This originally appeared at https://pubmed.ncbi.nlm.nih.gov/32856949/
Published 2020 Aug 28. doi: 10.1080/15389588.2020.1810246
PMCID: PMC7709737 NIHMSID: NIHMS1648312 PMID: 32856949
Abstract
Objective:
While attention has been given to how legalization of recreational cannabis affects traffic crash rates, there was been limited research on how cannabis affects pedestrians involved in traffic crashes. This study examined the association between cannabis legalization (medical, recreational use, and recreational sales) and fatal motor vehicle crash rates (both pedestrian-involved and total fatal crashes).
Methods:
We used crash data from the Fatality Analysis Reporting System (FARS) to calculate monthly rates of fatal motor vehicle crashes and fatal pedestrian-involved crashes per 100,000 people from 1991 to 2018. Changes in monthly crash rates in three states that had legalized cannabis (Colorado, Washington, and Oregon) were compared to matched control states using segmented regression with autoregressive terms.
Results:
We found no significant differences in pedestrian-involved fatal motor vehicle crashes between legalized cannabis states and control states following medical or recreational cannabis legalization. Washington and Oregon saw immediate decreases in all fatal crashes (−4.15 and −6.60) following medical cannabis legalization. Colorado showed an increase in trend for all fatal crashes after recreational cannabis legalization and the beginning of sales (0.15 and 0.18 monthly fatal crashes per 100,000 people).
Conclusions:
Overall findings do not suggest an elevated risk of total or pedestrian-involved fatal motor vehicle crashes associated with cannabis legalization.
Keywords: crash, FARS, fatality, ITS, drugs
INTRODUCTION
Deaths from motor vehicle crashes remain a major public health problem. In 2018 more than 36,000 people died in motor vehicle crashes (National Center for Statistics and Analysis 2019). Among youth (ages 10–19), fatal motor vehicle crashes are the leading cause of death, accounting for 20% of all deaths in this age group in 2016 (Curtin et al. 2018; Cunningham et al. 2018). Some of these deaths are from pedestrians involved in motor vehicle crashes. Nearly 6,000 pedestrians died in motor vehicle crashes in 2017 (National Center for Statistics and Analysis 2019). Pedestrian-involved crashes are more fatal; pedestrians are 1.5 times more likely to be die when involved in a traffic crash compared to motor vehicle occupants (Beck et al. 2007).
One of the main risk factors for fatal motor vehicle crashes is the use of psychoactive substances, such as alcohol, medications, and illicit drugs (Vingilis & Macdonald 2002; Taylor et al. 2010). While alcohol is well-established in the literature as a significant contributor to fatal motor vehicle crashes, less is known about the effects of other drugs, including cannabis. As more states legalize cannabis, this topic has become particularly relevant. However, it is important to distinguish policies that legalize medicinal cannabis used to treat qualifying medical conditions (e.g., seizures, intractable pain), from recreational cannabis legalization.
There is a growing body of evidence that suggests medical cannabis legalization is associated with traffic fatalities. Mark Anderson et al. (2013) showed an 8 to 11 percent reduction in traffic fatalities in the year following medical cannabis legalization among 19 states. Similarly, Santaella-Tenorio et al. (2017) showed reductions in traffic fatalities among states that had legalized medical cannabis between 1985 and 2014. In contrast, a study by Masten and Guenzburger (2014) found increases in proportion of fatally injured drivers and crash-involved drivers who tested positive for cannabis in several states following medical legalization. However, these increases may have been due to increased screening for the presence of cannabinoids post-mortem (Hall & Weier 2015).
A small number of recent studies have examined how the legalization of recreational cannabis is associated with motor vehicle crashes. Aydelotte et al. (2017) found no change in fatal motor vehicle crash rates three years after recreational cannabis legalization in Washington and Colorado. In contrast, a report from the Insurance Institute for Highway Safety showed a 5.2% higher rate of motor vehicle crashes in Colorado, Washington, and Oregon compared with neighboring states that had not legalized use and sales of cannabis (Monfort 2018). However, this report used only five years of data (2012 to 2016) and did not appear to account for seasonality or serial autocorrelation of crashes. Another study in these same three states using autoregressive integrated moving average (ARIMA) models found that the number of deaths from all traffic crashes increased by an average of one death per million residents after recreational cannabis legalization (Lane & Hall 2019). Because this study measured traffic fatalities rather than fatal crashes, we cannot know whether increases in deaths were attributable to an increase in crashes, the severity of crashes, or other related factors. Lee, Abdel-Aty and Park (2018) examined various changes in cannabis laws (e.g., medical legalization to full legalization, decriminalization and legalized medical to full legalization) and how they related to fatal traffic crashes. Besides a change to medical legalization, all other policy changes were associated with increases in fatal traffic crashes across several states. More research is needed to determine how recreational cannabis legalization affects rates of fatal motor vehicle crashes.
Only one study to date has investigated the effect of cannabis legalization on fatal pedestrian-involved crashes in the United States. A 2017 report from the Governors Highway Safety Association found that pedestrian fatalities in seven states that had legalized recreational cannabis (Alaska, Colorado, Maine, Massachusetts, Nevada, Oregon, and Washington D.C.) showed a collective 16.4% increase in pedestrian fatalities from motor vehicle crashes in the first months of 2017 (compared to the first six months of 2016) (Retting 2017). However, no other studies have examined this relationship further, nor verified these findings with newly available data or more rigorous analytic methods. In addition, no research has looked at effects based on both date of recreational cannabis legalization and the beginning of sales of recreational cannabis.
The purpose of this study is to address these gaps by determining whether legalization of recreational cannabis has had an effect on fatal pedestrian-involved crashes compared to states that have not legalized cannabis. We hypothesize that legalization of recreational cannabis will be associated with an increase in total fatal traffic crashes, but not pedestrian-involved crashes. While prior studies have shown evidence of a positive relationship between recreational cannabis legalization and total fatal motor vehicle crashes, there is no clear evidence to our knowledge that suggests recreational cannabis legalization puts pedestrians at greater risk.
METHODS
Study Design
We examined the relationship between cannabis legalization (medical, recreational use, and recreational sales) and pedestrian fatalities from motor vehicle crashes. This study used a quasi-experimental interrupted time series (ITS) design with control groups (Shadish et al. 2002). ITS designs are useful when randomization is not feasible; they allow for examination of longitudinal data while accounting for secular trends and many typical confounding variables (Biglan et al. 2000; Lopez Bernal et al. 2017). The current analyses included fatal crashes from 1991 to 2018 for a total of 336 repeated monthly measures. We chose this time interval to take advantage of the maximum amount of high-quality data on fatal motor vehicle crashes and provide a long enough time series for our analysis. At the time of this study, three states had legalized both medical and recreational cannabis with enough timepoints post-legalization for sufficient statistical power: Colorado, Washington, and Oregon (policy states). Table 1 displays the dates that each cannabis law went into effect for Colorado, Washington, and Oregon, and lists the states used as comparisons for the analyses.
Table 1.
Dates of Implementation for Cannabis Laws in Policy and Control States
| Medical cannabis | Recreational cannabis | Recreational cannabis sales began | |
|---|---|---|---|
| Policy states | |||
| Colorado | June 1, 2001 | December 10, 2012 | January 1, 2014 |
| Oregon | December 3, 1998 | July 1, 2015 | October 1, 2015 |
| Washington | November 3, 1998 | December 6, 2012 | July 8, 2014 |
| Control states | |||
| Arkansas | Not implemented a | Not legalized | Not legalized |
| Wisconsin | CBD/Low THC | Not legalized | Not legalized |
| Iowa | CBD/Low THC | Not legalized | Not legalized |
| Indiana | CBD/Low THC | Not legalized | Not legalized |
| Kentucky | CBD/Low THC | Not legalized | Not legalized |
aArkansas approved Issue 6 on November 8, 2016 to legalize medical cannabis; however, this was not implemented through licensed sales until May of 2019.
Each state was compared to a state that had not legalized cannabis. Potential control states were limited to those that did not border policy states and had not legalized medical (beyond CBD/low THC laws) or recreational cannabis. Including a control further strengthens the study design by adjusting for other policies or factors that may affect motor vehicle crashes. Controls were matched to policy states based on comparability of fatal crashes at baseline and parallel trends prior to the policy change. In models examining all fatal motor vehicle crashes, Arkansas served as Colorado’s control state, Wisconsin was a control for Washington, and Iowa was a control for Oregon. Different controls were selected for models of pedestrian-involved fatal crashes to match changes in crash trends: Wisconsin served as a control for Colorado, Indiana for Washington, and Kentucky for Oregon.
Data and Measures
Fatal motor vehicle crash rates
Two per-capita measures of fatal motor vehicle crashes were included in our analyses: total fatal motor vehicle crashes, and pedestrian-involved fatal crashes. Both measures were developed from the Fatality Analysis Reporting System (FARS), a database maintained by the National Highway Traffic Safety Administration which tracks fatal motor vehicle crashes for the 50 states and the District of Columbia. Our measure of total fatal motor vehicle crashes was created by summing each crash within state, year, and month to create a monthly count. Pedestrian-involved fatal crashes were any fatal crash that involved at least one pedestrian. These crashes were also summed within state, year, and month to create a monthly count of pedestrian-involved fatal motor vehicle crashes. These two count outcomes were then divided by the state’s population for a given year (using data from the U.S. Census Bureau) to get a rate of fatal crashes per 1,000,000 people. We then subtracted the crash rate of the control state from the crash rate of the respective policy state to obtain the difference in rates between policy and control states, which allowed us to estimate how the change in each policy state differed from the change in its respective control state over the same period of time (Penfold & Zhang 2013).
Cannabis policies
We examined the effects of three cannabis policies in our analyses: legalization of medical cannabis, legalization of recreational cannabis, and start of retail sales of recreational cannabis. Binary variables were created for each, coded 0 before the date the policy was implemented (legalization or sales), and 1 after legalization or when sales had been implemented. For example, the variable for Oregon’s recreational cannabis law was coded as a “0” prior to July 1, 2015, and a “1” thereafter. Dates that did not begin on the first of the month (e.g., Colorado’s recreational cannabis law was implemented on December 10) were treated as if they had begun on the first day of that month. Given these policies went into effect near the beginning of the month, we expected any effects could still be seen for the remainder of that month. A separate model was fit for each of the three policy variables.
Analysis
We used segmented regression with autoregressive terms (Penfold & Zhang 2013). This approach maximizes the benefit of having many repeated observations while accounting for secular trends and seasonality. Including autoregressive terms allowed us to account for serial autocorrelation due to repeated (i.e., monthly) observations within each state. We identified the most parsimonious model for each state and each policy using backward elimination of lags that were not statistically significant at p<0.05. We tested up to 12 lags (one for each month) in order to capture any potential seasonal trends in fatal crashes.
Our outcome variable for each model was the difference in crash rates between a policy state and its control state. For example, we subtracted the monthly number of crashes in Arkansas from the monthly number of crashes in Colorado. Using differenced values as our outcome variables allowed us to take a difference-in-differences approach, estimating how the change in crashes in a policy state compared to the change in the control state over the same time period. The two main predictors were a binary variable modeling the immediate effect of legalization and a sequential variable that captures the continuing effect of legalization (i.e., the change in trend). In total we fit 12 models; one for each type of crash (total fatal traffic crashes and pedestrian-involved fatal traffic crashes) and across three policy states, where each model included two policy changes (either medical legalization and recreational legalization, or medical legalization and legalized sales of recreational). Models were unconditional (i.e., did not use covariate adjustment) for several reasons: confounding was addressed using a long time series to account for any changes within states, policy and control states were matched based on pre-policy trends and intercepts in the outcome, and including covariates in these models would not function to increase comparability between policy and control states (rather, covariates could only adjust for differences within states).
We also conducted a robustness check by fitting models for each state and each policy using data limited to 36 months before and 36 months after a given policy change. There were 18 models for this check in total: one for each outcome, policy state, and cannabis policy. For example, we fit six of these models for Colorado: three using total fatal traffic crashes with medical, recreational, or sales legalization as the predictor variable; and three using pedestrian-involved fatal traffic crashes. All analyses were done using SAS version 9.4.
RESULTS
Results from the 12 segmented regression models are displayed in Table 2 (models incorporating medical and recreational legalization) and Table 3 (medical legalization and legalized sales of recreational). Figure 1 provides an example of one model (for all fatal traffic crashes in Colorado following medical and recreational legalization) plotted to show changes over time, with additional figures available in the Appendices. Estimates for immediate effect show the change in crashes associated with a given cannabis policy change immediately following implementation. The estimates for continuing effects measure how the trend in crashes changes in the months following implementation compared to the trend pre-implementation. Estimates should be interpreted as the difference in the rate of crashes between a policy state (i.e., Colorado, Washington, or Oregon) and its associated control state. Legalization of medical cannabis was associated with a significant immediate decrease in rate of total fatal motor vehicle crashes in Washington (between −3.37 and −3.79) and Oregon (between −6.01 and −6.02) (Figure 1). Colorado did not show a significant immediate decrease in crashes post-legalization, however in the model incorporating medical and recreational legalization there was a significant decrease in continuing effects for total crashes following medical legalization (−0.08). This estimate was not significant in the model incorporating medical legalization and sales, but magnitude and directionality were consistent. For pedestrian-involved crash rates, legalization of medical cannabis was not associated with any significant immediate or continuing effects (Supplementary material Figure 2).
Cannabis Laws and Total Fatal Motor Vehicle Crashes
Table 2.
Population Rates of Fatal Traffic Crashes by Medical and Recreational Legalization (Mean difference1; 95% CI)
| All motor vehicle crashes | Pedestrian-involved crashes | |||
|---|---|---|---|---|
| Immediate effect | Continuing effect | Immediate effect | Continuing effect | |
| Colorado | ||||
| Medical | 2.12 (−3.39, 7.63) | −0.08 (−0.15, −0.01) | 0.23 (−0.85, 1.31) | <0.01 (−0.02, 0.01) |
| Recreational | 3.56 (−2.83, 9.95) | 0.20 (0.07, 0.34) | 0.67 (−0.60, 1.94) | 0.01 (−0.02, 0.03) |
| Washington | ||||
| Medical | −3.37 (−6.46, −0.27) | 0.01 (−0.04, 0.07) | −0.01 (−0.85, 0.84) | −0.01 (−0.02, 0.01) |
| Recreational | 1.38 (−1.98, 4.74) | 0.03 (−0.04, 0.10) | −0.45 (−1.38, 0.48) | 0.01 (−0.01, 0.03) |
| Oregon | ||||
| Medical | −6.01 (−10.09, −1.92) | 0.01 (−0.06, 0.08) | −0.43 (−1.61, 0.76) | <0.01 (−0.02, 0.02) |
| Recreational | 1.47 (−3.65, 6.59) | 0.04 (−0.17, 0.25) | −0.84 (−2.36, 0.68) | 0.02 (−0.05, 0.08) |
1Refers to the beta coefficients that quantify the slope associated with a 1-unit change in the independent variable. For immediate effect, this refers to the change in the dependent variable immediately following the policy change. For continuing effect, this refers to the average change in the dependent variable for every timepoint following the change in policy.
Note: significant effects (p<0.05) are in bold
Table 3.
Population Rates of Fatal Traffic Crashes by Medical Legalization and Beginning of Recreational Sales (Mean difference1; 95% CI)
| All motor vehicle crashes | Pedestrian-involved crashes | |||
|---|---|---|---|---|
| Immediate effect | Continuing effect | Immediate effect | Continuing effect | |
| Colorado | ||||
| Medical | 1.10 (−4.33, 6.53) | −0.06 (−0.13, 0.02) | 0.12 (−0.93, 1.18) | <0.01 (−0.01, 0.01) |
| Sales | 2.50 (−4.26, 9.26) | 0.22 (0.05, 0.39) | 0.49 (−0.86, 1.83) | 0.01 (−0.03, 0.04) |
| Washington | ||||
| Medical | −3.79 (−6.84, −0.74) | 0.02 (−0.03, 0.07) | 0.07 (−0.76, 0.90) | −0.01 (−0.02, 0.01) |
| Sales | −0.14 (−3.82, 3.53) | 0.04 (−0.06, 0.15) | 0.01 (−1.03, 1.05) | 0.01 (−0.02, 0.04) |
| Oregon | ||||
| Medical | −6.02 (−10.10, −1.94) | 0.01 (−0.06, 0.08) | −0.44 (−1.61, 0.74) | <0.01 (−0.02, 0.02) |
| Sales | 1.55 (−4.59, 7.70) | 0.03 (−0.19, 0.26) | −1.36 (−3.21, 0.48) | 0.03 (−0.04, 0.10) |
1Refers to the beta coefficients that quantify the slope associated with a 1-unit change in the independent variable. For immediate effect, this refers to the change in the dependent variable immediately following the policy change. For continuing effect, this refers to the average change in the dependent variable for every timepoint following the change in policy.
Note: significant effects (p<0.05) are in bold
The effect of legalizing recreational cannabis on total fatal traffic crash was significant only in Colorado, with an increase in continuing effects (0.20). Washington and Oregon did not show significant changes in immediate changes or continuing effects for fatal crash rates. For pedestrian-involved fatal crash rates, there were no significant estimates for immediate or continuing effects. Legalizing the sale of recreational cannabis showed one significant association – an increase in continuing effects in Colorado for total fatal motor vehicle crashes (0.22). No other effects were significant.
Models for robustness check
In models limited to 36 months before and 36 months after a given policy change, the only significant estimates were for an immediate reduction in total fatal crashes following medical cannabis legalization in Oregon (−11.17) and an increase in continuing effects for fatal pedestrian-involved crashes following medical cannabis legalization in Washington (0.09) (Table A1). All other estimates for both total and pedestrian-involved fatal traffic crashes were non-significant. Some of these estimates changed in directionality and magnitude while others remained consistent, however there were no overt inconsistencies (e.g., prior significant estimates in one direction becoming significant in the opposite direction.
DISCUSSION
We did not find strong evidence of an association between recreational or medical cannabis legalization and fatal motor vehicle crashes. Only Colorado showed statistically significant changes in crashes associated with legalization of recreational cannabis – an increase in the trend for crashes post-legalization. Legalization of medical cannabis was associated with immediate decreases in Washington and Oregon. For pedestrian-involved crashes, there were no significant associations. When checking the robustness of our findings with data limited to 36 months before and 36 months after a given policy change, nearly all significant estimates became non-significant. One reason for this may be a decrease in power due to fewer time-points. While a portion of estimates changed in magnitude and direction, none were drastic. Overall, these findings do not suggest an elevated risk of motor vehicle crashes associated with cannabis legalization, nor do they suggest an increased risk of pedestrian-involved motor vehicle crashes.
We did not find an increase in fatal pedestrian-involved crashes as was found by the Governors Highway Safety Association report on pedestrian traffic fatalities in 2017. There are several possible reasons for this difference in findings. The 2017 report compared crash data from the first 6 months of 2017 to the first 6 months of 2016. We sought to use a more rigorous model that accounted for trends over time and state-level serial autocorrelation, using more data and incorporating appropriate comparators to account for history bias. In addition, while the 2017 report counted pedestrian deaths from motor vehicle crashes, we examined fatal crashes that involved at least one pedestrian who was killed. A crash that involved several pedestrian deaths would only be counted as a single event in our analyses. It is possible that while crashes did not increase, pedestrian deaths did (e.g., crashes involved an increasing number of pedestrians over time). Our datasets also differed: we used data from the FARS while the report examined preliminary data from the State Highway Safety Offices (FARS data for 2017 were not available at the time of the report) (Retting 2017).
Our findings regarding medical cannabis legalization showed a reduction in fatal motor vehicle crashes immediately following policy implementation with no changes in trend. Colorado differed from Washington and Oregon in that medical cannabis legalization was not associated with a statistically significant decrease in fatal motor vehicle crashes. One reason for this may be a rapid increase in Vehicle Miles Traveled (VMT) in 2001 in the Denver region of Colorado (Report on Traffic Crashes in the Denver Region 2017), the most populous area of the state. The timing of this increase coincided with the date of medical cannabis legalization, introducing a possible mixing of effects between policy and VMT. More driving would mean a greater risk to pedestrians. This is also consistent with a surge in population in the Denver region during that same year.
Notably, both Washington and Oregon showed large immediate decreases in total fatal motor vehicle crashes following legalization of medical cannabis. This finding is consistent with previous studies that have shown decreases in motor vehicle crashes (Monfort 2018; Lee et al. 2018). While estimates in Colorado states were non-significant, directionality suggests legalizing medical cannabis may be related to decreases in fatal motor vehicle crashes.
Only Colorado showed significant changes in fatal motor vehicle crashes following recreational cannabis legalization. Prior studies of crashes and cannabis legalization using FARS data have been mixed. Aydolette et al. (2017) found no significant changes in the rate of fatal motor vehicle crashes per vehicular miles traveled following recreational cannabis legalization in Colorado and Washington. However, a later study showed significant increases after commercial dispensaries began operating (Aydelotte et al. 2019). In contrast, other studies have used the number of cannabis-related fatal crashes as an outcome. Lee et al. (2018) found significant increases in cannabis-related crashes following recreational cannabis legalization in several states. Similarly, a study using FARS data for Hawaii found an increase in the proportion of fatal crashes involving cannabis-positive drivers (Steinemann et al. 2018). While examining the proportion of crashes involving cannabis would be more appropriate, cannabis testing varies considerably by state and driver injury status (e.g., transported to hospital, uninjured). For example, a study using 2013 FARS data found that only 10% of surviving drivers not at fault in a fatal crash were tested for cannabis (Slater et al. 2016). Estimates of cannabis-involved fatal crashes from FARS may be unreliable and ill-advised for studies of cannabis policy impact on cannabis-involved crashes (Romano et al. 2017). We chose all fatal and pedestrian-involved motor vehicle crashes as our outcomes to avoid this issue of measurement error. Future studies should examine whether drivers involved in crashes are using cannabis, and whether other factors – such as distracted driving – may also factor into changes in crashes over time.
Limitations
This study has several limitations. First, although there were 11 states that had legalized recreational cannabis at the time these analyses were conducted, we only included three states. The availability of FARS data coupled with the need to have enough post-policy implementation timepoints did not allow us to examine effects in states that have legalized more recently (e.g., Nevada, Massachusetts, Michigan). For an appropriate interrupted time series design, several timepoints are needed pre- and post-policy implementation to accurately model trends (Biglan et al. 2000). As more years of data become available, more states that have legalized cannabis should be included for greater power and replicability and more years of data included to better model trends over time. At the same time, a greater number of timepoints must be included with caution. While more timepoints increases power and helps to account for confounding, there is also risk of violating the assumption of linearity for ARIMA models when historical trends change substantially (Lopez Bernal et al. 2017). To address this concern, we conducted a robustness check by limiting the number of timepoints before and after each policy change.
Second, we did not directly model whether drivers or pedestrians involved in these fatal crashes were impaired by cannabis. There is a need for better testing for the presence of cannabis among drivers, to determine whether crashes are attributable to cannabis use. Third, control states selected for our models were non-equivalent controls, modeling the counterfactuals for policy states imperfectly despite efforts to match each policy state with an appropriate control. Fourth, there is the possibility of residual confounding or other threats to validity for time series data (e.g., history bias). However, an interrupted time series design is the most effective way of measuring intervention effects outside of randomized controlled trials due to its ability to account for several threats that other observational studies have (Soumerai et al. 2015).
Finally, we did not model heterogeneity of cannabis policies, meaning policies with fewer restrictions were considered equivalent to policies with more restrictions. Some of the variability in our outcome estimates across the three states may be explained by the differences in cannabis policies. Future research may wish to measure restrictiveness of recreational cannabis policies and whether this is associated with effect heterogeneity. In addition, future studies should examine heterogeneity in how policies relate to crashes across urban versus rural areas. Urban areas with greater access to public transit may differ from rural areas with fewer transit options and where a motor vehicle is more essential. Such differences could mean different rates of pedestrian and vehicular traffic, which could affect overall rates of fatal motor vehicle crashes.
Results from this study provide insight into how cannabis policies may relate to motor vehicle crashes with pedestrians, as well as contributing to the growing body of literature around cannabis and total motor vehicle crashes. Our findings may be informative for states that have legalized cannabis use. These states may wish to monitor changes in pedestrian-involved crashes over time, and examine whether changes are attributable to other factors outside of cannabis legalization. To determine the safety of policy changes that affect motor vehicle crashes, consideration must be given not just to the safety of drivers, but also to the safety of pedestrians who may be involved in motor vehicle crashes.
