Putting out fires: an examination of the determinants of state clean indoor-air laws.

1. Introduction

Few products have generated as much public scrutiny as tobacco. In light of the health consequences associated with smoking, the federal government took several steps in the 1960s (such as published health warnings and restrictions on cigarette advertising) to reduce the incidence of smoking. Beginning in the 1970s, state governments began to play an increasing role in the anti-smoking campaign by adopting clean indoor-air laws, which restricted smoking in particular locations, such as restaurants, bars, and work sites. (1)

Various studies, including Wasserman et al. (1991), Chaloupka (1992), Chaloupka and Saffer (1992), and Yurekli and Zhang (2000), generally confirmed that clean indoor-air laws are effective at reducing cigarette demand. However, the nature of the smoking restriction appears to matter. In particular, Chaloupka (1992) and Chaloupka and Saffer (1992) found that restrictions on smoking in public places reduced cigarette demand, whereas restrictions at private work sites do not significantly reduce demand. (2)

Due to variations in antismoking laws across states, only a few studies have addressed the demand for clean indoor-air laws. In a survey of individuals from San Luis Obispo, California, for example, Boyes and Marlow (1996) found nonsmokers and women to be more supportive of a smoking ban in restaurants and bars. Alternatively, Chaloupka and Saffer (1992) used state-level data and found that differences in smoking restrictions in public places and at private work sites depend on several factors, including cigarette prices, income, tobacco production, religious affiliation, and political activity. More recently, Hersch, Del Rossi, and Viscusi (2004) found that voting preferences of state residents and the political affiliation of lawmakers were the main factors that affect whether or not a restriction is adopted at a specific location (for example, restaurants, bars, malls, enclosed arenas, and hospitals). However, smoking habits could have an effect on voting preferences; therefore, these results could show the indirect effect of tobacco consumption on smoking restrictions.

There are several limitations of the existing literature that lead to a need for further analysis. First, studies have ignored the potential role of cigarette taxation as a determinant of smoking bans. For example, if states adopt a general antismoking position, taxes could be used in conjunction with smoking bans to reduce tobacco consumption. Thus, depending on the argument, states may view taxation as a policy substitute or a policy complement to clean indoor-air laws. Second, the potential endogeneity of right-hand-side variables has not been addressed. For example, tobacco consumption is often included as a determinant of smoking bans; however, tobacco consumption may potentially be endogenous. In general, studies have found that restrictions on smoking in public places reduce cigarette demand, which implies that cigarette consumption is potentially endogenous in the reverse regression of smoking on restrictions. It is unfortunate that the literature has not addressed this possibility in a systematic manner. (3) Third, except for Hersch, Del Rossi, and Viscusi (2004), studies have failed to account for differences in the demand for clean indoor-air laws across different restricted locations.

This study addresses these limitations by estimating the demand for clean indoor-air laws using a panel of state-level data. By using a panel data set, we were able to control for state-specific effects that made it more or less likely that a given state had a smoking ban in a particular area. Specifically, we accounted for the potential endogeneity of cigarette consumption and taxation within a random-effects Probit model of the adoption of smoking restrictions across six locations--public places, government buildings, private work sites, schools, health care facilities, and restaurants. Briefly, we found statistical differences across locations; however, the results revealed that the probability of a smoking ban being adopted tends to be lower in states with higher per capita cigarette consumption, more politically conservative lawmakers, higher metropolitan populations, lower per capita income, and higher tobacco production. However, the effect of cigarette taxes on the probability that a state will adopt a smoking ban depends on the handling of the endogeneity of cigarette consumption and cigarette taxes. For example, by treating cigarette consumption and cigarette taxes as exogenous, we found that taxes complement smoking restrictions. However, when endogeneity is accounted for, the role of cigarette taxes shifts toward being a policy substitute.

Section 2 of this paper outlines the empirical procedures. Section 3 discusses our estimation results. Concluding remarks are provided in section 4.

2. Empirical Specification and Econometric Issues

Empirical Specification

Our model addressed the factors that affect the probability that a state will adopt a clean indoor-air law. We obtained State Legislated Actions on Tobacco Issues (SLATI) data from the American Lung Association, which provide information about when a state has adopted a smoking ban (i.e., equals one if smoking ban is adopted, zero if not) in six particular locations (public places, government buildings, private work sites, schools, health care facilities, and restaurants) for 1980 through 2000. (4) These dichotomous variables were then treated as dependent variables in a series of Probit regressions. (5)

We hypothesized that the probability that a state will adopt a smoking ban will depend on several factors. First, we included per capita cigarette consumption as a regressor. Depending on the argument, however, cigarette consumption may increase or decrease the probability of a smoking ban being adopted. For example, it may be that higher smoking rates induce a state to adopt a smoking ban in order to reduce the incidence of smoking. Alternatively, higher smoking rates may signal greater political clout of smokers, who are less likely to favor smoking bans. (6)

Second, state-level tax rates on cigarettes are included to control for possible substitutability or complementarity between taxes and smoking bans. It may be, for example, that if states adopt a general anti-smoking position, taxes could be used in conjunction with smoking bans to reduce tobacco consumption. In this case, higher tax rates will correlate with a greater probability of adopting a smoking ban. Alternatively, it may be that states are particularly keen on raising tax revenue and view taxes as competing with smoking bans. Therefore, if higher tax rates are adopted in an effort to raise tax revenue, then states will be less likely to adopt smoking bans, which reduce demand and tax revenue. In anticipation of the estimation methods, therefore, it seems reasonable to consider cigarette taxes and the chance of adopting smoking restrictions as being simultaneously affected by certain characteristics of states.

Third, political pressure likely influences whether a state adopts a smoking ban. For example, Hersch, Del Rossi, and Viscusi (2004) found more politically conservative states are less likely to adopt a smoking ban, which they suggest may be tied to a lack of support for government involvement in markets. (7) In an effort to measure this, we included an index of conservatism in our model. This measure is an annual index of the voting pattern of every member of Congress regarding 20 key issues chosen by the American Conservative Union (ACU). Each member of Congress is awarded five points for each vote deemed "conservative" by the ACU; therefore scores can range from 0 to 100. A score of 100 would mean that this member of Congress had a perfect conservative voting record for that year. We derived a state-level annual score for a given year from the average of all members of Congress from that state.

Fourth, we included the percentage of the state's population living in a metropolitan area. This variable accounted for several possible scenarios. For instance, if secondhand smoke is particularly a nuisance in highly populated areas, then a higher share of the population living in metropolitan areas may lead to greater pressure to adopt smoking bans. Alternatively, since metropolitan areas have more businesses (e.g., restaurants, bars, shopping centers) for which smoking bans could be detrimental, political pressure may be against the adoption of smoking bans. (8)

Fifth, per capita income is included in our specification. Some studies (for example, those by Cremieux, Ouellette, and Pilon [1999] and Hitiris and Posnett [1992]), have found that income is positively correlated with health outcomes. Because of these findings, we expect states with higher per capita income to be more likely to take an anti-smoking position, and therefore have a higher probability of adopting a smoking ban.

Lastly, Chaloupka and Saffer (1992) and Hersch, Del Rossi, and Viscusi (2004) included measures of the importance of tobacco production to states because states that are heavy producers of tobacco are expected to be less supportive of smoking bans. Consequently, we included the value of tobacco production as a percentage of gross state product. In addition, there may be other variables that we did not capture, but any remaining unobserved heterogeneity can be controlled for in the panel framework. Descriptive statistics of all variables are provided in Table 1, whereas the Appendix provides the sources of our data.

Econometric Issues

Our data consisted of annual observations across the United States. Since a pooling model cannot control for unobserved state-specific effects that may exist, it is rational to use panel data techniques. (9) For panel choice models, however, several estimation options are available. One may consider, for instance, using a fixed effects Logit model based on the conditional likelihood function (see Chamberlain 1980). This model has the advantage of remaining consistent when individual state fixed effects are correlated with any regressors. However, a critical disadvantage of this model is that it cannot allow for time-invariant regressors. Results are sensitive to the inclusion of regressors with little variation, which is troubling in our model because some of our regressors exhibit little variation over time yet vary across states. As an alternative, the random effects Probit model is robust to the presence of time invariant variables, and as such, we deemed this model to be more appropriate for our purposes. (10)

Another issue that needs to be addressed is a potential endogeneity problem because our choice model may be contaminated with endogenous explanatory variables. In particular, the adoption of smoking restrictions can potentially be simultaneously determined with cigarette consumption and taxation, for strict regulations on smoking may affect cigarette consumption and taxation. (11) Failing to control for endogeneity of these variables can lead to biased estimates. For this purpose, we employed two-stage panel random effects Probit models, which required the use of instrumental variables. In particular, consider the following choice model:

[y.sup.*.sub.it] = [X.sub.it][beta] + [w.sub.it][gamma] + [[alpha].sub.i] + [[epsilon].sub.it], i = 1, ..., n; t = 1, ..., T, (1)

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.] (2)

where [y.sup.*.sub.it] is an index latent variable of state i in period t, [X.sub.it] are exogenous regressors, [[epsilon].sub.it] is the error term, [w.sub.it] are endogenous variables (such that cov([w.sub.it], [[epsilon].sub.it]) [not equal to] 0), and [[alpha].sub.i] is the unobserved heterogeneity of state i. In our version, [y.sub.it] corresponds to each of the six public smoking locations (public places, government buildings, private work sites, schools, health care facilities, and restaurants), equaling one if state i limits smoking in the respective location at time t and equalling zero if not. Included in [X.sub.it] is the ACU index of conservatism, the percentage of the state population living in a metropolitan area, per capita income, and tobacco production as a percentage of gross state product; per capita cigarette consumption and cigarette tax rates are in [w.sub.it].

Assuming that all regressors are uncorrelated with [[alpha].sub.i], our problem was to estimate [beta] and [gamma]. Lee (1981) suggested a two-stage estimation procedure that mimics the usual linear two-stage least squares. The required assumption is the existence of the following linear reduced-form equation:

[w.sub.it] = [Z.sub.it][alpha] + [u.sub.it], (3)

where [Z.sub.it] includes both exogenous variables [X.sub.it] and instrumental variables for wit. We then replaced wit in Equation 1 by its predicted value from the first-stage reduced form equation in (3), following Newey (1987), which yielded:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.] (4)

Whether there exists such a linear reduced-form equation is an open question, because [y.sup.*.sub.it] and [w.sub.it] enter Equation 1 in a nonlinear fashion. However, we are not aware of any parametric solution to the endogeneity problem that would not rely on the assumption that the linear Equation 3 exists. (12)

In the panel data model, we still had difficulty eliminating the effect of the state specific effects, [[alpha].sub.i]. For this, we follow the usual random effect treatment by adopting the procedure suggested by Butler and Moffitt (1982). That is, we treated [[alpha].sub.i] as a random variable and use a Gaussian-Hermite quadrature to approximate the integration in obtaining a proper likelihood function. This method is the usual procedure for a random effects model and the same procedure can be used in the two-stage random effects Probit model. Arendt (2001) also considered the same approach for the random effects Probit model with time invariant endogenous variables.

Rivers and Vuong (1988) suggested an alternative procedure to control for endogeneity. Specifically, assuming that

[[epsilon].sub.it] = [rho][u.sub.it] + [e.sub.it], (5)

we can rewrite Equation 1 as:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.] (6)

where we replace [u.sub.it] in Equation 5 by the residual from the regression in Equation 3. In linear models, the same estimates of [beta] and [gamma] are obtained from Equations 4 and 6, whereas the standard errors are different. However, this equivalence does not hold true in our case because the underlying choice models are nonlinear. Hence, the estimates from Equation 4 and 6 will differ. A key benefit of the Rivers and Vuong (1988) approach, however, is that we can test for the existence of endogeneity from Equation 6 by examining the significance of p using t, or likelihood ratio tests for the hypothesis [rho] = 0, which corresponds to the usual Hausman test for endogeneity (see also Vella and Verbeek 1999). (13) Accordingly, rather than simply pick an endogeneity correction method, we reported results from the estimations of both Equations 4 and 6. Also, as a point of comparison, we reported results by using the usual random effects Probit model not corrected for endogeneity.

3. Estimation Results

Results from the estimation of our models are provided in Tables 2-5. To begin, Table 2 provides results from the estimation of Equation 3 for both per capita cigarette consumption and cigarette tax rates. (14) As the table illustrates, the coefficients of many of the regressors are significantly different from zero. For example, with respect to per capita cigarette consumption, the reduced-form estimates reveal that consumption is higher in states that have a more conservative bent, smaller metropolitan population, lower per capita income, higher tobacco production, lower cigarette wage, and lower poverty rate. With respect to cigarette taxes, many of the coefficients switch signs compared with the cigarette consumption results, which could partly be tied to the inverse relationship between cigarette taxes and cigarette consumption. The results show that cigarette taxes are negatively related to conservatism, tobacco production, and cigarette wages, yet positively related to per capita income, cigarette advertising, and the poverty rate. Therefore, more conservative states with a higher presence of tobacco and cigarette wage earners are less apt to aggressively tax cigarettes.

The Probit results are reported in Tables 3 through 5. In each table, we report the estimated coefficients, t statistics, and marginal effects (in brackets). (15) We first estimated the random effects model without endogeneity correction, which is similar to the estimates made by Chaloupka and Saffer (1992) and Hersch, Del Rossi, and Viscusi (2004). Looking at Table 3, there are several interesting results. First, unlike Hersch, Del Rossi, and Viscusi (2004), who found that smoking rates play little role in the determination of smoking restrictions we found that increased cigarette consumption significantly reduces the probability that a state will limit smoking in several locations (except for private work sites and health care facilities). (16) Perhaps this correlates with cigarette smokers, who tend to be less supportive of smoking bans, having greater political clout. Second, for those coefficients that are significantly different from zero, cigarette taxes appear to complement smoking restrictions in that higher taxes correspond to a greater probability that states restrict smoking. (17) Therefore, by relying on a combination of smoking restrictions and taxes, these results are consistent with states using a two-pronged approach to reduce smoking incidence. Third, for those coefficients that are significantly different from zero, we found an increased tendency for smoking restrictions to be adopted in states that have lower metropolitan populations, higher per capita incomes, and a lower presence of tobacco production. (18) This finding was nearly uniform across all six locations. However, except for the significantly negative coefficient of conservatism in the public place regression, we found that political affiliation plays little role in determining whether or not a smoking ban is enacted. (19) Again, since we have not tested nor controlled for potential endogeneity of cigarette consumption and cigarette taxes, these are merely preliminary results. Therefore, the results from Table 3 should be interpreted with caution because we should control for endogeneity.

The results controlling for endogeneity of cigarette consumption and cigarette taxes are provided in Tables 4 and 5. Since estimation of Equation 6 allows us to test for endogeneity, we begin by discussing those results. As highlighted by the values of the likelihood ratio statistics at the bottom of Table 4, the null of no endogeneity of cigarette consumption and taxes can be decisively rejected. Controlling for endogeneity does appear to matter. By using the Rivers and Vuong (1988) procedure, we show in Table 4 that the probability of smoking restrictions being adopted is negatively influenced by per capita cigarette consumption. Also, with few exceptions, we continued to find the probability of adopting a smoking ban to be negatively (positively) related to metropolitan population and tobacco production (per capita income); in addition, for all six locations (although statistically insignificant for government buildings, schools, and health care facilities) more conservative states are less likely to adopt smoking bans. However, what is most interesting is that the role of cigarette taxes is reversed. That is, across all locations (although insignificant for government buildings and health care facilities) the coefficient of cigarette taxes is now negative. Unlike before, cigarette taxes now become a policy substitute for smoking restrictions.

When we correct for endogeneity by using predicted cigarette consumption and cigarette taxes in the estimation of Equation 4, we found results similar to those in Table 4. For example, as illustrated in Table 5, we again found an inverse relationship between the probability of adopting a smoking ban and cigarette consumption, conservatism, metropolitan population, and tobacco production. Also, per capita income continues to positively affect the decision to adopt a smoking ban. Lastly, the role of cigarette taxes remains the same as in Table 4. Therefore, the results without endogeneity correction suggest that cigarette taxes are complementary to smoking restrictions; however, the opposite is the case after we control for endogeneity. Instead, cigarette taxes are a substitute for smoking restrictions. This would be consistent with states viewing cigarette tax revenue as an important factor when deciding whether or not to restrict smoking. (20)

4. Concluding Remarks

In this paper, we used improved econometric techniques and an expanded and more comprehensive data set to show the impact of certain economic, demographic, and fiscal variables on state-level smoking restrictions. We found that the probability of a state adopting a smoking restriction is particularly sensitive to per capita cigarette consumption, cigarette taxes, political affiliation, metropolitan population, per capita income, and tobacco production.

These findings are of particular interest to policymakers for a number of reasons. First, understanding the determinants of state-level regulations (whether applied to tobacco, alcohol, or other products) can facilitate the political decision-making process. As our results show, states with higher per capita cigarette consumption are less likely to adopt smoking bans; therefore advocates of smoking bans in those states should be aware of this when evaluating the benefits and costs of engaging in a campaign to limit smoking. Second, since we show that it is important to acknowledge endogeneity in policy-choice models. For any policymaker concerned with abating the consumption of a product but also interested in maintaining the tax revenue derived from the product, it is important to understand the potential tradeoffs between these policies. Indeed, once we control for endogeneity, we found that smoking restrictions and cigarette taxes are, to some extent, competing against each other.

Appendix

Data Sources

Variables               Data Source

Smoking restrictions    American Lung Association (2000 SLATI)
                        (www.slati.tungusa.org)

Conservatism            American Conservative Union
                        (www.conservative.org)

Percentage of           U.S. Census Bureau (www.census.gov/)
metropolitan
population

Per capita disposable   Economagic.com (www.economagic.com)
income

Percentage of gross     U.S. Department of Agriculture
state product from      (www.ers.usda.gov/)
tobacco

Per capita cigarette    Tax Burden on Tobacco (1998); American
consumption             Lung Association (www.lungusa.org)

Per pack cigarette      Tax Burden on Tobacco (1998); American
tax                     Lung Association (www.lungusa.org)

Price of tobacco        U.S. Department of Agriculture
                        (www.ers.usda.gov/)

Hourly wage of          U.S. Bureau of Labor Statistics
cigarette industry      (www.bls.gov/)
workers

Cigarette advertising   Federal Trade Commission (www.ftc.gov/)
expenditures

Poverty rate            U.S. Census Bureau (www.census.gov/)

Unemployment rate       Economagic.com (www.economagic.com)

The authors wish to thank the co-editor and two anonymous referees for their helpful comments and suggestions.

Received December 2004; accepted November 2005.

References

Arendt, Jacob. 2001. Endogeneity and heterogeneity in LDV panel data models. Unpublished paper, Institute of Local Government Studies, Denmark.

Borland, Ron, Simon Chapman, Neville Owen, and David Hill. 1990. Effects of workplace smoking bans on cigarette consumption. American Journal of Public Health 80:178-80.

Boyes, William J., and Michael L. Marlow. 1996. The public demand for smoking bans. Public Choice 88:57-67.

Butler, J., and Robert Moffitt. 1982. A computationally efficient quadrature procedure for the one-factor multinomial Probit model. Econometrica 50:761-4.

Chaloupka, Frank J. 1992. Clean indoor air laws, addiction, and cigarette smoking. Applied Economics 24:193-205.

Chaloupka, Frank J., and Henry Saffer. 1992. Clean indoor air laws and the demand for cigarettes. Contemporary Economic Policy 10:72-83.

Chamberlain, Gary. 1980. Analysis of covariance with qualitative data. Review of Economic Studies 47:225-38.

Cremieux, Pierre, Pierre Ouellette, and Caroline Pilon. 1999. Health care spending as determinants of health outcomes. Health Economics 8:627-39.

Cremieux, Pierre, and Pierre Ouellette. 2001. Actual and perceived impacts of tobacco regulation on restaurants and firms. Tobacco Control 10:33-7.

Evans, William N., Matthew C. Farrelly, and Edward Montgomery. 1999. Do workplace bans reduce smoking? American Economic Review 89:728-47.

Gallet, Craig. 2003. Advertising and restrictions in the cigarette industry: Evidence of state-by-state variation. Contemporary Economic Policy 21:338-48.

Gallet, Craig. 2004. The efficacy of state-level antismoking laws: Demand and supply considerations. Journal of Economics and Finance 28:404-12.

Glantz, Stanton A., and Lisa R. Smith. 1994. The effect of ordinances requiting smoke-free restaurants on restaurant sales. American Journal of Public Health 84:1081-5.

Hersch, Join, Alison F. Del Rossi, and W. Kip Viscusi. 2004. Voter preferences and state regulation of smoking. Economic Inquiry 42:455-68.

Hitiris, Theo, and John Posnett. 1992. The determinants and effects of health expenditure in developed countries. Journal of Health Economics 11:173-81.

Lee, Lung-Fei. 198l. Simultaneous equation models with discrete and censored dependent variables. In Structural Analysis of Discrete Data with Economic Applications, edited by Charles Manski and Daniel McFadden. Cambridge, MA: MIT Press, pp. 346-64.

Mazare, Ioana R. 2004. State regulations of smoking in public places: Decisions on timing and restrictions. Unpublished paper, Weber State University.

Newey, Whitney K. 1987. Efficient estimation of limited dependent variable models with endogenous explanatory variables. Journal of Econometrics 36:231-50.

Rivers, Douglas, and Quang H. Vuong. 1988. Limited information estimators and exogeneity tests for simultaneous Probit models. Journal of Econometrics 39:347-66.

Tremblay, Carol, and Victor Tremblay. 1995. The impact of cigarette consumption on consumer surplus, profit, and social welfare. Contemporary Economic Policy 13:113-24.

Vella, Francis, and Mamo Verbeek. 1999. Two-step estimation of panel data models with censored endogenous variables and selection bias. Journal of Econometrics 90:239-63.

Wasserman, Jeffrey, Willard G. Manning, Joseph P. Newhouse, and John D. Winkler. 1991. The effects of excise taxes and regulations on cigarette smoking. Journal of Health Economics 10:43-64.

Yurekli, Ayda A., and Ping Zhang. 2000. The impact of clean indoor-air laws and cigarette smuggling on demand for cigarettes: An empirical model. Health Economics 9:159-70.

(1) According to Chaloupka and Saffer (1992), Arizona adopted the first clean indoor-air law in 1973 with the intent of reducing secondhand smoke exposure. In 1974, Connecticut limited smoking in restaurants. Then, in 1975, Minnesota adopted an extensive clean indoor-air law by restricting smoking in several types of locations. Subsequent to these early efforts, other states have adopted their own clean indoor-air laws.

(2) However, Borland et al. (1990) and Evans, Farrelly, and Montgomery (1999) found that bans on smoking in workplaces do reduce smoking rates.

(3) Chaloupka and Saffer (1992) did control for the endogeneity of antismoking laws in the demand for cigarettes but did not include smoking rates as a regressor; they considered a reduced form equation for smoking restrictions.

(4) For example, a fairly typical law restricting smoking in public places would state: "Nonsmoking areas must be designated by signs posted in places of public assembly, including enclosed theaters (except the lobby), opera houses, auditoria, classrooms, elevators, and other enclosed buildings with a seating capacity of 50 or more persons available to the public. Restaurants/food service establishments, bowling alleys and taverns are expressly excluded from this law. Nonsmoking areas shall be designated by the local fire authority or the assigned person in control of a place of public assembly. A nonsmoking area may include the entire place of public assembly."

(5) Some studies have addressed how various components of clean indoor-air laws affect smoking incidence. For example, Gallet (2004) estimated the impact of states adopting smoking bans, coupled with state-level enforcement efforts, on cigarette demand and supply. He found the mere passage of the law is sufficient to reduce cigarette consumption because enforcement efforts fail to significantly affect consumption. In light of this finding, our paper emphasizes what drives a state to adopt a smoking ban at a particular site, not the extent to which the state enforces its smoking bans.

(6) Indeed, Boyes and Marlow (1996) found that smokers and ex-smokers are less in favor of smoking bans in restaurants and bars. Hersch, Del Rossi, and Viscusi (2004) argued that smoking rates may affect the likelihood of a ban being adopted by affecting voter preferences.

(7) Furthermore, Mazare (2004) found that the influence of special interest groups affects the timing and severity of state regulations regarding smoking in public places.

(8) Interestingly, Boyes and Marlow (1996) found a higher percentage of business owners thought smoking bans in restaurants and bars would hurt their business. Nonetheless, several studies (such as those by Glantz and Smith [1994] and Cremieux and Ouellette [2001]) found reality often differs from perception because the actual impact of smoking bans on restaurant and bar revenue is insignificant.

(9) State-specific unobservable factors may include state-specific culture, historical background, and other factors that we could not measure.

(10) The random effects Probit model becomes inconsistent when regressors are correlated with unobserved heterogeneity. In the linear models, the usual Hausman test is often used to compare the random effects estimates with the fixed effects estimates in testing whether regressors are correlated with unobserved heterogeneity. In the panel Probit models, however, this task poses a problem because there is no fixed effects version of the panel Probit model. Adding state dummy variables can lead to the incidental parameter problem, and the resulting estimator is inconsistent. The situation is similar in the panel Logit models because there may be no fair comparison of the fixed effects and random effects estimates in the Hausman test setting. Nonetheless, we obtained and examined several estimation results based on other possible models. These results are generally consistent with the main results from the random effects Probit model, although some disparities do exist. All estimation results are available from the authors upon request.

(11) A reviewer pointed out that cigarette taxes are often driven more by state fiscal matters than by lawmaker preferences. This point is valid and it seems important to consider state fiscal matters as determinants of cigarette taxes. Noting that cigarette taxes and the chance of adopting smoking restrictions are simultaneously affected by certain characteristics of states; therefore, when correcting for endogeneity, we include a few important variables of state fiscal matters in the reduced form equation for cigarette taxes.

(12) For our model, the instrumental variables include the price of tobacco, the average hourly wage of a worker in the cigarette industry, cigarette advertising expenditures, the rate of poverty, and the unemployment rate. With respect to cigarette consumption in [w.sub.it], we expect increases in the tobacco price and cigarette wage to reduce cigarette consumption through reductions in supply. We hold no sign expectation for the effect of poverty and unemployment because these factors depend on the direction of the shift in demand. As for advertising, per capita consumption may increase with more advertising if advertising increases demand. On the other hand, per capita consumption may negatively correlate with advertising if advertising increases market power in the cigarette industry (see Tremblay and Tremblay [1995] and Gallet [2003]) or if cigarette producers redirect their spending away from advertising when consumption is already high. To maintain consistency, we also include these instrumental variables as determinants of cigarette taxes in [w.sub.it]. It is reasonable, for example, that the general fiscal health of a state (proxied by unemployment and poverty) influences tax policy. Also, industry attributes (proxied by costs and advertising) may affect tax policies directed at the cigarette industry.

(13) Because we are testing whether endogeneity exists with respect to cigarette consumption and cigarette taxes, our model includes the residuals from the estimation of Equation 3 for both per capita cigarette consumption and cigarette tax rates. Accordingly, we use a likelihood ratio test to test for the joint significance of the coefficients of these two residuals in Equation 6. One potential issue relates to the need to correct for standard errors when generated regressors are used in the second stage of maximum likelihood estimation and theft coefficients are significant. This task poses a technical difficulty in our models because the Gaussian-Hermite quadrature approximation does not permit us to recover necessary score vectors, and there are two generated regressors in the equation.

(14) We have examined the significance of potential time fixed effects in the reduced form equation in Equation 3. The likelihood ratio statistics for the joint insignificance of the coefficients of the time dummy variables are calculated as 0.913 (p-value = 0.57) and 1.332 (p-value = 0.149) in the reduced form regression for per capita cigarette consumption and cigarette tax rates, respectively. Therefore, we observe that the time fixed effects are not significant in both models; and, therefore, we exclude the time dummy variables in each reduced form model. To be consistent, we also drop time fixed effects in the choice models.

(15) Of course, the signs of the marginal effects are the same as those of the reported coefficients. The significance of the marginal effects can be gauged by their standard errors, which can be calculated by the delta method. However, the computed standard errors are often unreliable in the random effects Probit models and, in some cases, we failed to compute them. Therefore, we omitted them from the relevant tables.

(16) Hersch, Del Rossi, and Viscusi (2004) analyze the determinants of restrictions for one year. Because our results are quite different in that many of the coefficients are significantly different from zero, this suggests that the nature of the data may be important because we use a richer panel data set over a longer period of time.

(17) This is somewhat consistent with Chaioupka and Saffer (1992), who included the price of cigarettes (which includes state and federal taxes) as a determinant of smoking restrictions and found that higher cigarette prices increase the probability of smoking restrictions being adopted.

(18) Chaloupka and Saffer (1992) also found the probability of a smoking restriction being adopted increases with per capita income. Furthermore, Chaloupka and Saffer (1992) and Hersch, Del Rossi, and Viscusi (2004) found tobacco production lowers the likelihood of smoking restrictions.

(19) Hersch, Del Rossi, and Viscusi (2004) also found spotty significance of political affiliation (significance being most pronounced for restaurants and mails), with more conservative states being less likely to favor smoking bans.

(20) Interestingly, controlling for endogeneity also affects the coefficient of metropolitan population. That is, consistent with other locations, when cigarette consumption and taxes are treated as exogenous, the coefficient of metropolitan population is negative for public places. Yet when endogeneity is taken into account, the coefficient of metropolitan population switches to positive for public places.

Craig A. Gallet, * Gary A. Hoover, ([dagger]) and Junsoo Lee ([double dagger])

* Department of Economics, California State University at Sacramento, CA 95819-6082, USA; E-mail [e-mail].

([dagger]) Department of Economics, Finance and Legal Studies, Box 870224, University of Alabama, Tuscaloosa, AL 35487, USA; E-mail [e-mail].

([double dagger]) Department of Economics, Finance and Legal Studies, Box 87204, University of Alabama, Tuscaloosa, AL 35487. USA; E-mail [e-mail]; corresponding author.

Table 1. Descriptive Statistics

                                                      Standard
                                              Mean    Deviation
Smoking restriction locations
  Public place (a)                             0.45     0.50
  Government buildings (a)                     0.42     0.49
  Private work sites (a)                       0.23     0.42
  Schools (a)                                  0.38     0.48
  Health facilities (a)                        0.43     0.50
  Restaurants (a)                              0.32     0.47

Regressors
  Cigarette consumption, packs per capita    106.96     28.35
  Cigarette tax, cents per pack               23.68     16.45
  Conservatism (American Conservative
    Union index)                              48.79     22.14
  Tobacco production (percentage of Gross
    State Product)                             0.06      0.19
  Metropolitan population (percentage of
    total population)                         66.81     20.14
  Per capita income (thousands of dollars)    16.17      5.19

Instrumental variables
  Tobacco price (cents per pound)            170.21      9.44
  Cigarette wage (dollars per hour)           18.79      5.16
  Cigarette advertising (billions of
    dollars)                                   4.18      2.20
  Poverty rate                                13.26      4.11
  Unemployment rate                            6.14      2.19

(a) Equals 1 if smoking is restricted and equals 0 if not.

Table 2. Estimation Results of Reduced Form Regressions (a)

                                     Dependent Variables

Variables                 Cigarette Consumption     Cigarette Tax

Conservatism                0.106 * (3.50)         -0.161 * (-9.42)
Metropolitan population    -0.177 * (-4.59)          -0.014 (-0.62)
Per capita income         -0.697 ** (-1.71)         2.438 * (10.63)
Tobacco production         55.573 * (15.67)       -16.300 * (-8.16)
Tobacco price                 0.095 (1.21)            0.027 (062)
Cigarette wage             -1.807 * (-4.69)        -0.862 * (-3.97)
Cigarette advertising        -1.185 (-1.64)       0.985 *** (2.42)
Poverty rate                -1.12 * (-4.87)       0.318 *** (2.45)
Unemployment rate             0.389 (0.91)            0.180 (0.75)
Constant                  156.984 * (12.40)          -3.978 (-0.56)
# obs                             1050                  1050
F statistic                      101.00                114.17
[R.sup.2]                         0.466                 0.497

(a) t statistics in parentheses are below coefficient estimates.

* Significant at 1% level.

** Significant at 10% level.

*** Significant at 5% level.

Table 3. Estimation Results of Random Effects Probit Model (a)

                                  Dependent Variables

                            Public     Government     Private
Variables                   Places     Buildings    Work Sites

Cigarette consumption      -0.047 *    -0.013 **      -0.008
                           (-4.91)      (-1.65)       (-0.82)
                           [-0.018]     [-0.005]    [-0.67 E-4]
Cigarette taxes            0.280 *     0.040 ***       0.020
                            (7.15)       (2.43)       (1.26)
                           [0.104]      [0.013]     [0.15 E-3]
Conservatism              -0.015 ***     0.009         0.004
                           (-2.04)       (1.39)       (0.49)
                           [-0.006]     [0.003]     [0.28 E-4]
Metropolitan population    -0.049 *     -0.047 *     -0.043 *
                           (-4.28)      (-4.79)       (-3.88)
                           [-0.018]     [-0.016]    [-0.35 E-3]
Per capita income          0.988 *      0.532 *       0.470 *
                            (7.44)      (12.58)       (8.88)
                           [0.368]      [0.176]       [0.004]
Tobacco production          0.151        0.936      -13.951 ***
                            (0.11)       (0.63)       (-2.54)
                           [0.056]      [0.309]      [-0.112]
# obs                        1050         1050         1050
Log likelihood             -147.23      -155.41       -144.47

                                  Dependent Variables

                                         Health
Variables                  Schools     Facilities    Restaurants

Cigarette consumption     -0.018 ***    -0.75 E-4     -0.024 *
                           (-2.22)       (-0.01)       (-2.74)
                           [-0.005]    [-0.21 E-4]    [-0.001]
Cigarette taxes            0.025 **      0.060 *      0.033 ***
                            (1.82)       (3.19)        (2.12)
                           [0.007]       [0.017]       [0.001]
Conservatism                0.006         0.005         0.002
                            (1.05)       (0.76)        (0.30)
                           [0.002]       [0.001]     [0.89 E-4]
Metropolitan population    -0.043 *     -0.038 *     -0.025 ***
                           (-5.20)       (-3.72)       (-2.28)
                           [-0.011]     [-0.011]      [-0.001]
Per capita income          0.517 *       0.496 *       0.455 *
                           (14.28)       (12.02)       (10.22)
                           [0.138]       [0.139]       [0.019]
Tobacco production           1.47       -11.212 *     -13.296 *
                            (1.61)       (-3.46)       (-2.79)
                           [0.392]      [-3.153]      [-0.560]
# obs                        1050         1050          1050
Log likelihood             -200.26       -186.13       -167.36

(a) Not corrected for endogeneity. t statistics in parentheses
are below coefficient estimates, whereas marginal effects are
provided in brackets.

* Significant at 1% level.

** Significant at 10% level.

*** Significant at 5% level.

Table 4. Estimation Results of Random Effects Probit Model (a)

                              Dependent Variables

                     Public        Government     Private
Variables            Places        Buildings    Work Sites

Cigarette             -0.040 *     -0.078 **     -0.060 *
  consumption          (-2.01)      (-3.34)       (-1.98)
                     [-0.25 E-3]    [-0.025]    [0.21 E-3]
Cigarette taxes       -0.23 **       -0.056      -0.199 **
                       (-2.76)      (-0.88)       (-2.61)
                      [-0.002]      [-0.018]    [-0.69 E-3]
Conservatism          -0.057 **    -0.30 E-3     -0.029 *
                       (-3.51)      (-0.03)       (-2.06)
                     [-0.36 E-3]   [0.99 E-4]   [-0.10 E-3]
Metropolitan           0.022 *     -0.047 **     -0.050 **
  population           (2.48)       (-4.54)       (-4.03)
                     [0.13 E-3]     [-0.015]    [-0.17 E-3]
Per capita            1.992 **      0.546 **     0.787 **
  income               (6.71)        (4.12)       (4.49)
                       [0.013]      [0.178]       [0.003]
Tobacco                 1.316         2.23      -17.538 **
  production           (0.62)        (0.94)       (-2.78)
                       [0.008]      [0.726]      [-0.061]
Residual cigarette    0.106 **      0.073 **     0.059 ***
  consumption          (4.18)        (2.93)       (1.75)
                       [0.007]      [0.024]     [0.20 E-3]
Residual cigarette    0.414 **     0.101 ***     0.222 **
  taxes                (4.35)        (1.68)       (3.00)
                       [0.003]      [0.033]     [0.77 E-3]
# obs                   1050          1050         1050
Log likelihood         -151.79      -184.79       -138.31
Likelihood ratio        26.15        10.80         11.04
  endogeneity          (0.00)        (0.01)       (0.00)
  test (p-value)

                            Dependent Variables

                                   Health
Variables             Schools    Facilities   Restaurants

Cigarette            -0.091 **    -0.056 *     -0.063 *
  consumption         (-3.91)     (-2.25)       (-2.31)
                     [-0.022]     [-0.015]     [-0.002]
Cigarette taxes      -0.153 **     -0.075      -0.196 **
                      (-2.59)     (-1.14)       (-2.81)
                     [-0.037]     [-0.020]     [-0.006]
Conservatism          -0.014       -0.012      -0.031 *
                      (-1.31)     (-1.00)       (-2.43)
                     [-0.004]     [-0.003]     [-0.001]
Metropolitan         -0.044 **   -0.039 **     -0.028 *
  population          (-5.04)     (-3.75)       (-2.45)
                     [-0.011]     [-0.010]     [-0.001]
Per capita           0.681 **     0.600 **     0.809 **
  income              (5.33)       (4.43)       (4.98)
                      [0.166]     [0.160]       [0.025]
Tobacco                2.384     -13.488 **   -17.858 **
  production          (1.36)      (-3.49)       (-3.22)
                      [0.582]     [-3.592]     [-0.562]
Residual cigarette   0.075 **     0.062 *        0.043
  consumption         (3.23)       (2.40)       (1.46)
                      [0.018]     [0.017]       [0.001]
Residual cigarette   0.178 **     0.131 *      0.229 **
  taxes               (3.12)       (2.12)       (3.38)
                      [0.044]     [0.035]       [0.007]
# obs                  1050         1050         1050
Log likelihood        -189.82     -181.61       -159.94
Likelihood ratio       19.23        9.90         13.60
  endogeneity         (0.00)       (0.01)       (0.00)
  test (p-value)

(a) With added residuals for endogeneity. t statistics in
parentheses are below coefficient estimates, whereas marginal
effects are provided in brackets.

* Significant at 5% level.

** Significant at 1% level.

*** Significant at 10% level.

Table 5. Estimation Results of Random Effects Probit Model (a)

                               Dependent Variables

                       Public      Government     Private
Variables              Place        Building       Work

Predicted cigarette    -0.104 *     -0.079 *      -0.064 **
consumption           (-5.19)      (-3.44)       (-2.04)
                      [-0.034]     [-0.024]      [-0.99 E-4]

Predicted              -0.299 *     -0.110 ***    -0.222 *
cigarette taxes       (-4.24)      (-1.87)       (-3.04)
                      [-0.098]     [-0.033]       [0.34 E-3]

Conservatism           -0.057 *     -0.004        -0.030 **
                      (-4.40)      (-0.36)       (-2.14)
                      [-0.019]     [-0.0011       [0.47 E-4]

Metropolitan            0.016 **    -0.046*       -0.049 *
population             (2.23)      (-4.54)       (-4.01)
                       [0.005]     [-0.014]      [-0.76 E-4]

Per capita income       1.379 *      0.63 *        0.826 *
                       (7.30)       (5.02)        (4.70)
                       [0.452]      [0.191]       [0.001]

Tobacco production     -5.463 *      0.730       -21.673 *
                      (-2.77)       (0.29)       (-3.25)
                      [-1.790]      [0.222]      [-0.033]

# obs                   1050         1050          1050

Log likelihood        -157.98       -189.03       -139.11

                            Dependent Variables

                                   Health
Variables             Schools     Facility     Restaurant

Predicted cigarette    -0.088 *    -0.069 *     -0.067 **
consumption           (-3.85)     (-2.92)      (-2.44)
                      [-0.020]    [-0.013]     [-0.001]

Predicted              -0.171 *    -0.143 **    -0.222 *
cigarette taxes       (-3.06)     (-2.40)      (-3.40)
                      [-0.039]    [-0.026]     [-0.004]

Conservatism           -0.016      -0.015       -0.034 *
                      (-1.51)     (-1.38)      (-2.70)
                      [-0.004]    [-0.003]     [-0.001]

Metropolitan           -0.040 *    -0.033 *     -0.028 **
population            (-4.63)     (-3.41)      (-2.36)
                      [-0.0091    [-0.006]     [-0.43 E-3]

Per capita income       0.705 *     0.664*       0.831
                       (5.72)      (5.14)       (5.41)
                       [0.160]     [0.123]      [0.013]

Tobacco production      0.918     -19.236 *    -24.617 *
                       (0.51)     (-4.68)      (-4.07)
                       [0.209]    [-3.556]     [-0.384]

# obs                   1050        1050         1050

Log likelihood        -193.56     -186.55      -164.03

(a) With predicted cigarette consumption and taxes for endogeneity.
t statistics in parentheses are below coefficient estimates, whereas
marginal effects are provided in brackets.

* Significant at 1% level.

** Significant at 5% level.

*** Significant at 10% level.