Cannabis legalization is a hot topic in the United States. Since California adopted a medical cannabis program in 1996, 38 States have legalized medical cannabis and 25 have legalized adult-use recreational cannabis. This policy change has led to a plethora of research on the different externalities of legalization (public health, economic, social, etc.). However, the results of this research tend to be mixed, depending on the outcome.

One persistent feature of prior scholarship on the topic is that cannabis legalization has primarily been measured as a dichotomy, i.e., whether medical and adult use cannabis has been legalized at the state level or not. What this approach misses is the variation in the design of each state’s laws. Importantly for researchers, that variation is directly related to externalities that we care about. For example, home cultivation should relate to changes in emergency room visits by minors.

To leverage this policy variation, our cannabis policy bundles dataset consists of 36 unique characteristics of state medical and recreational cannabis policies, measured from 1994 to 2022. Rather than only measuring variation in the laws that were adopted, this dataset also captures when each policy component was implemented in each state. This is important for research on the effects of legalization, as some policies become effective upon legislative adoption, but others take considerable time to implement. The policy bundles were developed from a theoretical perspective, but confirmatory factor analysis supports the presence of three bundles: pharmaceutical, permissive, and fiscal.

The pharmaceutical policy bundle includes policy characteristics that approach cannabis more like other regulated prescription medicines, with limits on who and how one can obtain the product. The permissive bundle represents state policy regimes focused on the deregulation of cannabis more in line with a typical legal product, such as over-the counter medicines or vitamins. Finally, the fiscal policy bundle reflects policy characteristics related to the state earning revenue from the sale of cannabis.

Creating this dataset was time intensive. It took multiple researchers hours of combing through not only the original legislation, but the websites and policy documents of state regulators, state courts, non-governmental organizations, and print and digital media. All these sources were necessary for capturing the dates of implementation for all the policy components, as well as when some of them were overturned by courts or removed by subsequent legislation or regulation. Each source is documented in the Dataverse for this project (https://doi.org/10.7910/DVN/2SB7ZF).

There are several ways in which our cannabis policy datasets can be used by researchers. The primary approach is to enable a nuanced understanding of the effects of policy design variation on different social, political, economic, or health outcomes. Let us take youth mental health as an example. Studies using either pre- and post-legalization data within states or comparisons between legal and non-legal states have found concerning, though inconsistent, links between legalization and poor youth mental health, including psychosis and suicidality. However, a more nuanced study (with respect to policy design) using the cannabis bundles showed that it is fiscal policies that are associated with increased poor youth mental health. The pharmaceutical bundle, as well as even the permissive bundle, were associated with reductions in poor mental health (https://doi.org/10.1080/10826084.2025.2466208). Similarly, an analysis of traffic crashes and pedestrian deaths found the same concerning outcomes for the fiscal bundle (https://doi.org/10.1111/add.16638).

But this is not the only use of the cannabis policy bundles dataset. Researchers can also re-cluster the bundles or use individual policy components, depending on their research question. For example, one could create a scale of patient-centric regulations and one of business-centric regulations. While the pharmaceutical bundle captures many of the policy components centered on patients, there are components in the permissive bundle that could be relevant to clinical research. Our measurement of both adoption and implementation, including their timing, could also be used to test for anticipatory effects of a policy in the period between adoption and implementation. Lastly, the bundles can be used for state-specific analyses and comparisons.

These are only some possibilities, but researchers can find the bundles described in our related  article (https://www.nature.com/articles/s41597-025-05284-2) and the datasets themselves (https://doi.org/10.7910/DVN/2SB7ZF).