Last week, I was at the Association for Public Policy Analysis and Management annual conference in Boston. This conference is where policy researchers come together to share their research and discuss specific policy interventions. This is the place to be if you want to learn more about specific policy impacts.
Utah’s unique blood alcohol limit policy
In 2017, Utah passed a law that lowered the legal blood alcohol content limit from 0.08 to 0.05. When the law came into effect the following year, Utah became the only state in the country that had a legal limit for driving lower than 0.08.
This is an interesting policy change because it is unclear what the potential impacts of what they might be. It is natural to think that this policy change would reduce drunk driving, but there are two reasons why I’m skeptical.
First, people don’t generally know their own blood alcohol content. Most people make the decision to drive based on heuristics: how many drinks have they had, how long has it been since their last drink, etc. Second, alcohol affects people differently. While being above the legal limit is unsafe for everyone, differences in individual biology could make a car crash more or less likely at the same blood alcohol level.
The one impact this policy could have is to send a signal that drinking and driving enforcement is a policy priority. While I don’t think that people sitting at bars regularly attempt to calculate their blood alcohol content, I do think a policy change like this might make them believe that police enforcement is more likely. It would follow that if the probability of getting arrested increases, the expected value of driving after drinking would decrease, maybe to the point where calling a cab provides better value.
Did 0.05 reduce traffic deaths?
In the aftermath of this policy change, research came out that suggested it was a major improvement for road safety in Utah. One recent paper used a difference-in-differences approach, comparing changes in crash fatalities in Utah with changes in neighboring states. The researchers found reductions in alcohol-related crash fatalities associated with the law.
The basic idea behind difference-in-difference research design is that simply comparing Utah before and after the policy change isn’t enough. Traffic fatalities could change for reasons unrelated to the law. Looking at what happened in other states helps us account for changes that might have occurred independent of the policy change. The key assumption is that without the policy, Utah and the comparison states would have followed similar trends.
However, choosing that comparison group can be difficult. Just because a state borders Utah doesn’t necessarily mean that it provides a good estimate of what would have happened in Utah without the law. Additionally, there is a concern that we only have a sample size of one state that enacted this policy change.
What is a synthetic control?
At the conference, I was listening to a presentation of new research that attempts to better understand this policy change. The researchers revisited this policy using synthetic difference-in-differences, an approach that combines elements of traditional difference-in-differences and “synthetic control” methods.
A synthetic control is a weighted combination of places that did not implement the policy. Instead of choosing one state to represent Utah, researchers combine information from several states to construct a “synthetic Utah.” The weights are chosen to make the comparison resemble Utah’s outcomes before the policy change. Synthetic difference-in-differences also weights the periods before the policy to improve the comparison.
This can be especially useful when only one state implements a policy. There might not be any individual state that provides a convincing comparison, but a combination of states could better approximate its previous trends. Having only one treated state doesn’t make difference-in-differences invalid. It does make the choice of comparison particularly important.
Using data from 2014 through 2023, the researchers found no statistically detectable effect on alcohol-involved traffic fatalities or fatal crashes when compared to the synthetic control they designed. Their results did not support the conclusion that the law reduced these outcomes.
That finding still requires careful interpretation. Failing to detect an effect does not prove that the effect is exactly zero. It also doesn’t tell us whether the law affected outcomes outside the study’s scope, such as crashes where nobody died.
What can policy analysts learn from this?
For policy analysts, this is a useful reminder that our estimates depend on the research behind them. If we wanted to estimate the benefits of adopting a similar law in another state, our choice of effect estimate could substantially change the result.
We shouldn’t automatically prefer a study because it uses a more complicated method or because its conclusions match our expectations. We need to understand whether its comparison is credible, how much uncertainty remains, and whether it measures the outcomes relevant to our question. When studies disagree, understanding why is part of our job.
Strong research includes studies that find no detectable improvement. These findings help us question assumptions and avoid treating a plausible benefit as an established one. Policy analysis relies on researchers doing this careful work, and on analysts being equally careful about how we use it.

