We conducted a descriptive cross-sectional study by using survey data collected in the 2007 Missouri County-Level Study (CLS). The Missouri Department of Health and Senior Services and the Missouri Foundation for Health conducted the CLS to determine county-level prevalence of behavioral risk factors, chronic diseases, and preventive practices among adults (www.health.mo.gov/data/cls/). Survey administration followed Centers for Disease Control and Prevention (CDC) Behavioral Risk Factor Surveillance System (BRFSS) methods; the survey consisted of items from the BRFSS and CDC Adult Tobacco Survey. Data were collected from February 2007 through April 2008 via telephone interviews of adults aged 18 years or older. The response rate was 60.3% based on response rate formula 2 (RR2) from the American Association of Public Opinion Research (24).

There were 49,513 completed interviews; no information was collected on nonrespondents. Of the 49,513 participants, 30,398 indicated that they were currently employed for wages or self-employed; 23,923 indicated that they worked indoors most of the time, and 23,820 (99.6%) of indoor workers responded to the question assessing secondhand smoke in the workplace. We conducted analyses on these 23,820 participants. Data were weighted to be representative of the Missouri adult noninstitutionalized population based on distributions of age, sex, race, and county of residence from the 2000 census.

We selected the following individual-level characteristics to study: sex, age, race, annual income, education level, and smoking status. A review of studies on the reliability and validity of BRFSS measures found age, sex, race/ethnicity, education, income, and smoking status to have high reliability and moderate to high validity (25). Although occupation has been identified as a factor in secondhand smoke exposure, it was not assessed in the CLS. Age was measured in years. Race was identified by 4 categories: non-Hispanic white, non-Hispanic black, Hispanic, or other. Annual income was measured in 6 increments, starting with less than $15,000 as the lowest category and $75,000 or more as the highest category. Income was treated as continuous in the models. Education was categorized into less than high school graduate, high school graduate or General Educational Development certificate, some college, or college graduate. Smoking status was collapsed from current, former, and never smoker status into current smoker and nonsmoker. The county-level predictor was rurality, and was determined by the Rural-Urban Continuum Codes (RUCC) (www.ers.usda.gov/briefing/rurality/ruralurbcon/), which we classified into 4 categories (urban, large rural, small rural, or isolated).

Dependent variables were workplace secondhand smoke exposure and support for a smokefree workplace law. The workplace secondhand smoke exposure variable was measured by determining how many participants answered yes to both of the questions "While working at your job, are you indoors most of the time?" and "As far as you know, in the past 7 days, has anyone smoked in your work area?" Support for a smokefree workplace law was measured by response to the question "Some cities and towns are considering laws that would make workplaces smokefree by prohibiting smoking in all indoor workplaces, including restaurants, bars and casinos. Would you support such a law in your community?" Even though these questions are specific to support of smokefree workplaces laws at the local level, support for local laws can increase support and demand for a statewide law by increasing awareness, demonstrating the ease of implementation, and changing social norms (23,26). Perceptions of the health effects of secondhand smoke was measured by response to the item, "Do you think that breathing smoke from other people's cigarettes is: 1-Very harmful to one's health; 2-Somewhat harmful to one's health; 3-Not very harmful to one's health; and 4-Not harmful at all to one's health." The variable was collapsed into a binary variable where yes represented responses 1 and 2 and no represented responses 3 and 4.

People with missing data for any of the independent variables were not included in the final model. We used HLM 6.08 for Windows (Scientific Software International, Inc, Lincolnwood, Illinois) for multilevel modeling and SPSS 17.0.1 (SPSS, Inc, Chicago, Illinois) for all other analyses.

We employed bivariate statistics to describe the sample and examine differences in workplace secondhand smoke exposure related to geographic and demographic characteristics. To determine whether a multilevel modeling strategy was appropriate, we used 2 strategies to test the heterogeneity of proportions, that is, does workplace secondhand smoke exposure vary significantly by county? First, we calculated an intraclass correlation coefficient (ICC [ρ]) specifically developed for binary outcomes (27) between workplace secondhand smoke exposure and county of residence. Second, we conducted a χ² test of county and workplace secondhand smoke exposure (28). In addition to examining the significance of the χ² statistic, we examined the standardized residuals to determine any significant differences. Standardized residuals with a magnitude greater than 2 are major contributors to significant χ² results; they indicate that the observed frequency was much higher than expected, and standardized residuals less than −2 indicate that the observed frequency was much lower than expected.

Although we found a low ICC value (ρ_workplace = 0.014), the χ² analyses found significant differences across Missouri counties for exposure to secondhand smoke in the workplace (χ²(114) = 334.4; P < .001). Because of this heterogeneity of proportions, we used multilevel modeling to examine secondhand smoke exposure in Missouri, accounting for individual and county-level differences. Specifically, we developed multilevel logistic regression random intercepts models consisting of the following: 1) null model, 2) individual-level characteristics (model 1), and 3) model 1 plus county-level characteristics (model 2). We assessed measures of model significance (χ2) and model fit (Akaike Information Criterion [AIC], deviance) and compared models by using the likelihood ratio (LR) test. AIC and deviance are used with nested models to quantify the lack of fit in 1 model relative to the other; lower AIC values represent better fitting models with less lack of fit. LR tests are used with nested logistic models to determine which model is a better fit. The LR statistic is the difference between the measures of deviance (lack of fit) of 2 models, and it follows a χ² distribution with degrees of freedom equal to the difference between the parameters in the 2 models (29). A significant LR test indicates that the larger model is a better fit. As a demonstration, we calculated the probability of secondhand smoke exposure in specific groups of Missourians by substituting values representing the groups into our final model.

Finally, to determine the feasibility of adopting and implementing a statewide smokefree workplace law in Missouri, we conducted bivariate analyses to examine support for a workplace law across demographic and geographic groups. Where there were significant differences, we examined standardized residuals to determine which groups were significantly different in terms of their support for a smokefree law.

Of the 23,820 indoor workers who responded to the question assessing secondhand smoke in the workplace, 2,740 respondents (11.5%) reported being exposed to secondhand smoke at work in the last week. Extrapolated to more than 2 million Missourians employed in indoor settings, this finding indicates that approximately 236,000 Missourians are exposed to secondhand smoke in their workplaces. Demographic characteristics of those exposed and not exposed to secondhand smoke at work are shown in Table 1, and a few of the notable differences are demonstrated in Figure 1. Among those exposed to secondhand smoke in the workplace, 60.9% were nonsmokers, and 88.9% believed secondhand smoke exposure was harmful to health; however, 62.0% declared that they were not in favor of a local smokefree workplace law. Close to half (41.2%) of those not in favor were current smokers (data not shown). There was a significant association between smoking status and support for a local workplace law among those who were exposed (χ²(1) = 419.7; P < .001); 13.7% of current smokers and 53.7% of nonsmokers supported a workplace law. Support for a local workplace law was high (60.3%) among people who had not been exposed to secondhand smoke at work.

Exposure to secondhand smoke varied in counties across the state. A few of the counties with lower than expected secondhand smoke exposure (standardized residuals less than −2) have communities that enacted comprehensive smokefree workplace ordinances shortly before the CLS data were collected (Figure 2). All of the counties with lower than expected secondhand smoke exposure had high population densities and were clustered around the 3 largest metropolitan areas, Kansas City, Columbia, and Saint Louis. Counties with higher than expected secondhand smoke exposure rates were scattered throughout the state, but were particularly concentrated in the southeast boot-heel region, which is mostly rural.

All 3 models demonstrated statistical significance. A comparison of fit indices (deviance and AIC) indicated that the model including individual and county predictors was the best fit. LR tests comparing model 1 to the null model and comparing model 2 to model 1 yielded significant results. Given the measures of fit and significant LR test (χ²(3) = 18.0; P = .001), model 2 was adopted as the final model (Table 2). Age, sex, income, education, smoking status, and rurality were all predictors of the probability of secondhand smoke exposure. Race, however, was not. Living in a small rural or isolated area significantly increased the likelihood of exposure to secondhand smoke at work compared with living in an urban setting. However, living in a large rural area was not significantly different than living in an urban setting. Increased age and increased income were associated with a reduced likelihood of secondhand smoke exposure in the workplace. Male sex, lower levels of education, and current smoking were significantly associated with increased secondhand smoke exposure. The more rural the county a person lived in and the less education they had, the more likely they were to be exposed to secondhand smoke at work (Figure 3).

Overall exposure to secondhand smoke at work across Missouri is 11.5%. However, among young white men with low incomes and limited education living in small rural areas, 40% of nonsmokers and 56% of smokers may be exposed to secondhand smoke at work. On the basis of our model, the highest exposure category is smokers who were young black men making less than $15,000 a year with less than a high school education and living in isolated areas. This group has a 66% chance of being exposed to secondhand smoke in the workplace. This same group with nonsmoking status would have a 50% chance of exposure based on our model.

Among employed Missourians, there were significant differences in the proportion of respondents supporting a smokefree workplace law by sex (χ²(1) = 113.3; P < .001), education level (χ²(3) = 528.2; P < .001), income category (χ²(5) = 78.3; P < .001), race (χ²(3) = 41.5; P < .001), exposure to secondhand smoke at work (χ²(1) = 468.9; P < .001), and smoking status (χ²(1) = 3,426.2; P < .001). There were no significant differences (P > .05) in support for a local smokefree law by age or rurality. Overall, 58% of Missourians support a local smokefree workplace law, regardless of exposure and smoking status.

Residual analyses indicated that, although our multilevel model showed that the following groups were more likely to be exposed to secondhand smoke, they were also significantly less likely than expected (standardized residuals less than −2) to support a local smokefree workplace law: men, smokers, people making less than $75,000 per year, and people with less than a college education. Respondents exposed to secondhand smoke in the workplace were also significantly less likely than expected to support local smokefree workplace laws. Conversely, many of those groups who were already protected (ie, women, college graduates, people with annual incomes of more than $75,000, people not exposed to secondhand smoke at work, and nonsmokers) were more likely than expected to indicate support for a local smokefree workplace laws. Race was not a significant predictor of secondhand smoke exposure in the multilevel model; however, there were a few notable differences in the bivariate analyses. Specifically, Hispanic and other race Missourians were more likely than expected to support local smokefree workplace laws.

CDC considers people of low socioeconomic status and African American, Hispanic/Latino, Asian American/Pacific Islander/Native Hawaiian, American Indian/Alaska Native, and lesbian/gay/bisexual/transgender populations as facing tobacco-related disparities. The largest limitation of this study is the exclusion of the last 3 of these populations. In addition, the sample of Missourians analyzed here did not include residents with wireless telephone numbers only (30) or no telephone service at all (31). Recent studies have shown some demographic and health-related differences between people who use only a cellular telephone and those who use a land-based telephone line. People not reachable during the calling hours may represent important segments of the workforce, such as the service industry who are exposed to secondhand smoke. Because the location of the participant's employment was not collected, we were also unable to account for the existence of local smokefree laws in our analysis. However, given the very small number of cities (n = 4) that had already adopted comprehensive smokefree workplace laws at the time of data collection, we do not think that this omission affected our results. Although the data were weighted to be representative of the general population, the distribution of income was skewed toward higher income levels, limiting generalizability. However, because exposure to secondhand smoke is more common in lower income levels, we believe this skew may have precipitated an underestimation of the number of Missourians exposed to secondhand smoke at work.

Although comprehensive smokefree workplaces policies adopted in some Missouri communities appear to have decreased secondhand smoke exposure on a local level, many areas in Missouri have been reluctant to adopt comprehensive smokefree workplace laws, allowing disparities to persist throughout the state. For example, St. Louis County implemented a smokefree workplace law on January 2, 2011 (32), but the law is not comprehensive. Exemptions leave those who work in small bars, casinos, hotels and motels, and several other venues at risk for workplace secondhand smoke exposure. Statewide comprehensive law is the only method that will ensure that those most at risk in Missouri are protected (33-35) and, on the basis of prior research, will reduce the overall rate of workplace secondhand smoke exposure by more than half (19). This reduction would provide an additional 133,000 Missourians with smokefree workplaces. Given the evidence related to exposure rates and disparities and the support for a smokefree law among Missourians, it may be a good time for policy advocates to push for a statewide comprehensive smokefree workplace law.