Introduction

The United States currently does not carry any sort of federal smoking ban, having decided it was an issue best left to the individual states to decide. The first statewide smoking ban was enacted by California in 1995 and since then many states have followed, primarily during the mid-2000’s. It is at least common for cigarette usage to be banned in places with obvious health risks such as hospitals, airplanes, and nursing homes across the country, even in states without a state enforced smoking ban. When referring to state smoking bans they cover one of three categories, restaurants, bars, and non-hospitality workplaces. 28 states have decided to enact bans in all three of these categories, while five states have laws covering two of these categories, and four states have laws covering one of these three. Finally, there are 13 states that carry no form of public smoking ban. Note that local municipalities are not restricted from having their own smoking bans by the states that do not carry smoking bans. Texas for example does not carry a state smoking ban but has at least 90 municipalities that ban smoking inside of bars/restaurants including Houston, San Antonio, and Dallas. However, this research is only concerned with states that have statewide smoking bans versus states that do not.

Electronic cigarettes are a growing sector of the “nicotine” market that involve devices that superheat a liquid (Often times a flavored mixture that includes nicotine among other chemicals) into a vapor form that can be inhaled directly into the lungs of the user. They have been advertised as being a “healthier” alternative to cigarettes and even a means to help quit smoking. Surprisingly enough, there are some professionals in the health field that support e-cigarettes from the standpoint of them being a “harm reduction” strategy to combat health effects from the tobacco industry. (Cahn, Stiegel, 12/09/2010) It certainly is not something that is seen as “healthy” for an individual to partake in but in comparison to regular cigarette smoking is comparatively much better for you. Some researchers are still speculative about their ability to help users quit smoking (Dutra, Glantz, 07/2014). E-Cigarettes began gaining popularity among consumers in the late 2000’s, similar to the time period that states began enforcing smoking bans. Smoking bans do not mention e-cigs or other “vaping” type products as they are rather new and have only been banned for indoor usage in a small number of states, 11 to be exact with six of those states being in the northeast. In states with indoor smoking bans, one could viably use them as a means to not leave their favorite bar or restaurant to smoke outside.

This study plans to investigate any differences in e-cigarette usage between states that carry smoking bans and states that do not. The study will use a weighted population to take into effect states that had fewer responses because of population. The study will also hold age constant, as that is a factor that could influence the likelihood of one using e-cigarettes with older demographics being less likely than younger people. This can be seen in research that has catered specifically to adolescent usage (Dutra, Glantz, 07/2014) as well as studies that show increasing adult awareness of the products themselves. (Regan, 01/2013) Since the hypothesis of this study directly relates to cigarette users finding a loophole to state smoking bans, we will be focusing on respondents that are smokers or former smokers. Even former smokers are important in case they have recently quit and are using e-cigs as a “harm reduction” strategy.

Methods

Data

Data from this analysis came from the [2016 Behavioral Risk Factor Surveillance System] (https://www.cdc.gov/brfss/annual_data/annual_2016.html). This is a nationwide survey conducted by the CDC. The variable CURECIG is the key dependent variable in the analysis. CURECIG shows whether or not the respondent is a current e-cig user. This is coded as an 0/1 variable, 0 meaning not a current user, 1 meaning they are a user. This survey was only conducted to those 18 and older so any underage usage of e-cigarettes were not recorded. The mean rate of e-cig usage for the entire survey was 4.26% in the weighted response. Note that this includes those that have not smoked 100 cigarettes in their lifetime, who are being excluded from the data set.

State.ban is the primary binary variable equal to 1 for any state that carries a statewide smoking ban, while a 0 represents states without one. Severity.of.ban is an ordinal variable meant to be used separately from State.ban to elaborate on how extensive the ban is. A 0 corresponds to no ban, 1 meaning they have one of the three types mentioned before in place, 2 meaning two of the three types, and 3 meaning a ban in all three major categories. State smoking ban information came from the following [map] (https://en.wikipedia.org/wiki/List_of_smoking_bans_in_the_United_States). Figures 1 and 2 Shows a map of all states with a smoking ban, as well as the severity of the ban.

Figure 1 States with Smoking Bans

Figure 1 States with Smoking Bans

Figure 2 Key

Figure 2 Key

The BRFSS also recorded the age of the survey taker, showing them in the variable, ageg5yr, which is done in an interval scale. Age was held constant in one of the models to see if there was a difference in usage by age. The hypothesis being that younger age demographics are going to have a higher usage of e-cigs than older age demographics due to it being a new trend with younger generations. The survey also gives us the variable, smoke100. Smoke100 asked respondents if they have smoked at least 100 cigarettes in their life. Those that responded no or didn’t answer were excluded from the data set. This will be used to help focus solely on whether state smoking bans affect current or former smokers’ usage of e-cigarettes as it seems reasonable to think if someone has smoked at least 100 cigarettes they have been a “user” in at least one point in their life.

After eliminating responses with missing data, the analytic sample contains 185,303 observations. All of the descriptive statistics and regressions apply BRFSS sampling weights finalwt due to differences seen later from applying the weights. Table 1 below shows the basic regression done to compare CURECIG with State.ban. This is followed by Table 2 showing CURECIG with severity.of.ban.

smoking.ban meancurecig
No 0.098
Yes 0.088
severity.of.ban meancurecig
No Ban 0.098
One Type 0.100
Two Types 0.095
Full Ban 0.086

Empirical Strategy

For trying to estimate whether smoking bans have an impact on e-cigarette usage, I compare the basic rate of e-cig usage among individuals in states with and without smoking bans. Estimating Model 1, is a simple regression using Ordinary Least Squares (OLS) and then weighted least squares later on to help account for the sampling structure complexity of the data. The null for this study is that there is no difference in e-cig usage in smokers/former smokers between states that have enacted a smoking ban and states that have not. The alternate hypothesis is that there is a difference between the two. Even though I believe the e-cig rate will be higher, a two sided test is more appropriate in case there is a negative effect on e-cig usage with a smoking ban.

Model 1 Simple Regression: State Ban

\[curecig_{i} = \beta_{0} + \beta_{1}state.ban_{i}+\epsilon_{i}\] B1 is the estimated difference in population frequency usage of e-cig usage between those living in states with smoking bans to those living in states without a statewide smoking ban. A positive estimate for B1 would show higher e-cig usage in states with a smoking ban in place. On the other hand, a negative estimate for B1 would show lower e-cig usage in states with a smoking ban in place.

Model 2 Simple Regression: Severity of Ban

\[curecig_{i} = \beta_{0} + severity.of.ban_{i}\theta_1+\epsilon_{i}\]

Model 2 is nearly identical to model 1 but instead of state.ban being a 0/1 variable, it is examined from an ordinal scale perspective since there are different levels of smoking bans depending on the state. These results don’t reveal anything that doesn’t line up with what model 1 provides but does show that there are is not necessarily a uniform state policy when it comes to smoking bans. Since they do provide similar numbers though, I will be elaborating on model 1 for simplicity’s sake when it comes time to deal with the multiple regressions later on.

Since model 1 attributes any difference between the states to being solely because of whether or not they have a smoking ban in place there is a chance of there being a bias in place if there are other things that affect the difference in frequency of e-cig usage between states. When looking at model 1, if a state had a higher response rate from a specific age demographic, that could cause a bias on the model depending on which age group was over or under represented. For example if states with smoking bans had a larger response from a population of middle aged to elderly adults that took the survey, they could have a negative bias towards the model, as one can assume older demographics are much less likely to use e-cigs. This negative bias would cause the estimated difference to be lower in states with a smoking ban than what it actually would be. When comparing average ages of states that have smoking bans and states that do not, there does not appear to be any major differences between them but considering the relationship between age and e-cig use itself could be significant, it is addressed as following.

To help address this possible bias source, I control for age in the third model by holding it constant. Age is an interval scale variable with 13 categories (starting with 18-24, 25-29, 30-34, 35-39, 40-44, 45-49, 50-54, 55-59, 60-64, 65-69, 70-74, 75-79, 80+). The 18-24 age group is omitted for comparison reasons. Those that did not answer, refuse to answer, or did not know were omitted from the data set.

Model 3 Multiple Regression Controlling for Age

\[curecig_{i} = \beta_{0} + \beta_{1}state.ban_{i}+ ageg5yr_{i}\theta_1 + \epsilon_{i}\]

Finally, Model 4 has the variable ageg5yr interact with state.ban for cross interaction analysis. This lets us see both how ageg5yr can affect e-cig usage as well as when ageg5yr is combined with states that have e-cig bans.

Model 4 Multiple Regression Comparing Age with State Ban

\[curecig_{i} = \beta_{0} + \beta_{1}state.ban_{i}+ ageg5yr_{i}\theta_1 + state.ban_{i}*ageg5yr_{i}\theta_2 + \epsilon_{i}\]

Results

Table 3 shows the results of estimating Model 1 by OLS and weighted least squares as well as Model 2’s weighted least squares. Seeing that there is a small difference between the OLS and weighted model, the other estimates made will use the sampling weights. The intercept can be interpreted as the % of e-cig users within the smoker/former smoker population in states without a smoking ban (9.8%). The estimate indicates the probability of e-cig use is 0.9 percentage points lower among persons living in states with a ban relative to those living in states without a ban. The estimate was precisely estimated enough to show a difference. We can either reject the null then or assume the sample was unusual since we got a statistically significant result. When compared to the rate in states without a smoking ban, that the 0.9% point difference was actually a 9.2% decrease in regards to the rate of e-cig users in states without a smoking ban (9.8%). It appears from this simple regression there is a negative impact on e-cig usage in states with a smoking ban in place.

Table 3 Results for Models 1 and 2 Simple Regression
Dependent variable:
as.numeric(curecig)
Model 1: OLS Model 1: Weighted Model 2
(1) (2) (3)
Ban -0.007*** -0.009***
(0.001) (0.001)
One Type 0.003
(0.003)
Two Types -0.003
(0.003)
Full Ban -0.012***
(0.002)
Constant 0.072*** 0.098*** 0.098***
(0.001) (0.001) (0.001)
Observations 185,303 185,303 185,303
R2 0.0001 0.0002 0.0004
Adjusted R2 0.0001 0.0002 0.0004
Residual Std. Error 0.248 (df = 185301) 6.364 (df = 185301) 6.363 (df = 185299)
F Statistic 24.648*** (df = 1; 185301) 37.687*** (df = 1; 185301) 27.104*** (df = 3; 185299)
Note: p<0.1; p<0.05; p<0.01

Table 4 contains the results of estimating Models 3 and 4. By controlling for age in Model 3, the impact from state.ban is fairly close to Model 1’s difference of states with smoking bans (-0.7% points). This is also statistically significant although slightly less so since it is a smaller difference. Even though we can estimate e-cig usage as varying among the different age groups, it appears that as expected, age did not vary much between states that carry smoking bans and states that do not.

Model 4 is shown in column 2 and shows both effects of age and state.ban on likelihood of e-cig usage. There are only large differences in usage within the 25-29 and 50-54 age groups where they are more likely to use e-cigs in states without smoking bans then they are in states that do have them. These two age groups don’t necessarily have much in common from a logical point of view besides the fact that they represent both extremes of the typical work force. With that being said, the age groups between them (as well as 55-59) are higher in states without, just not as high as the differences on the two bookends. It is someone surprising to see the 18-24 age group not follow this trend but considering many of those respondents might not be in the workforce yet or not “settled down” yet (attending school out of their home state, etc.), it makes sense that these general lifestyle differences could affect their numbers. You can also see that in the graph below the hypothesis of younger demographics being more likely to use e-cigs was correct as there is a clear downward slope as the age groups get older.

Conclusion

To reiterate the purpose of this study, I was attempting to see if there was a difference in the usage of e-cigarettes within the population of smokers/former smokers between states that carried state level smoking bans and states that do not. There has not been research done on this topic since e-cigs are a rather new trend. There has however been research done to compare e-cig usage to regular cigarette usage as well as adult awareness of them. I hypothesized that e-cig usage would be higher in states with smoking bans as a “loophole” for smokers. By using cross sectional data from BRFSS, there is actually a statistically significant negative effect from states that do carry smoking bans when it comes to their rates of E-Cigarette usage. Perhaps because those states have less people that use nicotine laced products in general, since the part of the country that largely does not carry smoking laws is the southeast where tobacco is predominantly grown. I held age constant in my second model in case of a possible bias but still found a similar negative effect. It appears this effect comes largely from the working force population as the age groups that yielded generally significant differences in usage came between 25-59.

There are several limitation with this type of research. A problem with this research is that almost all the states that do not have any smoking ban are in the same region of the country, where tobacco is predominantly grown. Tobacco/nicotine in general usage could be very different in these states as well, most likely higher. Another thing and perhaps the most important to note when it comes to comparing smoking bans between states is the issue of states without smoking bans that have widespread municipal bans. These states could skew the data by making a state that appears to be a “no-ban” state actually have numbers closer to a “smoking ban” state. As mentioned before, a state like Texas is listed in this study as having “no ban” but its largest cities and many of its smaller ones carry municipal smoking bans. In an ideal scenario, you could compare e-cig usage within states before and after smoking bans were put into place but since e-cigs have only been a more recent trend and largely after smoking bans were common, this is not possible. For future studies on this, it would be worth looking into how income could affect these variables, as lower levels of income have been associated with higher smoking rates. It might also be worth seeing if there is a difference in e-cig usage by gender, and if there is, to hold that constant as there could be a bias there as well.

Works Cited

Cahn, Zachary, and Michael Siegel. “Electronic cigarettes as a harm reduction strategy for tobacco control: A step forward or a repeat of past mistakes?” Journal of Public Health Policy 32, no. 1 (February 2011): 16-31. Accessed March 01, 2018. https://link.springer.com/article/10.1057/jphp.2010.41#citeas.

Center for Disease Control and Prevention. Behavioral Risk Factor Surveillance System. August 3, 2017. Raw data. U.S.A.

Dutra, Lauren M., ScD, and Stanton A. Glantz, PhD. “Electronic Cigarettes and Conventional Cigarette Use Among US Adolescents A Cross-sectional Study.” JAMA Pediatrics, July 2014, 610-17. Accessed March 01, 2018. https://jamanetwork.com/journals/jamapediatrics/fullarticle/1840772.

Regan, Annette K, Gabbi Promoff, Shanta R Dube, and Rene Arrazola. “Electronic Nicotine Delivery Systems: Adult Use and Awareness of the ‘e-cigarette’ in the USA.” Tobacco Control 22, no. 1 (2013): 19-23. http://www.jstor.org/stable/43289292.

Wikipedia. “List of smoking bans in the United States.” November 6, 2012. Accessed February 22, 2018. https://en.wikipedia.org/wiki/List_of_smoking_bans_in_the_United_States#References.

Wooldridge, Jeffrey M. Introductory Econometrics: A Modern Approach. 6th ed. South-Western College Publisher, 2016.