Study: Smoking bans saved countless lives — could they have increased drunk driving?

Alcohol has really bad health consequences, and the US doesn’t take that seriously enough. While America may have taken smoking more seriously, it continues to kill nearly half a million people every year. Alcohol-related deaths number almost 100,000 annually.

A new job market paper by Anne Burton, an economics PhD candidate at Cornell University, looked into how smoking bans at bars and restaurants affected alcohol consumption, violent crime, and fatal drunk-driving crashes. Her most notable finding was a 4 percent increase in fatal drunk-driving crashes associated with the implementation of these bans in areas with high levels of smoking. If accurate, this would be a significant harm policymakers need to consider when designing other policies that attempt to curb the use of harmful substances.

While smoking bans indoors have become pretty ubiquitous, this research may have new relevance as states begin liberalizing marijuana laws and cracking down on e-cigarettes and vaping. Policymakers have to determine if tamping down or easing up on drug restrictions affects the use of other potentially dangerous substances. If legalizing marijuana has the unintended consequence of increasing cigarette usage or cracking down on e-cigarettes pushes those consumers to smoke cigarettes more, that’s a big problem. Burton’s research looks to see if a policy that worked to stop smoking actually could have led people to drink more and led to measurable increases in drunk driving fatalities.

But first — from one drinker to another (probably) — why drinking and smoking are actually very, very bad and why we should care if more of it happens:

From 1999 to 2017, a study by the National Institute on Alcohol Abuse and Alcoholism (NIAAA) tracked nearly a million alcohol-related deaths in the United States. In a 2010 National Bureau of Economic Research paper, researchers systematically reviewed the existing evidence linking alcohol and violent crimes and concluded that the “extensively documented correlations” between alcohol and crime represent “‘true’ causal effects of alcohol use on crime commission.” That’s a big deal. That means the researchers believe it’s not just an association and that some other factor is making people both drink heavily and commit crimes — it’s that alcohol is a big factor in whether some crimes even happen at all.

“The number one substance that is involved in arrests and incarceration is alcohol in the United States,” Keith Humphreys, a drug policy expert and professor of psychiatry at Stanford University, told me. “In terms of the damage — people think of illegal drugs as the drivers of the criminal justice system, [but] none of them come close to alcohol.”

However, as a country, we’ve been overwhelmingly focused on regulating other drugs. My colleague German Lopez wrote a piece titled “Imagine if the media covered alcohol like other drugs.” Here’s my favorite excerpt:

What’s worse, public use of this drug has become widely accepted in some circles. In New Orleans, several men and women in their 20s and 30s shouted that they’re going to get “wasted” — a slang term for coming under the effects of alcohol. Some have even turned drinking alcohol into a game that involves ping pong balls and cups. One popular holiday, St. Patrick’s Day, appears to celebrate the dangerous drug...

No other drug comes close to the staggering fatalities of these two. Illicit fentanyl, which has consumed widespread media attention due to the opioid epidemic in the past few years, was linked to fewer than 30,000 overdose deaths in 2017. And marijuana — another drug that federal law enforcement officials have warned is dangerous — reportedly caused zero overdose deaths in the past few thousand years.

Cigarettes are also extremely bad for health — though we’ve done more as a country to combat their use. They remain the leading cause of preventable death in the US, killing more than 480,000 people every year. The Centers for Disease Control and Prevention adds that “for every person who dies because of smoking, at least 30 people live with a serious smoking-related illness.”

Burton wasn’t the first to study smoking bans and their relationship with alcohol consumption as well as events associated with drinking like drunk driving and violent crime.

Economists Scott Adams and Chad Cotti of the University of Wisconsin Milwaukee and UW Oshkosh, respectively, looked into the relationship between smoking bans and drunk driving between 2000 and 2005. They began their study by noting that “the expected effect of smoking bans on drunk driving is ambiguous” since smokers might choose to go out less, lowering the number of people drinking in total. However, their research indicates that “fatal accidents involving a drunk driver increase by about 13 percent” following the implementation of a smoking ban — an astonishing finding.

Adams and Cotti looked into what might be happening here, and after reviewing case studies they settled on two theories. The first is “cross-border shopping,” where smokers are willing to drive farther to go to a bar in a neighboring jurisdiction that allows smoking, thereby increasing the number of miles driven after drinking. One example they point to is Minneapolis and St. Paul, Minnesota, which is one large metropolitan area split into two counties. Hennepin County, where Minneapolis is, enacted a smoking ban in 2005, after which there was a 12 percent increase in fatal accidents in Ramsey County, which contains St. Paul.

The second theory is that bars are differentiating themselves within jurisdictions with smoking bans by providing outdoor seating or by not enforcing the ban, and that is leading smokers to drive more as well.

Burton’s study, which builds on this research, looked at smoking bans in bars and restaurants from 2004 to 2012. She pulled from the Behavioral Risk Factor Surveillance System (BRFSS) and the Nielsen Consumer Panel to measure alcohol consumption and the location of alcohol consumption by smoking status. She turned to the Uniform Crime Reports (UCR) for information on alcohol-related crimes.

Burton found no statistically significant effect from the bans on whether people remained smokers or on violent crime. However, she did find that for those who drink, the bans increased how much they drank and that in high-prevalence smoking areas, there was a 4 percent increase in fatal drunk-driving crashes but only among high-prevalence smokers. Among all subjects, there was no increase in accidents. This means in the other subgroups, there were slight decreases or no effect on drunk driving. However, Burton told me the 4 percent finding among high-prevalence smokers was the only statistically significant finding.

Burton inferred that the 4 percent observed increase in drinking was likely happening at bars and restaurants because there was no effect or a slight decline in alcohol purchased for home consumption. She said she isn’t very concerned about the increase in drinking observed in her study — it amounts to about one drink a month, and there was no evidence that people were all of a sudden binge drinking or engaging in especially dangerous drinking behavior.

But there are concerns with the drunk driving findings, which other researchers have pointed out.

Before I get into the arguments against Adams, Cotti, and Burton’s findings, I want to make something clear: At the end of the day, the documented health benefits to smoking bans like reductions in secondhand smoke exposure largely outweigh any of the costs, like a small increase in drunk driving in some places. Additionally, there are simple ways to eliminate potential increases in drunk driving.

Following the release of Adams and Cotti’s research (which originally showed an association between smoking bans and fatal drunk driving accidents), several researchers dove into the question in order to test their findings.

NIAAA-funded research in 2013 examined the effect of New York’s and California’s statewide bar and restaurant smoking bans on “alcohol-related car crash fatalities” and found no association. They tested Adams and Cotti’s hypothesis about jurisdiction shopping by looking at communities along the Pennsylvania-New York border but found no effect on drunk driving accidents.

Andrew Hyland, chair of the department of health behavior at Roswell Park Comprehensive Cancer Center who was one of the authors of the 2013 paper, told me that the biggest concern his team had with Adams and Cotti’s findings was that their research did not include counties where they found zero drunk driving fatalities, which could have biased their observed effect upward.

“Zero events is a very important piece of information,” Hyland told me.

Importantly, Hyland’s research only looked at statewide bans. That means it doesn’t rule out the most plausible jurisdiction-shopping story: one where a city has banned smoking but the state has not, so it’s easy to find a bar just a few miles away that lets people indulge in both vices.

As statewide smoking bans have proliferated over the years, this jurisdiction-shopping effect is becoming less plausible, highlighting an easy way to eliminate this concern — passing statewide bans everywhere.

Looking to Burton’s research, concerns with her findings can be boiled down to measurement error and whether she was actually able to isolate the effect of smoking bans.

Self-reported measures of alcohol and cigarette consumption can be unreliable. First, because of something called “social desirability bias,” people will often say what reflects better on them rather than the truth. In this case, smokers ashamed of the habit might report fewer cigarettes smoked than is true; heavy drinkers could do the same with their alcohol intake.

Second, there’s the issue of “recall bias” — it’s actually just really hard to remember exact numbers when it comes to things like this.

“If you add up all the alcohol that Americans say that they drink, you come to the conclusion that roughly half of all alcohol is poured down the drain, because nobody claims it. ... That means people aren’t very good at reporting their drinking,” Humphreys told me.

For tobacco use, researchers have been able to corroborate survey data with biomarkers (which can’t lie to pollsters). But for alcohol, this type of analysis isn’t possible.

“She’s trying to do the best that she can with the data that she has available,” Hyland said of Burton’s research. “The question is, is that sufficient to make a statement that smoke-free policies are causing a 4 percent increase in [drunk driving] fatalities.”

Burton defends her use of survey data here by saying that even though there could be errors in self-reporting, those errors are likely uncorrelated with the implementation of a smoking ban. She believes the implementation of a smoking ban shouldn’t affect how people respond to survey questions about their alcohol consumption and thus the change between those two points in time should still show the effect of smoke-free policies.

Measurement error could also exist in Burton’s reliance on data from the BRFSS and Nielsen. These sources provide data at the individual and household level, respectively. In order to compare county-by-county numbers, Burton used provided “weights” to extrapolate the given information. But the problem is, neither of these data sets is designed to be representative at the county level. So her county-by-county data could be fuzzy.

As with most research, the biggest problem is trying to isolate the cause of the effect being studied. Aren’t jurisdictions that implement smoking bans likely to have implemented other public health measures around the same time?

Burton controls for a few of these, like state blood alcohol concentration limits for driving under the influence and whether the state has a cigarette tax, but says in the future she hopes to go further and include workplace smoking bans and other anti-smoking measures that could be driving the results she finds here.

Notably, Burton did not control for the cost of alcohol or taxes on alcohol in her results. This is troubling since these have a well-documented effects on demand and could be confounding her results.

“I don’t think that small increases in alcohol consumption by themselves are a bad thing,” Burton told me. “The biggest concern is trying to compare the protective effects from secondhand smoke exposure, particularly for bar and restaurant workers ... against the cost of those potential increases in drunk driving fatalities.”

She’s right — and luckily, the small increase in drinking and the ambiguous effect on drunk driving she found are far outweighed by the benefits of smoke-free laws.

“Smoke-free policies in worksites and hospitality venues have been one of the greatest public health successes of the last 25 years,” Hyland told Vox.

While there is still some debate around the potential increase in drunk driving, there is a vast, peer-reviewed, scientific literature around the harms of secondhand smoke inhalation, and around the massive health benefits associated with the sharp decline in smoking in part due to smoke-free policies.

We know that smoking bans have been effective at reducing secondhand smoke exposure. Bans in restaurants, bars, and other hospitality establishments have the added benefit of ensuring that workers are not forced to carry the health costs against their will simply due to their place of employment. Bans have also been effective at reducing smoking and “reducing opportunities to smoke, changing smoking norms, and reducing smoking rates.”

Smoking and exposure to secondhand smoke increases the risk of cardiovascular disease, pulmonary disease, cancer, and death. Research has shown that heart attack admissions “rapidly declined” after the implementation of 100 percent smoke-free laws.

All of this to say that if there was in fact a small increase in fatal drunk driving accidents as a result of these bans, the bans were still worth it.

It can seem callous to think about policy in these terms, but it cannot be understated how much anti-smoking policies have improved health and general well-being. And that benefit far outweighs the potential that there could be some small increase in drunk driving fatalities. Especially since that risk could be curbed with other anti-drunk driving measures and by universalizing anti-smoking laws at bars, restaurants, and other hospitality establishments.

“Reading this paper makes me more confident in the value of smoke-free laws,” said Humphreys. “Because what it shows is there is no effect on violence, there is no effect on drunk driving ... and the claimed increases in drinking amount to a teaspoon of wine a day. So even if I believed that we could measure the alcohol consumption that accurately in big panel studies, which I don’t, I don’t care.”