The legalization of recreational cannabis in Washington state (I-502) and Colorado (A-64) created a natural experiment with ancillary unknowns. Of these unknowns, one of the more heavily debated is that of the potential effects on public health and safety. Specific to public safety, advocates of legalization expected improvements in police effectiveness through the reduction in police time and attention to cannabis offenses, thus allowing them to reallocate resources to more serious offenses. Using 2010 to 2015 Uniform Crime Reports data, the research undertakes interrupted time-series analysis on the offenses known to be cleared by arrest to create monthly counts of violent and property crime clearance rate as well as disaggregated counts by crime type. Findings suggest no negative effects of legalization on crime clearance rates. Moreover, evidence suggests some crime clearance rates have improved. Our findings suggest legalization has resulted in improvements in some clearance rates.
Proponents of marijuana legalization assert that legalization will allow the police to reallocate resources away from possession arrests to the prevention of property and violent crimes (Trilling, 2016). This “resource reallocation,” they argue, will improve the effectiveness and efficiency of police operations. In fact, legalization proponents made this argument in every one of the 12 states where citizens voted directly on marijuana legalization ballot measures predicting that legalization would improve clearance rates.
However, despite the widespread use of this argument, little research exists showing the relationship between the legalization of either recreational or medical marijuana and the ability of agencies to reallocate resources. As an example, over the last decade, many municipalities have passed city ordinances and implemented initiatives mandating that police agencies treat minor marijuana possession as a low-priority offense (Ross & Walker, 2016). However, as Ross and Walker (2016) demonstrate with their research on deprioritization, there is limited understanding of how police prioritization has influenced police outcomes, specifically with respect to clearance rates.
As Cullen (2016) highlights in his broader discussion of the relationship between resource allocations and crime, there is much we do not know. Traditional research on resource allocation and crime has primarily concerned case-level management (i.e., specialized units and deployment of officers), organizational factors (agency size and centralization or decentralization), and contextual or environmental factors (Doerner & Doerner, 2012). Yet, there is also evidence that legal changes and political decisions can affect police outcomes (White, 2003), in addition to standard case and organizational explanations. However, few studies have been able to examine the relationship between substantial policy changes and the potential for agencies to reallocate resources. In fact, the most analogous policy change commensurate with the legalization of recreational marijuana would be the repeal of alcohol prohibition in 1933. While research has examined the relationship between marijuana deprioritization mandates and resource reallocations, measured by way of changes in clearance rates, these studies are often limited in scope to a specific city.
Given the arguments that legalization would result in resource redistribution, and the substantial reduction in number of marijuana-related arrests witnessed each year in Washington and Colorado, we undertake an interrupted time-series analysis of state-level crime clearance data in Washington and Colorado to determine if, and how, legalization of marijuana influenced clearance rates. We start by summarizing the context for Initiative 502 (I-502) and Amendment 64 (A-64)—the ballot measures that legalized retail sales and recreational use of marijuana for adults in November 2012 in Washington and Colorado, and the theoretical explanations and existing research suggesting why this legal change would be associated with changes in clearance rates.
I-502 and A-64
In November 2012, voters in Colorado and Washington authorized the retail sale of recreational marijuana. Specific to police practice, these ballot measures included the following language “in the interest of the efficient use of law enforcement resources . . . ” (A-64 which added Section 16 to Article XVIII of the constitution of the state of Colorado) and “allows law enforcement resources to be focused on violent and property crimes,” (I-502 in Washington); these measures legalized the possession, production, and retail sale of marijuana in both states. After passage of A-64 and I-502, it has not been a crime for adults aged 21 or older to purchase or possess an ounce or less of marijuana (or 16 ounces of marijuana-infused solids/72 ounces of marijuana-infused liquids [R.C.W. 69.50.360(3)] in Washington) or to grow up to six plants for personal use (in Colorado).
As commonplace as marijuana sales have become, legalization in both states is strictly limited. It is still illegal to possess marijuana if you are under 21, to use it in a public space, to transport it unsealed in your car or across state lines, to send it through the mail, and to drive while under the influence of marijuana may result in a DUI. The two states differ on how much cannabis can be in one’s possession, and Colorado allows those residents over 21 to grow up to six plants for recreational use while Washington state does not permit home growth (with an exception for medical marijuana). In sum, while the aspects of legalization differ between Colorado and Washington, the result is the same—consumption and possession of marijuana is now, in many instances, legal.
According to available police statistics, the liberalization of marijuana availability in Colorado and Washington coincided with declines in arrest rates for its possession. For the sake of consistency despite police agencies’ lack of diligence in reporting arrest statistics (Maltz, 1999), we examined aggregate year-to-year changes in arrest rates only for “zero-population” police agencies that reported drug arrests in each of the last 12 years (2004–2015) to the Uniform Crime Reports (UCR; Federal Bureau of Investigation, 2009a, 2009b, 2009c, 2010a, 2010b, 2011, 2012, 2013, 2014, 2015, 2016, 2017).
In Washington, these records include 131 city and county agencies which cover slightly more than half of the population (57.8%) and, unfortunately, exclude many of the larger agencies in the state (e.g., sheriff’s offices in King, Snohomish, and Spokane counties and city police in Seattle and Spokane). With the exception of a few larger city agencies (i.e., Boulder and Pueblo), the Colorado data are much more complete with 86 agencies (holding jurisdiction over 79.6% of the state’s population) fully reporting drug arrests over those 12 years. We also compared Washington and Colorado’s rates with the aggregate annual rates for states outside of the Pacific census region, so as to have a rough indication of trends in marijuana possession arrests in places where the criminalization of marijuana persisted; these rates account for arrests by police agencies serving 56.6% of the population in the region.
For the police agencies that reliably reported drug arrests to the UCR, Figure 1 indicates that (a) the marijuana possession arrest trends in Washington and Colorado were quite different than what occurred across the aggregate of non-Pacific census region states and (b) the largest declines in arrests occurred when marijuana was legalized. As might be expected given the growth of the medical marijuana market in both Colorado and Washington, decreases in marijuana possession arrests began well before the drug was legalized. Whereas the aggregated rate for the non-Pacific region states was essentially unchanged in the 9 years prior to legalization (2004–2012), over that period Washington and Colorado recorded declines in marijuana possession arrest rates of 25.4% and 21.6%, respectively.
These declines, however, pale in comparison to the dramatic decreases in arrests for marijuana possession in Washington and Colorado following legalization in late 2012. In Washington, the rate change of 71 fewer arrests per 100,000 population in the year after legalization (2013) was 5.5 times greater than the 2004 to 2015 trend of 12.9 fewer arrests per 100,000 population per year. Similarly, the single-year decline following legalization in Colorado of 84.4 fewer arrests per 100,000 population was 6 times greater than the 2004 to 2015 trend of 14 fewer arrests per 100,000 population per year. Seen another way, the 71 fewer arrests per 100,000 population in the year after legalization was almost two thirds (63.7%) of the overall decline of 111.5 fewer arrests per 100,000 population that occurred between 2004 and 2015.
This overall decline in arrests for marijuana possession represents a considerable change in police activity following the multiple statutory changes enacted in Colorado and Washington over the past decade, and the most substantial declines in arrests were concomitant with the legalization of marijuana possession. If we can assume that the police had not somehow become less proficient in making marijuana possession arrests when attempting to do so, then by extension we would expect that they have redirected their efforts toward resolving other violations of the law. Considering the pace of the decline in arrests for marijuana possession, it appears that the scaling back of enforcement was greatest after the legalization of possession and that the decline was a continuation of a trend in the reduction of enforcement corresponding with the earlier liberalization of marijuana controls. Clearly, police agencies have experienced a substantial shift in marijuana cases. Using the estimate provided by Warburton, May, and Hough (2004) that police officers spend an average of 5 hours on cannabis offenses, that shift would have allowed an agency to allocate those hours to other activities. While we do not take one estimate as proof of the substantial investment in time, it is clear that after legalization officers were able to focus on other activities. Proponents of marijuana legalization argued that this would directly translate into improved clearance rates. As such, we present the existing research on factors influencing clearance rates to explore the theoretical foundation for the relationship between legalization, resource reallocations, and clearance rates.
In broad terms, existing research suggests that a wide range of factors, falling into three categories, influence clearance rates: case-level factors, environmental or contextual factors, and organizational factors (Doerner & Doerner, 2012). For the purposes of this study, we focus on organizational factors. This aligns with the marijuana law reform proponents’ assertion that legalization of recreational marijuana would produce clear benefits to the police organization, and one of the most asserted benefits is that it would improve clearance rates for violent and property crime.
Research on organizational factors primarily includes agency size, workload, organizational structures (centralization or decentralization of duties and specified units), number of detectives, depth of training, and management style; these all have been examined with respect to their influence on police outcomes (Cordner, 1989; Geberth, 1996; Greenwood, Chaiken, & Petersilia, 1977; Jang et al., 2008; Sanders, 1977). For example, a well-known and widely read study conducted by RAND (Greenwood, Chaiken, Petersilia, & Prusoff, 1975) found that increases in the number of detectives, enhanced training, reduced workload, and the change of management practices were not associated with higher clearance rates. Such findings are at odds with the more recent “effort-result hypothesis,” which holds that more focused police effort will lead to an increase in the crime clearance rate (Braga & Dusseault, 2018, p. 5).
Notwithstanding the results of RAND’s study, some evidence does exist to suggest that the extent of investigative efforts made does affect clearance rates. For instance, Greenwood et al. (1977) noted that investigation quality is positively associated with homicide clearance rates. Similarly, after examination of 798 homicide cases, Welford and Cronin (1999) found that high-quality police initial investigations and provision of sufficient follow-up resources do, in combination, increase the likelihood clearance of major crimes.
Regarding workload, scholars have reasoned that reduced workload would increase crime clearance rates because the police have more time and resources that can be used in solving open cases (Bayley, 1994). This argument has been supported empirically by several research studies (e.g., Chaiken, 1975; Cordner, 1989; Jang et al., 2008). For example, Cordner (1989) found that a heavy workload (as measured by index crimes per sworn officer) was negatively associated with clearance rates for property crimes. Similarly, Jang et al. (2008) also detected a negative relationship between heavy workloads and clearance rates for property offenses. It is worth noting the effects of workload on clearance rates may vary across different types of crime with respect to their perceived seriousness. As Paré et al. (2007) have argued, in the face of a heavy caseload, the police may focus on crimes that are more serious and “screen out” those viewed as minor, a factor that may lead to different results in clearance rates for crimes of varying levels of perceived seriousness.
The legalization of marijuana undoubtedly resulted in the opportunity for agencies to reallocate resources, and as mentioned earlier, the level of resources available in police agencies is one important organizational factor that may influence clearance rates. Although some research finds little evidence of a resources-to-clearance rate connection (Cloninger & Sartorius, 1979; Greenwood et al., 1975), considerable evidence does exist in the research literature that resource availability does make a difference. For example, Stolzenberg, D'Alessio, and Eitle (2004) found that an increase in the number of police officers is associated with improved clearance rates for violent crime. In one of the earliest studies undertaken in this area, Chaiken (1975) found that officers’ effectiveness in solving crime improved with the increase in departmental resources used for criminal investigation. Similarly, a more recent study conducted by Wong (2010) revealed that resources available—as measured by police expenditures—are positively associated with the clearance capability of the police.
These studies demonstrate that police resources do matter in the provision of public safety outcomes. However, the likelihood of clearance of a crime is contingent on the availability of policing resources devoted to investigation, including the ability to actively search for evidence and to spend time on the development of leads (Benson, Rasmussen, & Kim, 1998; Borg, Parker, & Karen, 2001). Indeed, as Cooney (1994) has noted, police resources typically are not evenly distributed across cases even within the same type of crime.
Although there are studies examining the effects of police resources on clearance rates, there are few studies specifically examining the impact of major policy changes which provide the opportunity for substantive shifts in the ability to reallocate police resources. I-502 and A- 64 represent such major policy change, allowing police agencies a profound ability to allocate resources to other areas. Legalization prevented police from making formerly commonplace arrests, allowing arrest only under very narrow conditions. In the analyses described later, we explore what, if any, influence the legalization of marijuana in Washington and Colorado had on police clearance rates.
Methods and Data
We use multigroup interrupted time-series modeling to examine the short-term effects of legalization on clearance rates in Washington and Colorado. In the absence of a true experimental design, interrupted time series have long been regarded as a strong quasiexperimental alternative (Campbell, 1969; Cook, Campbell, & Shadish, 2002). Whereas true experiments use randomization to create comparison groups, interrupted time-series designs use change in trends to induce comparative logic. Specifically, interrupted time-series modeling compares trends in some process or outcome change before and after some demonstrable intervention point (the so-called interruption).
Applied to the study of marijuana legalization in Washington and Colorado, our focus is on whether trends in crime clearance rates changed following the intervention of I-502 and A-64, that is, did these policies create an interruption and change clearance rates for an array of violent and property crimes? Moreover, we make use of a multiple group comparisons by examining the trends in clearance rates for Washington and Colorado, as compared with the rest of the country during the same time periods. Given that most environmental factors that we might also expect to be associated with clearance are slow to change, this interrupted time-series approach uses pre- and postintervention trends as comparison groups under the logic that significant shifts in the trend are much more likely to be the result of the intervention than of other slower, more gradual changes taking place in both states.
Although marijuana possession, pre I-502 and A-64, was a misdemeanor in the majority of cases, proponents of legalization suggested that legalization frees the police to focus their attention on more serious crimes, thereby resulting in increases in crime clearance rates. Our models examine the degree to which clearance rates for serious crime in Washington and Colorado changed following legalization compared with states which did not decriminalize or legalize marijuana. Specifically, we construct interrupted time-series models to examine trends of clearance for Part I violent and property crimes as well as disaggregated models that examine clearance rate trends for rape, robbery, aggravated assault, burglary, larceny, and motor vehicle theft. We use 2010 to 2015 UCR data on the offenses known to be cleared by arrest to create monthly totals of violent and property crime clearance rates as well as disaggregated counts by crime type. Although National Incident-Based Reporting System data would allow for the examination of clearance rates for a broader set of crimes, National Incident-Based Reporting System data are not broadly adopted by agencies and so the UCR data provide better coverage. We selected these years because 2010 allows us to track the trend prelegalization (which occurred in late 2012 in Colorado and WA) and because 2015 is currently the most recent year of data available.
While there are several different approaches to interrupted time-series modeling, we adopt the multiple group Interrupted Time-Series Analysis (ITSA) approach described by Linden and colleagues (Linden, 2015; A. Linden & Adams, 2011). Although autoregressive integrated moving average (ARIMA) models are often applied to interrupted time-series designs, Linden (2015) argues ARIMA models are highly sensitive to specification choices. In contrast, Linden’s regression approach is both robust and simple to implement. As described in detail by Linden (2015), the multiple group interrupted regression series model is defined as follows:
The first four terms of this model ( through ) are the standard regression-based single-group interrupted time-series model, where is the outcome variable measured at each time period, is the number of time units that have passed since the intial measurement, and is a dummy variable where prior to the intervention and at and after the point of intervention. represents the linear trend in the outcome prior to the intervention, while represents the immediate treatment effect of the intervention and the treatment effect overtime. The four following terms all include Z, which is a dummy variable indicating whether an observation is in the treatement or control groups. As such represent the interaction between being in the treatment group (in this case, a state that legalized recreational marijuana) and the previously defined regression terms. Roughly speaking, these estimates are the difference between the treatment group and the control group in their preintervention slopes, immediate treatment effects, and posttreatment slopes.
Linden (2015) notes, however, that the posttreatment difference in slope coefficients () should not be interpretted as the raw difference in slopes between the treatment and control groups but instead as the difference between the treatment and control groups relative to their pretreatment differences. This type of analysis of trend data allows for an examination of prelegalization trends, immediate legalization effects, postlegalization trends, and the differences in these trends and the comparison of effects on crime clearance for states which did and did not legalize marijuana.
The natural interruption point used for this study is the month of legalization—December 2012 for Washington and November 2012 for Colorado. We use the date of legalization as the intervention point because it marks the time point at which law enforcement investment in pursuing marijuana possession offenses would be largely fruitless. We estimated these models in Stata 14 using the ITSA package (Linden, 2015). Given that the interruption period was not the same for Colorado and Washington, it was necessary to estimate ITSA models for each state separately, using the rest of the country as the control group. For each of the Colorado models we omitted Washington, and for each of the Washington models we omitted Colorado, as the legalizaiton process occuring in these states made them inappropriate as part of the control average for the other. In addition, we omitted Alabama and Florida from the control average due to missing data concerns. As a robustness check, we reestimated the earlier models using only the neighboring states of Kansas and Idaho as the control groups for Colorado and Washington, respectively. These models are substantively similar in terms of both sign and significance, indicating that the patterns described in the results hold regardless of whether Colorado and Washington are compared with the whole country or with neighboring states which, during the research period, investigated did not enact either medical or recreational marijuana laws.
For evidence on trends in crime clearance rates, we present our results both visually and in table form. Figures 2 and 3 present the trends of aggregated clearance rates for violent and property crimes, respectively. As shown in Figures 2 and 3, UCR evidence suggests that violent and property crimes clearance rates shifted at the point of intervention. Interpretation of the Colorado and Washington results for the immediate and posttreatment coefficients must be done in reference to the rest of the country. For example, the immediate Washington effect of I-502 on motor vehicle theft is 2.997 + 2.029.
Prior to legalization, clearance rates for violent and property crimes were declining in both Colorado and Washington. However, immediately after legalization, the slope of the clearance rate trends shifted upward for violent crime in both of the treatment states. Conversely, while there was a jump in the trend line for average violent clearance rate at the point of intervention at the national level, postintervention clearance trends did not shift upward as occurred in the treatment states. This set of findings suggests that right around the time of legalization, clearance rates trends seemed to increase for violent crime in general for both Colorado and Washington, though no similar shifts are noted for the country as a whole. In terms of property crime clearance rates, there is a sharp increase for both Colorado and Washington after the point of intervention, though Colorado’s trend continues upward while Washington’s property crime clearance rates appear to regress back to prelegalization levels. The United States as a whole, however, remained essentially stable during this time period and followed a relatively predictable cyclical pattern of property crime clearance rates.
Figures 2 and 3 show a general change in clearance rates around the point of legalization. However, it is possible that any resultant resource allocation might affect specific types of crimes more directly than others. Although proponents suggested that legalization would allow police to spend more time on more serious crimes, it is likely that police already spent a significant amount of resources investigating more serious crimes, and, therefore, any changes might be most visible by conducting this analysis on disaggregated crime clearance trends. Figures 4 and 5 present overtime clearance rates for the two disaggregated offenses that show the most striking changes since the interruption point—burglary and motor vehicle theft. Visually, this analysis shows that while the percentage of burglary and motor vehicle theft offenses cleared by arrest per month was consistently declining in Colorado and Washington prior to legalization, the clearance rate for these two offenses increased dramatically postlegalization; in contrast, national trends remained essentially flat. This change is most notable in Colorado, where the clearance rate trend remained upward as of the end of 2015.
As it can be difficult to discern the changes presented as visual trends on graphs, we also present the interrupted time-series regression results for violent and property clearance rates as well as the clearance rate for crime-type disaggregated interrupted time models in Colorado and Washington in Tables 1 and 2. The multiple group ITSA regression approach produces coefficients for time trends prior to intervention, immediate treatment effects, and posttreatment effects overtime as well as providing coefficients describing the difference between the treatment (Colorado or Washington) and control (the rest of the country) trends.
Table 1. Interrupted Time-Series Analysis Results on Crime Clearance Rates per Month for CO.
Motor vehicle theft
U.S. trend before I-502
Pretreatment intercept difference between CO and United States
Pretreatment slope difference between CO and United States
Immediate average legalization effect
Posttreatment average slope
Immediate CO effect
Posttreatment CO effect
Note. CO = Colorado.
*p < .01. **p < .05. †p< .1.
Table 2. Interrupted Time-Series Analysis Results on Crime Clearance Rates per Month for WA.
Motor vehicle theft
U.S. trend before I-502
Pretreatment intercept difference between WA and United States
Pretreatment slope difference between WA and United States
Immediate average legalization effect
Posttreatment average slope
Immediate WA effect
Posttreatment WA effect
Note. WA = Washington.
*p < .01. **p < .05. †p < .1.
The row “Immediate Average Legalization Effect” in Tables 1 and 2 displays the immediate shift in clearance rates at the intervention point. The coefficients for this variable are statistically significant and positive for violent crime, property crime, robbery, aggravated assault, burglary, larceny, and motor vehicle theft models for both states. This implies that there was a significant increase in the clearance rates for each of these crimes in late 2012. This is an important finding, as it suggests that a simple jump in clearance rates for Colorado or Washington would not indicate a treatment effect, as there was an average increase in clearance rates for included states for these crimes at the point of intervention.
To determine if there were significant immediate shifts in Colorado or Washington, the “Immediate CO Effect” and “Immediate WA Effect” rows in Tables 1 and 2 must be referenced. These coefficients represent the difference in immediate treatment effects between CO or WA and the control states. For Colorado, there were no statistically significant treatment effects (see the null results for the “Difference in Legalization Effect between the Colorado and United States” rows in Table 1). Similarly, Washington did not differ from the control states in terms of immediate treatment effects in the vast majority of models. The only significant difference is for motor vehicle theft clearance rates, in which Washington’s clearance rate increased by nearly 3% (b = 2.997) more than the control states at the point of intervention. This is a noteworthy response, given that clearance rates for motor vehicle thefts increased by about 2% (b = 2.029) for states on average, thus suggesting that the rate in Washington increased by approximately 5% immediately following legalization, a jump which is clearly indicated on the right-hand side of Figure 5. These results suggest that legalization generally did not result in an immediate change in clearance rates for either Colorado or Washington, with the exception of a large shift in motor vehicle theft clearance rates in Washington.
These results are not surprising, as it is expected that resource shifts might take some time to produce measurable results. A key strength of the interrupted time-series approach is the ability to identify long-term treatment effects of interventions. To do this, trends following legalization for all states are estimated (this is the Post-Treatment Average Slope row in Tables 1 and 2) and then in trends between the treatment states and the control states are also estimated (these are the Post-Treatment CO/WA Effect rows). Statistically significant findings in the Post-Treatment Average Sloperows would indicate a significant shift in clearance rates for all states in the analysis, while a significant finding in the Post-Treatment CO/WA Effect rows would indicate that clearance rates were increasing or decreasing faster than average in the treatment states.
For Colorado, there were several significant differences in slope postintervention, suggesting that clearance rates for violent crime, property crime, rape, robbery, burglary, and larceny were all increasing faster in Colorado than in the rest of the country. For example, the Post-Treatment CO Effect result for larceny indicates that larceny clearance rates were increasing by .149 on average more per month than in the control group states. Therefore, over the course of a year, our model suggests that larceny clearance rates increased by 1.79% more than rest of the country (which remained essentially flat, as noted by the null results for the Post-Treatment Average Slope coefficient). These significant increases can be seen visually in Figures 2 to 5. These results are particularly noteworthy given that prelegalization the slope for violent crime, rape, robbery, and aggravated assault clearance rates in Colorado was lower than in the rest of the country.
For Washington, the postintervention trend results are similar, albeit somewhat less pronounced. As with Colorado, this pattern changed postintervention, with Washington slopes for violent crime (b = .183) and burglary (b = .064) growing at a significantly greater rate than the rest of the country following the point of intervention. Unlike Colorado, where there were no significant differences in the immediate legalization effects between Colorado and the rest of the nation, the immediate increase for motor vehicle theft clearance in Washington was much greater for the rest of the country.
The results noted here suggest that while there were both immediate and longer term differences between states which legalized and the rest of the country in terms of crime clearance rates, the long-term differences are much more pronounced, especially in Colorado. While there was an average immediate increase in most clearance rates in late 2012 for both the treatment and the control states, the trends into 2013 and onward suggest that crime clearance rates were increasing more rapidly in states that legalized (and especially in Colorado). These results reflect a large shift in clearance rates, as clearance rates had been declining in comparison to the rest of the country for a variety of crime types in both states.
Finally, a visual inspection of the clearance rates for various crimes suggests that there may be important monthly fluctuations that our standard ITSA model would not capture. To examine these possibilities, we estimated a set of generalized least squares models with autoregressive error terms and monthly control variables to account for seasonal variations (Maggin et al., 2011). Finally, we also estimated models in which Colorado and Washington were compared with a single control group (specifically, Kansas and Idaho, as these are neighboring states with no recreational or medical marijuana laws). Both sets of results, available upon request, were substantively similar in terms of sign and significance.
This study examined clearance rates among police agencies’ in Washington state and Colorado. Specifically, the study sought to determine what, if any, influence recreational marijuana legalization had on clearance rates. As advocates for marijuana legalization argued, legalization would allow police agencies to prioritize other activities, which in turn would increase clearance rates and reduce crime (Trilling, 2016).
While our research does not model changes on crime, our results suggest that, just as marijuana legalization proponents argued, the legalization of marijuana influenced police outcomes, which in the context of this article is modeled as improvements in clearance rates. Specifically, clearance rates grew more in Colorado than in the rest of the country for all crime types except aggravated assault and motor vehicle theft and similarly rose more in Washington than in the rest of the country for violent crimes and burglary. There were no crime types in either state for which legalization appeared to have a negative impact on clearance rates. In addition to these inferential results, the time-series plots are also remarkably dramatic, showing clear visual evidence of both an immediate jump in clearance rates and a later upward trend.
At the most basic level, it could be surmised that police agencies are allocating their resources to other crimes, and those crimes are being cleared at higher levels because they no longer dedicate time to minor marijuana offenses. From a theoretical perspective, legalization may have influenced case level and environmental factors by allowing agencies to reprioritize to index crimes with lower clearance rates. However, this explanation assumes police agencies are homogenous, when in reality police resources, expertise, expectations, and prioritizations are often extremely diverse.
Reflecting on these results, the most basic explanation may be associated with measures of individual performance within organizational performance models driven by arrest rates. These marked improvements in clearance rates could very well reflect the practical reality that as arrests remain a key indicator for officer performance, legalization, for pragmatic reasons, meant officers needed to reprioritize and what we are seeing here is a natural response to a change not merely to the law, but a response to a need to demonstrate continued laudable performance. As Brodkin (2011) would argue, these streel-level practitioners, or as Lipsky (1980) would offer street-level bureaucrats, are responding and adapting to this policy change within an environment where agencies have not readjusted their performance metrics. It is unlikely that agencies had redesigned their performance metrics following legalization, and as a consequence officers are adjusting to meet existing performance goals in an environment where they are no longer able to “produce numbers” using marijuana offenses.
While we are confident, our analysis can explain to some degree why clearance rates have improved after the reform of marijuana laws in Washington and Colorado, we cannot speak to the factors associated with how they have improved. Returning to the work of Brodkin (2011), how these rates occurred is tantamount to understanding the broader impact of this policy. Are police officers working more overtime hours? Are we seeing new strategies being implemented or used? Or perhaps, are we witnessing the return of older, more aggressive crime clearance strategies? We cannot answer these questions with our current data.
There are several noteworthy limitations associated with this study. As noted earlier, interrupted time-series models make use of trends pre- and postintervention in treatment and control observations. The interruption point acts as a natural pre- and postintervention comparison point, and the estimation of trends before and after I-502 and A-64 allows us to control for other changes in Washington and Colorado before and after the legalization of marijuana. But it is possible that some other shift happened in or around November 2012 and December 2012 that affects our results. Although we are not aware of any specific policy changes in Washington or Colorado relevant to crime clearance rates, aside from the legalization amendments that might explain shifts in clearance rates, this possibility must still be considered.
Furthermore, it is possible that different external factors occurred in or around November 2012 and December 2012 causing shifts in clearance rates. For example, it could be that several law enforcement agencies implemented automated license plate detection systems around this time, and this might also explain the immediate and large treatment effects for motor vehicle theft clearance rates. We have made direct inquiries concerning such public policy-related changes, and no such dramatic changes are to be noted in either Washington or Colorado. Nonetheless, some other combinations of factors might explain changes in other crime clearance rates.
In addition, it is also important to note that there are potential seasonality effects in these models. Some cyclical seasonality can be seen in the time-series plots presented in Figures 2 to 5. As a robustness check, we examined seasonal single-group interrupted ARIMA models for Colorado and Washington and found quite similar results, though to our knowledge a multigroup seasonal ARIMA is not possible to construct. Moreover, there is likewise a potential seasonality effect which is confounded with legalization itself. It is possible that agencies push officers to clear more crimes in December of each year. If so, this would confound our results, given that our treatment point occurs in December in the Washington models. We explored this possibility in supplementary analyses (available upon request) by conducting bivariate regression models on clearance rates disaggregated by crime types where we used a dummy indicator for December (with the rest of the year as a reference category) to see if clearance rates tended to increase significantly in December. Results of this supplementary analysis were decidedly mixed, with some crimes (e.g., aggravated assaults) showing increased clearance rates in December and others showing a significant decline in December (burglary), while there was no “December effect” for others (motor vehicle thefts).
Given these mixed results in our supplemental analyses, we do not believe that the December 2012 intervention effect is simply a standard December end of the year effect. There is a clear December effect in evidence for some offenses, and this is an important qualification to our findings and suggests that our broader results could be potentially biased by maturation effects. Although we acknowledge this possibility, the fact that trends did not significantly decline following December 2012 lends considerable credence to the argument that there has been a genuine and persistent upward shift in clearance rates.
Finally, it is worth noting that the way clearance rate data are collected may result in some measurement error. Specifically, the UCR presents information on the number of crimes that occur each month and the numbers of crimes which are cleared during that period. The crimes cleared need not be the same crimes which occurred, as one can easily envision scenarios in which a crime took longer than a month to occur or in which a crime occurred late in the month and was cleared early the following month. Although the focus on months is necessary, as there are not enough yearly time periods to examine trends pre- and postlegalization, we investigated the potential effects of this measurement error by examining time to clearance using 2013 National Incident-Based Reporting System data. Supplementary analyses (available upon request) suggest that 74.8% of clearances happen the same day as the crime is reported, and over 90% of the clearances occur within 30 days of reporting, resulting in about 88% of crimes being cleared in the same month that they occur. Thus, while there is some underlying error in monthly clearance rates, we argue that this error is likely small in size and could not plausibly explain the substantial shift in trends documented earlier. Although these models offer strong comparative logic, they do not have the strength of random assignment and therefore cannot definitively account for external influences.
There is an unfortunate lack of systematic research in the clearance rate literature concerning resource allocation. While a few studies have explored the influence that marijuana deprioritization mandates had on clearance rates (Ross & Walker, 2016), the only analogous research on a large-scale policy change would be the termination of the New York Police Department’s stop-and-frisk policy. However, research on what effect termination of stop-and-frisk had in New York City has been restricted to crime trends (Cullen, 2016; see also Ferrandino, 2013). The only loosely relevant empirical work is that on the relationship between broken windows and clearance rates, which shows an inconsistent impact of broken windows enforcement on clearance rates for different types of crime (Jang et al., 2008). Therefore, we believe it is proper to explore this relationship with our present analytical strategy.
Unfortunately, we lack a longer period of analysis for the interrupted time-series results reported here. It is possible that these results would lessen in significance with added data points past 2015. Moreover, with a longer time period, it would be possible to examine other factors related to legalization. For example, the retail sale of marijuana might have created new pressures and challenges for law enforcement officers, such as diversions to minors and transport out of state, new pressures which might subsequently induce a decrease in clearance rates. Alternatively, it is possible that as marijuana revenues increase, added funding is funneled to criminal justice agencies resulting in greater latitude for resource allocations and increases in clearance rates as a consequence. Based on the findings and limitations reported here, we recommend that future research replicates our design with added years of data and includes the states which legalized recreational marijuana in 2016—namely, California, Massachusetts, Nevada, and Maine.
While our results cannot specifically explain why police clearance rates have increased in Colorado and Washington, we think the argument that legalization did in fact produce a measurable impact on clearance rates is plausible. This reallocation is striking even though some realigning of resources by police departments away from enforcement of marijuana offenses likely took place well before legalization (i.e., when medical marijuana laws were passed). For example, in 2003, the largest city in Washington, Seattle, implemented a citizens’ municipal ordinance initiative that directed the police to regard marijuana offenses as a law enforcement priority. In 2009, the third largest city in the state, Tacoma, passed a similar municipal ordinance by local initiative relating to marijuana possession and police priorities. Moreover, in 2009, Washington enacted a major further liberalization of its medical marijuana law and allowed a wide variety of persons to qualify as “medical providers,” a change which meant there was much more “legal” marijuana available among the citizens throughout the state. In Colorado, we observed nearly identical initiatives involving deprioritization mandates and medical marijuana. For example, Denver voters approved a deprioritization mandate in 2007.
Our models show no negative effects of legalization and, instead, indicate that crime clearance rates for at least some types of crime are increasing faster in states that legalized than in those that did not. This result is strong, as the multiple group ITSA approach controls for both preintervention clearance rates in the treatment states and compares trends to a control average made up of states which did not legalize. That we found similar positive results for Colorado and Washington is particularly noteworthy and supportive of a potential resource allocation explanation. These trends are particularly strong in Colorado, which might be reflective of a more generous allocation of state marijuana-derived revenues to state and local law enforcement than is the case in Washington (Caulkins et al., 2015).
While we limit our analysis to clearance rates, it would seem vital to figure out what effect, if any, considerable improvements in clearance rates have on overall crime trends within a city, or in our unit of analysis, the state. As we document here, prior to legalization, several crimes clearance rates were either flat or decreasing. However, in the postlegalization period, we see considerable improvement. We cannot offer with absolute certainty that these changes are entirely the result of marijuana legalization, though we are quite certain that legalization has not unduly hampered police performance, at least as measured by clearance rates. Moreover, in the absence of other compelling explanations, the current evidence suggests that legalization produced some demonstrable and persistent benefit in clearance rates, benefits we believe are associated with the marijuana legalization proponents’ prediction that legalization would positively influence police performance.
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This project was supported by Award No. NIJ-2016-9090, awarded by the National Institute of Justice, Office of Justice Programs, U.S. Department of Justice. The opinions, findings, and conclusions or recommendations expressed in this publication/program/exhibition are those of the author(s) and do not necessarily reflect those of the Department of Justice.
Bayley D. H. (1994). Policing the future. New York, NY: Oxford University Press.
Caulkins J.P., Kilmer B., Kleiman M., MacCoun R., Oglesby P., Pacula R., Reuter P. (2015). Considering marijuana legalization: Insights for Vermont and other jurisdictions. Santa Monica, CA: RAND Corporation.
Federal Bureau of Investigation. (2009a). Uniform Crime Reporting Program Data: Arrests by age, sex, and race, summarized yearly, 2005. Ann Arbor, MI: Inter-university Consortium for Political and Social Research [distributor].
Federal Bureau of Investigation. (2009b). Uniform Crime Reporting Program Data: Arrests by age, sex, and race, summarized yearly, 2006. Ann Arbor, MI: Inter-university Consortium for Political and Social Research [distributor].
Federal Bureau of Investigation. (2009c). Uniform Crime Reporting Program Data: Arrests by age, sex, and race, summarized yearly, 2007. Ann Arbor, MI: Inter-university Consortium for Political and Social Research [distributor].
Federal Bureau of Investigation. (2010a). Uniform Crime Reporting Program Data: Arrests by age, sex, and race, summarized yearly, 2003. Ann Arbor, MI: Inter-university Consortium for Political and Social Research [distributor].
Federal Bureau of Investigation. (2010b). Uniform Crime Reporting Program Data: Arrests by age, sex, and race, summarized yearly, 2008. Ann Arbor, MI: Inter-university Consortium for Political and Social Research [distributor].
Federal Bureau of Investigation. (2011). Uniform Crime Reporting Program Data: Arrests by age, sex, and race, summarized yearly, 2009. Ann Arbor, MI: Inter-university Consortium for Political and Social Research [distributor].
Federal Bureau of Investigation. (2012). Uniform Crime Reporting Program Data: Arrests by age, sex, and race, summarized yearly, 2010. Ann Arbor, MI: Inter-university Consortium for Political and Social Research [distributor].
Federal Bureau of Investigation. (2013). Uniform Crime Reporting Program Data: Arrests by age, sex, and race, summarized yearly, 2011. Ann Arbor, MI: Inter-university Consortium for Political and Social Research [distributor].
Federal Bureau of Investigation. (2014). Uniform Crime Reporting Program Data: Arrests by age, sex, and race, summarized yearly, 2012. Ann Arbor, MI: Inter-university Consortium for Political and Social Research [distributor].
Federal Bureau of Investigation. (2015). Uniform Crime Reporting Program Data: Arrests by age, sex, and race, summarized yearly, 2013. Ann Arbor, MI: Inter-university Consortium for Political and Social Research [distributor].
Federal Bureau of Investigation. (2016). Uniform Crime Reporting Program Data: Arrests by age, sex, and race, summarized yearly, 2014. Ann Arbor, MI: Inter-university Consortium for Political and Social Research [distributor].
Federal Bureau of Investigation. (2017). Uniform Crime Reporting Program Data: Arrests by age, sex, and race, 2015. Ann Arbor, MI: Inter-University Consortium for Political and Social Research [distributor].
Linden A., Adams J. L. (2011). Applying a propensity score‐based weighting model to interrupted time series data: Improving causal inference in programme evaluation. Journal of Evaluation in Clinical Practice, 17, 1231–1238.
Maggin D. M., Swaminathan H., Rogers H. J., O'keeffe B. V., Sugai G., Horner R. H. (2011). A generalized least squares regression approach for computing effect sizes in single-case research: Application examples. Journal of School Psychology, 49, 301–321.
Reiner R. (1998). Process or product? Problem of assessing individual police performance. In Brodeur J. P. (Ed.), How to recognize good policing: Problems and issues (pp. 55–72). Thousand Oaks, CA: Sage.
Warburton H., May T., Hough M. (2004). Looking the other way: The impact of reclassifying cannabis on police warnings, arrests and informal action in England and Wales. British Journal of Criminology, 45, 113–128.
David A. Makin is the lab director of the Complex Social Interactions Lab and an assistant professor in the Department of Criminal Justice and Criminology at Washington State University. His research explores the impact of technology and policy on public safety; sustainable technology integration; systemic social observation and event modeling; and refinement of, and evaluation of, public safety education and training.
Dale W. Willits is an assistant professor in the Department Criminal Justice and Criminology at Washington State University. His research interests include policing, violence, and public policy.
Guangzhen Wu is an assistant professor of criminology and criminal justice in the Department of Sociology at the University of Utah. He received his PhD in criminal justice and criminology from Washington State University. His research interests include comparative criminology/policing, police integrity, police stress and job satisfaction, and police–public relations.
Kathryn O. DuBois is an associate professor in the Department of Criminal Justice and Criminology at Washington State University Vancouver. Her research deals with the causes of and responses to violence against women in rural, remote, and indigenous communities.
Ruibin Lu is an assistant professor of criminal justice at Stockton University. Her research work focuses on law and court, specialty courts, drug and alcohol, and policy evaluation.
Mary K. Stohr, PhD, is a professor in the Department of Criminal Justice and Criminology at Washington State University. Her research of late is focused on drug policy implementation.
Wendy Koslicki, PhD, is an assistant professor in the Department of Criminal Justice and Criminology at Ball State University. Her research interests include U.S. police culture, police equipment and technology, and mixed methodology research designs.
Duane Stanton is a PhD candidate with Washington State University’s Department of Criminal Justice and Criminology. He serves as a research associate and agency outreach coordinator for the department.
Craig Hemmens, JD, PhD, is a professor in the Department of Criminal Justice and Criminology at Washington State University. He has published more than 20 books and 200 articles and other writings regarding a variety of issues in criminal justice.
John Snyder, JD, is a clinical professor in the Department of Criminal Justice and Criminology and works for the Office of Civil Rights, both at Washington State University. For many years, he was a practicing criminal defense attorney and his research interests are in the courts and law areas.
Nicholas P. Lovrich is Regents Professor Emeritus in the school of politics, philosophy and public affairs and research affiliate in the department of criminal justice and criminology at Washington State University.
Download to reference manager
If you have citation software installed, you can download article citation data to the citation manager of your choice