On Friday I wrote about a new study by Liz Klein of Ohio State University that found no impact on employment in restaurants and bars following the passage of smoking bans in Minnesota cities. While noting the study’s many limitations, I concluded that it was “interesting and credible.” In light of new information I have revised that evaluation downward.
My main criticism of the study was that it tracks changes only at the community level, missing potentially catastrophic impacts on individual businesses. At the time I thought this was an inherent limitation of the data. But as Michael McFadden (whose own site is here) noted in the comments, Klein could have easily made finer distinctions about the bans’ impact. Specifically, her data codes restaurants and freestanding bars separately, so if there was a differential effect on the two kinds of businesses it should be easy to spot. I posed the following question to Klein asking her why she didn’t include such an analysis in her paper:
A commenter on my blog alerted me to the fact that free-standing bars are coded separately from full-service restaurants in the NAICS data. Given that one would predict bars to suffer the brunt of any decline in business following a smoking ban, it would be of interest to see the data from bars separately rather than pooled together with restaurants. If no statistically significant decline appeared, it would make the conclusions of your study even stronger. On the other hand, if bar employment did decline, the change would be obscured by pooling the data with that from full-service restaurants. Why does your paper only publish the pooled data?
Two reasons we looked at data on the overall hospitality industry. We were not interested in looking at bars only because we wanted to make the comparison for the overall hospitality industry, since a) a lot of restaurants derive a substantive portion of their profit from alcohol sales, and b) the policies apply to both. The other reason is more practical – at least one of our communities didn’t code free-standing bars, even though there were bars in that community. Although the coding scheme should be followed as such, the Department of Employment and Economic Development cannot force a community to use the North American Industrial Classification System codes 7224 instead of 7221.
Obviously there’s nothing she can do about the community that doesn’t use the coding, but that doesn’t mean there’s no worth in examining the data in communities that do or trying to expand the sample. I wrote to Klein again asking her if I could see the data. She directed me to the Minnesota DEED website where employment data is publicly available. However this site doesn’t include information about smoking policies, nor does her study reveal the names of the cities included, so there’s no way for me to replicate her data.
Fortunately, at least one previous study has examined the question in depth. Scott Adams (University of Wisconsin-Milwaukee) and Chad Cotti (now at University of Wisconsin Oshkosh) published a 2007 paper that uses data sources very similar to Klein’s and they took the effort to differentiate between bars and restaurants. I recommend reading the entire paper for the full results, but the basic finding is that comprehensive bans decrease bar employment by a little more than 4% while having a neutral or slightly positive effect on restaurant employment.
In the discussion section of the paper, Adams and Cotti explain why this effect is plausible:
Although bars and restaurants are similar industries, there are important reasons why smoking might matter differently to both. One might argue that a restaurant is primarily selling food, with drinks secondary, and environment or atmosphere of lesser concern. Clean air is more conducive to enjoying food, especially among non-smokers who may be more likely to come to a restaurant following a ban. Bars, on the other hand, sell environment and atmosphere first, with perhaps drinks second and food third. Given that a smoking ban fundamentally changes the environment of an establishment, the observed negative impact on drinking establishments is not surprising. Moreover, part of the bar environment is the fellow patrons, which in many cases attract customers to a particular drinking establishment. It is therefore possible that a smoking ban alters the environment for non-smokers, leading them to shy away from bars following a ban. [...] The main point is that there are plausible explanations for the different impacts of smoking regulations on these similar industries.
Thus there are good reasons to be interested in the potential differential impact of smoking bans on bars and restaurants. To summarize with regard to Klein’s study:
1) Existing research based on similar data finds that differences do exist.
2) Her paper examines the question of whether partial bans have a smaller impact on employment than comprehensive bans, the relevant exemptions applying to freestanding bars.
3) Much of her data is conveniently already coded to differentiate between bars and restaurants.
4) A finding that there was no differential impact would strengthen her paper’s conclusion.
Given all of the above, her insistence that only data from bars and restaurants together are of interest simply makes no sense. She is saying, in effect, that this pooled information is more relevant than data exclusively about bars, even though the exemptions in question apply to bars only. The mind boggles to see the logic in this.
To take a broader review, there are several biases in Klein’s paper that should make one skeptical of the conclusion that communities needn’t worry about the economic impact of comprehensive smoking bans:
1) Existing research shows that bars and restaurants are affected differently, yet in her study the data are pooled.
2) Existing research shows that local bans are less harmful than bans imposed statewide, yet her study only examines local ordinances.
3) Her study examines relatively sticky employment numbers rather than data that more fluidly reflects market changes, such as total hours worked or business revenues.
4) The study period is one of economic growth in which bars are most likely able to withstand the shock of a smoking ban.
For all of those reasons, the applicability of her paper to public policy is very limited and other research that examines the differential impact on bars ought to be considered along with it. It also bears repeating that no study of community-level effects will account for individual smoking-oriented businesses that may suffer drastic drops in revenue even while the overall hospitality industry thrives; the rights of those business owners deserve protection.
Regardless of whether one supports or opposes smoking bans, I hope we can all agree that communities considering them should be informed by an unbiased, comprehensive understanding of how bans and exemptions will affect local businesses. They won’t find it looking at Klein’s study alone.