Information Diffusion in the Evaluation of Medical Marijuana Laws’ Impact on Risk Perception and Use

Up to date, 25 US states have legalized medical marijuana use through state legislation, but no consistent information has been provided to policymakers, parents, and the general public to assess whether the passage of medical marijuana laws (MMLs) may increase or reduce the risk of marijuana use. With data from the 2004 to 2014 National Survey on Drug Use and Health (NSDUS), the article “Young people’s more permissive views about marijuana: local impact of state laws or national trend,” published in the August issue of AJPH, documented a progressive decline in the perceived risk of marijuana use among adolescents and young adults living in MML states.¹ However, upon further analysis by controlling unmeasured between-state differences, this MML-related risk disappeared, as observed in a number of other studies also using national survey data.^2–4 On the other hand, studies using hospital data and criminal records consistently indicate a positive association between MMLs and marijuana use.^5,6 Unfortunately, findings from these studies cannot be generalized to the US population because the study samples are not representative.

Despite the strengths of national survey data, using such data to assess MML must consider the process by which laws and regulations affect people. It is well established that information diffusion is a key process for public health laws and regulations to exert their impact. This mechanism becomes more salient in the information era with increased amount and speed of effective information exchange. Just like the effect of cross-contamination in interfering with the evaluation of a behavioral intervention trial, the effect of MML may be masked, to a great extent, by information diffusion. No valid conclusion would be possible without considering this diffusion process if national survey data were used to assess MML.

In MML research practice, it is neither easy to directly measure the diffusion process nor simple to analyze it with designs and statistical methods commonly used in research. Two approaches can be potentially employed to control for diffusion effect without a direct measurement of the diffusion process: individual-based informational correlation and population-based diffusion modeling. In the first approach, the diffusion effect can be modeled through informational correlation by randomly pairing participants in MML states with participants in non-MML states using a method we developed to assess between-participant communications in an intervention trial.⁷ In the second approach, the effect of MMLs can be effectively detected by incorporating a nonlinear component that characterizes the informational correlation.