alcohol discourage heavy drinking and reduce motor vehicle fatalities [3; 6; 8; 11]. State-level legislation, such as minimum drinking age laws or preliminary-breath-test laws, has been shown to significantly lower motor vehicle fatalities [2; 13]. Although local legislation is examined less frequently in the literature, there is some evidence that county-level restrictions on alcohol sales lower the rate of alcohol-related motor vehicle fatalities. Saffer and Grossman [14; 15], Chaloupka, Saffer and Grossman [2], and Wilkinson [17] control for the proportion of a state's population residing in counties where alcohol sales are prohibited and find a negative impact on state-level motor vehicle fatality rates. Winn and Giacopassi [18] report that Kentucky counties which prohibit alcohol sales have significantly lower alcohol-related motor vehicle accident rates. Using county-level data on the state of Texas, Jewell and Brown [7] find that limitations in the number of licensed alcohol vendors could decrease alcohol-related motor vehicle accidents and fatalities.

All of this literature focuses on the effects of a particular policy - for instance, raising the excise tax on alcohol or reducing its availability - on some measure of drinking and driving, treating the observed policy instrument exogenously in the model. A problem with this approach, however, is that these policies are chosen either through direct referendum or by elected officials who presumably react to the preferences of their constituents. Therefore, failing to account for the public choice determinants of the observed policy introduces a potential bias in the empirical results: Observed variations in alcohol-related behavior may reflect differences in the characteristics of voters, rather than differences in how individuals react to policy-induced changes in their constraints as implied in previous literature. Without accounting for the endogeneity of the observed policy, the researcher cannot distinguish the degree to which the policy itself induces changes in individual behavior.

Evaluating the effectiveness of a policy requires that the empirical methodology accounts for the collective decision-making process by which the policy is established. This paper provides estimates of the impact of county-level alcohol prohibition on fatal alcohol-related motor vehicle accidents, taking into account the public choice nature of observed prohibition policies. Data on the 254 counties in Texas allow for a unique opportunity to empirically investigate both of these issues. These counties exhibit diverse demographic characteristics and considerable variation in the legalization of alcohol sales; residents in the 53 "dry" counties have voted to prohibit all alcohol sales, while those in the remaining 201 "wet" counties have voted to allow the sale of some form of alcohol.

Since alcohol prohibition in Texas results from the collective decision-making of the county electorate, the median voter model is a convenient framework in which to examine the determinants of prohibition legislation. Whether alcohol prohibition, in turn, impacts alcohol-related motor vehicle fatalities is determined by the sensitivity of alcohol consumption to changes in the full price of alcohol, which includes the dollar price of alcohol plus the travel cost associated with acquiring alcohol. Prohibiting alcohol sales in a county raises the travel-cost component of the price of alcohol; if consumption is sufficiently responsive to this full price increase, then fatal alcohol-related motor vehicle accidents should decrease. Alternatively, if alcohol consumption is insensitive to changes in the full price of alcohol, then prohibition only increases the distance consumers must travel to acquire alcohol and, therefore, may actually raise the risk of being involved in a fatal motor vehicle accident.

The paper proceeds as follows. First, we apply the median voter model to investigate a county's choice to prohibit alcohol as a function of the characteristics of its electorate, such as religious affiliation, race, political ideology, alcohol availability in surrounding counties, and other economic factors. Our results show that there are characteristic differences between wet and dry counties, indicating the importance of accounting for the public choice nature of a policy when assessing its impact on alcohol-related behavior. Second, we estimate the effect of alcohol prohibition on fatal alcohol-related motor vehicle accidents using a two-stage estimation to correct for the endogeneity of the observed prohibition policy in a county. Finally, the two-stage estimates are compared with those in which the observed prohibition policy is assumed to be an exogenous event. In general, treating prohibition legislation exogenously leads to biased estimation of the impact of alcohol restrictions on drinking and driving.

II. The Effect of Prohibition on Fatal Alcohol-Related Motor Vehicle Accidents

We assume that an individual derives utility from some composite choice of consuming alcohol and driving, either directly or indirectly as inputs into some final good such as entertainment. The costs of drinking and driving include the dollar price of alcohol plus the travel cost associated with acquiring alcohol. Beyond some level of alcohol consumption, however, drinking and driving becomes illegal and dangerous, so that the cost of drinking and driving must also include the expected probabilities of being caught drinking and driving as well as being involved in an accident. In this sense, the consumption decision of whether to drink and then drive can be viewed as a demand function for drinking and driving.

The county-level demand for drinking and driving is a function of the full price of alcohol - the dollar price plus its travel-cost component - as well as income, population, preferences toward alcohol consumption, local driving conditions, and local resources devoted to detecting drunk drivers. Prohibition legislation attempts to reduce alcohol consumption by increasing the travel cost required to acquire alcohol. If consumption is sufficiently responsive to full price increases, then fatal alcohol-related motor vehicle accidents should be lower in counties where alcohol sales are prohibited, other factors constant.

Suppose that consumption of alcohol in a county, A, depends on the dollar price of alcohol, P, and the average distance to the nearest vendor, D, or

A(P, D), (1)

where [Delta]A/[Delta]P [less than] 0 and [Delta]A/[Delta]D [less than] 0. Let fatal alcohol-related motor vehicle accidents in a county, F, increase with both alcohol consumption and the total miles driven to purchase alcohol, M, or

F(A, M), (2)

where [Delta]F/[Delta]A [greater than] 0 and [Delta]F/[Delta]M [greater than] 0. In turn, total miles driven depends on the number of purchase trips and the distance of each trip. Assuming that the number of trips is monotonically increasing in alcohol consumption, total miles driven can be written as

M(A, D), (3)

where [Delta]M/[Delta]A [greater than] 0 and [Delta]M/[Delta]D [greater than] 0.(1) To establish the relationship between fatal alcohol-related motor vehicle accidents and distance to the nearest vendor, substitute equations (1) and (3) into equation (2) and rewrite as the following:

F {A(P, D), M[A(P, D), D]}. (2a)

A county that prohibits alcohol sales effectively raises the required travel distance to the nearest vendor. The total effect of prohibition on fatal alcohol-related motor vehicle accidents can be distinguished by differentiating equation (2a) with respect to distance:

([Delta]F/[Delta]A)([Delta]A/[Delta]D) + ([Delta]F/[Delta]M)([Delta]M/[Delta]A)([Delta]A/[Delta]D) + ([Delta]F/[Delta]M)([Delta]M/[Delta]D) (4)

The first term in equation (4) represents the direct effect on fatal accidents from changes in the travel-cost component of the full price of alcohol; increasing travel distance reduces alcohol consumption and, in turn, reduces fatal alcohol-related motor vehicle accidents. The remaining terms collectively represent the effect of prohibition-induced changes in the total miles traveled. The second term represents the effect of increased travel distance on the number of trips taken to obtain alcohol; as the required travel distance increases, consumers take fewer trips to acquire alcohol which reduces fatal alcohol-related accidents. As seen in the third term, increased travel distance raises the miles driven per trip and, thereby, raises the number of fatal alcohol-related accidents. The sign of equation (4) is indeterminate, depending on the magnitudes of these three terms. For example, if alcohol consumption is completely insensitive to full price changes brought on by changing distance (i.e., [Delta]A/[Delta]D = 0), then equation (4) is positive. In this case, increasing travel distance to the nearest vendor simply induces a consumer to drive a longer distance to obtain alcohol. Alternatively, if consumption is sufficiently responsive to increases in travel-distance costs, then alcohol prohibition will reduce fatal alcohol-related motor vehicle accidents.

III. The Collective Decision to Prohibit Alcohol

Legislation to prohibit alcohol sales in a county is established through the collective decision-making of the electorate. The median voter theory provides a useful framework in which to examine the determinants of prohibition legislation. In this respect, our application is similar to Deacon and Shapiro [4], who model local voting outcomes to establish the California Coastal Commission as a function of the characteristics of local electorates, and to Richer [12], who relates the decision to propose and/or pass local growth control measures to the characteristics of California cities. In the case of Texas, the county-level choice of whether to prohibit alcohol sales reflects the characteristics and preferences of county voters, such as religious affiliation, political ideology, economic interests associated with alcohol availability, alcohol restrictions in surrounding counties, and demographic factors.

For instance, alcohol prohibition may result from strong religious convictions of the electorate: Those who are affiliated with conservative Protestant denominations have been found to favor alcohol restrictions, while Catholics tend to be more opposed to such restrictions.(2) Political ideology may also affect the county-level choice for alcohol restrictions, perhaps reflecting the electorate's willingness to impose government restrictions on personal behavior. In addition, political pressure against alcohol restrictions can be exerted by groups with an economic interest in alcohol availability, such as the entertainment or tourism industries.

Voters in Texas have the opportunity to make specific, and incremental, decisions regarding the types of alcohol sales in their county. Local option elections can be used to prohibit or legalize the sale of beer, the sale of both beer and wine, the sale of all alcoholic beverages, or the sale of all alcoholic beverages including mixed drinks.(3) A group of voters may request a local option election at any time to prohibit (or legalize) an alcoholic beverage sale by filing an application, signed by 10 or more registered voters, with the county clerk. The applicants then have 30 days to circulate copies of the "petition to prohibit" (or legalize) and collect signatures from at least 35 percent of the county's registered voters. Expenses associated with the election are paid by the county unless there is more than one election regarding a specific type of alcoholic beverage in a single year; additional elections must be financed by the applicants in the form of a nonrefundable deposit. The outcome of an election is immediately binding if the majority of voters decides to legalize the sale of some beverage type and takes effect in 30 days if the decision is to prohibit a beverage sale. These decisions can be overturned in subsequent local option elections after one year [16]. Historically, referendums to change the legal status of alcohol are relatively rare; for example, only 3 counties voted to prohibit alcohol sales between 1991-93.

The legal status of alcohol in Texas counties as of 31 August 1993 is compiled by the Texas Alcoholic Beverage Commission. Of the 254 counties in Texas, 53 prohibit the sale of alcohol and are considered dry. The remaining 201 counties allow some form of alcoholic beverages to be sold and are considered wet. Of these 201 wet counties, 96 allow the sale of all types of alcoholic beverages including mixed drinks, 11 counties allow the sale of beer (with 4 percent alcohol or less by volume), 5 counties allow the sale of both beer and wine (containing 14 percent alcohol or less), and 89 counties permit the sale of distilled spirits as well as beer and wine but not mixed drinks.

IV. Empirical Methods 

We estimate a demand function for drinking and driving, applying a two-stage estimation to account for the endogeneity of the electorate's observed prohibition choice. In the first-stage regression, the median voter framework allows us to test for an effect of various county-level characteristics on prohibition status. The dependent variable measures wet/dry status - the 201 wet counties are assigned a one, and the 53 dry counties receive a zero - so that a probit estimation is appropriate. The assumption is that counties are in equilibrium regarding alcohol prohibition status. Since it is possible to amend this status at any time, a county's present status is presumably the one most preferred by its voters.

Following the median voter model, we include median county-level income and education as explanatory variables. In addition, the percentage of county households below the poverty level and the percentage of residents in a census-defined urbanized area are included. The percentage of county residents who are Black and Hispanic are used to control for the effect, if any, of racial differences in alcohol preference. The two largest religious groups in Texas are the Roman Catholic Church and the Southern Baptist Church; the percentage of county residents who belong to these groups are used to proxy the nature and degree of religious beliefs.

The percentage of voters registered in each political party is typically used to measure voter ideology or sentiment not reflected in other taste variables. However, since Texas voters do not register with a party, these data are unavailable. Instead, the percentage of voters in each county who voted for the Democratic candidate in the 1993 runoff election for a vacant U.S. Senate seat captures political ideology.(4) The percentage of county population employed in the tourism industry, presumably highly dependent on alcohol sales, is used to measure the extent to which industry forces can affect alcohol availability. Finally, we use the average number of retail alcohol licenses in bordering counties, divided by a county's own population, to control for the relative availability of alcohol in surrounding counties. This variable is included since the cost and benefit of alcohol prohibition within a given county may be affected by the availability of alcohol in adjacent counties.

In the second-stage regression, we estimate the effect of alcohol prohibition on fatal alcohol-related motor vehicle accidents, other factors constant; the predicted probabilities from the first-stage probit regression are used as instruments in the second-stage.(5) Fatal alcohol-related motor vehicle accidents are averaged over five years (1988-92), in order to minimize any random yearly fluctuations, and then divided by the number of lane miles of road in a county. Also included are demographic factors which influence alcohol consumption as well as drinking and driving. Religious affiliation is included to control for unobservable sentiment towards drinking and driving.(6)

County-level information on the dollar price of alcohol is unavailable and, in any case, would be unobservable in dry counties. Omitting the dollar price is not a serious problem to the degree that the dollar price of alcohol and its travel-cost component are highly correlated. Furthermore, if the dollar price of alcohol is constant among counties with equivalent alcohol availability, then variations in the travel-cost component of price across counties, measured by alcohol availability, effectively capture full price changes. In this sense, whether alcohol is prohibited captures the observable component of price changes across counties. State tax rates have been used in state-level studies to reflect changes in the dollar price of alcohol; however, the tax rate is invariant across counties in Texas. The accessibility of alcohol in bordering counties also affects the price of acquiring alcohol in a county. This effect is captured by including the border alcohol availability measure in the estimation.

Police spending per person in a county proxies the expected probability of detection and conviction of drunk drivers. The number of registered vehicles captures traffic density; higher traffic density should increase accident frequencies, although it may reduce serious accidents since travel speeds tend to decrease with traffic density. County vehicle miles driven reflects the demand for driving, expected to positively impact alcohol-related accidents. The percent of county roads that are centerlane (i.e., non-rural) is also included as a measure of local road conditions. In addition, a dummy variable is included if the county has an interstate highway within its boundaries. Interstate highways typically have higher speed limits but are also constructed to higher safety standards; therefore, the impact of interstate highways on fatal accident rates is ambiguous. Summary statistics (means and standard deviations) for all the variables are given in Table I.

V. Results

The results from the first-stage probit regression are reported in Table II. Counties with a higher percentage of Baptists tend to be dry, while counties with more Catholics tend to be wet. These results confirm previous research showing that religious affiliation affects preferences toward alcohol consumption and alcohol restrictions. More urbanized counties tend to be wet rather than dry. Politically liberal counties, as measured by percent Democratic voters, are more likely to be wet, although it is unclear whether political liberals have less aversion to alcohol or more aversion to government limits on individual behavior. The percent of county population employed in tourism [TABULAR DATA FOR TABLE I OMITTED] is positive and significant, suggesting that economic dependency on alcohol sales plays a role in determining county level availability.(7)

Table III reports the estimates of the impact of alcohol prohibition on fatal alcohol-related motor vehicle accidents per lane mile of county road. The first column corrects for the endogeneity of the observed prohibition legislation in a county, using the predicted probabilities from the first-stage estimates as instruments.(8) The results suggest that the demand for the composite good drinking and driving is sensitive to the lower travel cost of acquiring alcohol in wet counties, independent of the electorate's sentiment toward alcohol: Wet counties experience approximately 0.003 more alcohol-related fatal motor accidents per lane mile than do dry counties. Since the average county in Texas has 715 lane miles of road, this crudely translates into 2.145 more fatal alcohol-related motor vehicle accidents per year for the average county. Furthermore, in 1992 the average county in Texas had 1.16 fatalities per fatal alcohol-related motor vehicle accident; according to these estimates, wet counties experienced approximately 2.49 more fatalities per year.

The coefficient on median household income is significant and positive, as is the coefficient on households living in poverty. These results may suggest that individuals in higher income brackets consume more alcohol as income increases, while individuals in poverty may simply substitute cheaper brands of alcohol and not decrease consumption as income decreases. The coefficient on border availability is negatively and significantly related to alcohol-related fatal accidents. Perhaps this implies that individuals who drive outside of their county to drink have a higher probability of being involved in an accident while drinking and driving in a wet county, supporting the prediction that counties with more availability have more accidents.(9) As expected, road conditions and traffic density impact accidents; an increase in the percent of centerlane roads significantly lowers fatal alcohol-related accidents, while an increase in vehicles miles driven significantly raises the number of fatal accidents per lane mile.

Table II. Probit Regression First-Stage Results - Dependent Variable
is Wet/Dry Status

Constant                                        6.3812
                                               (1.454)
Median Income                                  -0.00006
                                              (-0.080)
Percent Below Poverty Level                    -0.0465
                                              (-0.857)
Median Education                               -0.4048
                                              (-1.165)
Percent Urbanized                               1.1812
                                               (2.493)(***)
Percent Black                                  -0.0205
                                              (-1.148)
Percent Hispanic                               -0.0244
                                              (-1.239)
Percent CathoLic                                0.03278
                                               (1.704)(*)
Percent Baptist                                -0.05423
                                              (-4.501)(***)
Percent Democrat                                0.0350
                                               (2.363)(**)
Percent Tourism                               264.64
                                               (1.871)(*)
Border Availability                             4.0426
                                               (0.778)
Chi-Square                                     75.02

(t-statistics in parentheses)

*** significant at 0.01

** significant at 0.05

* significant at 0.10

It is also interesting to note that the religious measures are insignificant in the first column of Table III, contradicting the results from previous studies. However, since alcohol prohibition is [TABULAR DATA FOR TABLE III OMITTED] endogenous, religious affiliation may impact alcohol prohibition and not necessarily further affect alcohol-related motor vehicle accidents. Comparing the results from Table II to those presented in the first column of Table III suggests that preferences for or against alcohol due to religious affiliation are seen more directly in whether or not to prohibit alcohol.

The second column of Table III reports an OLS regression treating the observed prohibition legislation exogenously, as is the typical approach in the literature. Falling to account for the public choice nature of prohibition would bias the estimation of its effect on alcohol-related accidents. According to the OLS results, the coefficient on wet/dry status is biased downward.(10) The OLS estimates would lead one to conclude that alcohol prohibition in Texas counties is marginally effective in reducing drunk driving, both quantitatively and statistically. According to the two-stage results, an average wet county would have approximately 2.145 more fatal alcohol-related motor vehicle accidents than a dry county; the OLS results suggest that an average wet county would only have 0.505 more fatal accidents. In addition, the coefficient on percent Baptist is strongly significant with OLS; however, OLS does not address the issue of sentiment for prohibition legislation.

VI. Conclusion

Policies aimed at reducing drunk driving typically attempt to raise the costs of drinking and driving in one of two ways. First, deterrence policies raise the probability of detecting and convicting drunk drivers, but are not necessarily designed to reduce alcohol consumption. Several studies show that increasing the severity of punishment as well as the resources devoted to detection and conviction reduce motor vehicle accidents.(11) Second, policies directed at reducing alcohol consumption attempt to raise the full price of alcohol - either the dollar price or the travel-cost component; these policies include taxation and availability restrictions such as minimum drinking age laws and limiting the number or type of alcohol vendors. The policy-related literature generally estimates the impact of a particular policy on some measure of motor vehicle fatalities, assuming that the observed policy instrument is exogenous in the estimation. The potential problem with this approach is that drinking and driving legislation is ultimately chosen either through direct referendum or by elected representatives and, therefore, reflects the preferences of the electorate where the policy is imposed. Failing to account for the collective decision-making determinants of a policy introduces a potential bias in the results.

This paper addresses the impact of county-level alcohol prohibition - an area less examined in the literature - on fatal alcohol-related motor accidents, taking into account the endogeneity of the prohibition legislation. The empirical results provide evidence that the consumption of drinking and driving is responsive to prohibition-induced changes in individuals' constraints. The results suggest that dry counties have 2.145 fewer fatal alcohol-related motor vehicle accidents than wet counties per year. Failing to account for the collective decision-making process by which prohibition legislation is established may lead to biased estimation of the policy effect. The results presented here show that the corrected effect is over four times as large as the uncorrected effect. Furthermore, this endogeneity issue raises potentially important questions concerning the effectiveness of other anti-drinking and driving policies established at the local, state, and federal levels.

We thank an anonymous referee for helpful comments. This paper was completed while Jerrell Richer was a Gilbert F. White Fellow at Resources for the Future.

1. Hoarding, which can occur in counties allowing sales for off-premise consumption, can diminish the effects on total miles driven from changes in the travel distance per trip or the number of trips.

2. In discussing the historical determinants of federal and state alcohol tax rates, Meier [10] shows that Baptists, i.e., conservative Protestants, are associated with higher taxes and Catholics are associated with lower taxes. Similar results are found in various studies on alcohol-related behavior. For example, see Chaloupka, Saffer and Grossman [2].

3. We consider only choices made at the county-level. The voters of a town, incorporated area, or justice precinct can impose stronger restrictions in their jurisdictions, creating the possibility of islands of greater or lesser availability within any county.

4. On 5 June 1993, the state-wide runoff election pitted Republican Kay Bailey Hutchison against Democrat Robert Krueger, with Hutchison winning the election.

5. The model can also be viewed as the solution to a selectivity problem known as a treatment effects model. The interpretation is that the coefficient on observed prohibition does not measure the impact on alcohol-related deaths if counties with observed prohibitions would have relatively more or fewer alcohol-related deaths whether or not there were prohibitions. Greene [5] further explains this selectivity model, and Maddala [9] presents various applications of the model.

6. Chaloupka, Saffer and Grossman [2], Coate and Grossman [3], and Saffer and Grossman [14; 15] present the argument for including religious affiliation to measure aversion to alcohol or alcohol sentiment among U.S. states but do not correct for the public choice aspects of observed policy instruments.

7. Percent Democrat and Percent Tourism are excluded from the second-stage fatal accident equation as identification restrictions.

8. The reported t-statistics are corrected for endogeneity.

9. The dependent variables in Tables II and III may be spatially autocorrelated since dry counties in Texas tend to be surrounded by dry counties and wet counties tend to be surrounded by wet counties [1, 71-2]. The border availability measure may capture this spatial autocorrelation.

10. A Hausman specification test is used to test for exogeneity of wet/dry status. The null hypothesis of exogeneity is rejected at a 10 percent level of significance.

11. Some examples of deterrence policies include revoking offenders' drivers licenses, imposing mandatory fines, and preliminary-breath-test laws [2; 14; 17].

References

1. Berry, William D. Understanding Regression Assumptions. Newbury Park, Calif.: Sage Publications, 1993.

2. Chaloupka, Frank J., Henry Saffer, and Michael Grossman, "Alcohol-Control Policies and Motor Vehicle Accidents." Journal of Legal Studies, January 1993, 161-86.

3. Coate, Douglas and Michael Grossman, "Effects of Alcohol Beverage Prices and Legal Drinking Ages on Youth Alcohol Use." Journal of Law and Economics, April 1988, 145-71.

4. Deacon, Robert T. and Perry Shapiro, "Private Preferences for Collective Goods Revealed Through Voting on Referenda." American Economic Review, December 1975, 943-55.

5. Greene, William H. Econometric Analysis, Second Edition. New York: Macmillan Publishing, 1993.

6. Grossman, Michael, Jody L. Sindelar, John Mullahy, and Richard Anderson, "Policy Watch: Alcohol and Cigarette Taxes." Journal of Economic Perspectives, Fall 1993, 211-22.

7. Jewell, R. Todd and Robert W. Brown, "Alcohol Availability and Alcohol-Related Motor Vehicle Accidents." Applied Economics, August 1995, 759-65.

8. Laixuthai, Adit and Frank J. Chaloupka, "Youth Alcohol Use and Public Policy." Contemporary Policy Issues, October 1993, 70-81.

9. Maddala, G. S. Limited-Dependent and Qualitative Variables in Econometrics. Cambridge, UK: Cambridge University Press, 1983.

10. Meier, Kenneth J. The Politics of Sin: Drugs, Alcohol, and Public Policy. Armonk, N.Y.: M. E. Sharpe Publishing, 1994.

11. Mullahy, John and Jody L. Sindelar, "Do Drinkers Know When to Say When? An Empirical Analysis of Drunk Driving." Economic Inquiry, July 1994, 383-94.

12. Richer, Jerrell, "Explaining the Vote for Slow Growth." Public Choice, March 1995, 207-23.

13. Saffer, Henry and Frank J. Chaloupka, "Breath Testing and Highway Fatality Rates." Applied Economics, July 1989, 901-12.

14. Saffer, Henry and Michael Grossman, "Beer Taxes, the Legal Drinking Age, and Youth Motor Vehicle Fatalities." Journal of Legal Studies, June 1987, 351-74.

15. -----, "Drinking Age Laws and Highway Mortality Rates: Cause and Effect." Economic Inquiry, July 1987, 403-17.

16. Texas Alcoholic Beverage Commission. Texas Alcoholic Beverage Code. Austin, Texas: Texas Alcoholic Beverage Commission, 1993.

17. Wilkinson, James T., "Reducing Drunken Driving: Which Policies are Most Effective?" Southern Economic Journal, October 1987, 322-34.

18. Winn, Russell G. and David Giacopassi, "Effects of County-Level Alcohol Prohibition on Motor Vehicle Accidents." Social Science Quarterly, December 1993, 783-92.