Context: The opioid epidemic is a major US public health crisis. Its scope prompted significant public outreach, but this response triggered a series of journalistic articles comparing the opioid epidemic to the crack cocaine epidemic. Some authors claimed that the political response to the crack cocaine epidemic was criminal justice rather than medical in nature, motivated by divergent racial demographics.
Methods: We examine these assertions by analyzing the language used in relevant newspaper articles. Using a national sample, we compare word frequencies from articles about crack cocaine in 1988–89 and opioids in 2016–17 to evaluate media framings. We also examine articles about methamphetamines in 1992–93 and heroin throughout the three eras to distinguish between narratives used to describe the crack cocaine and opioid epidemics.
Findings: We find support for critics' hypotheses about the differential framing of the two epidemics: articles on the opioid epidemic are likelier to use medical terminology than criminal justice terminology while the reverse is true for crack cocaine articles.
Conclusions: Our analysis suggests that race and legality may influence policy responses to substance-use epidemics. Comparisons also suggest that the evolution of the media narrative on substance use cannot alone account for the divergence in framing between the two epidemics.
The opioid epidemic is one of the most significant public health crises in the United States today. Of the more than 70,200 drug overdose deaths in 2017, around 68% were attributed to opioid abuse (CDC 2018). The statistical trends regarding the epidemic are equally alarming; in 2017, the number of overdose deaths involving opioids was six times higher than in 1999 (CDC 2018). The worrying scope of the epidemic has prompted significant public outreach, including the White House declaring the epidemic a public health emergency while calling for better controls of prescription medications and access to substance-use treatment facilities (Davis 2017). Congress likewise allocated funding to treat the opioid epidemic as a public health concern and not a criminal justice issue in the 21st Century Cures Act, enacted in December 2016. State governments have also responded to the opioid epidemic through public health initiatives. For example, in 2017 Illinois released an Opioid Action Plan not through its law enforcement agencies but through its Department of Public Health, noting “The focus of our efforts is to save lives” (Rauner 2017). The messaging is clear: substance use is a medical, not a law enforcement, issue.
The rise in public health–focused responses to the opioid epidemic triggered a series of journalistic articles and opinion pieces comparing the epidemic to the earlier crack cocaine epidemic. Titles of these pieces included “America's Racist Response to the Crack Epidemic Must Inform the Way We Tackle Opioids” (Sharpton 2018) and “Why Didn't My Drug-Affected Family Get Any Sympathy?” (Bailey 2018). The authors of these pieces compared the criminal justice response to the crack cocaine epidemic of the 1980s with the public health approach to the current opioid epidemic and concluded that the differences in framing and public policy solutions were motivated by racism. Specifically, they were concerned that the differential responses were due to an association between crack cocaine and African Americans while the opioid epidemic is now seen as a problem afflicting white communities.
This article examines these assertions by analyzing the language used in newspaper articles from each period. Drawing on newspapers across the country, we compare word frequencies from articles about crack cocaine in 1988–89 and opioids in 2016–17 to evaluate how each epidemic was framed in public discourse. We particularly examine language indicating a medical model of addiction, such as health and treatment, as compared to terms of social control, such as police, enforcement, and arrest. We find support for the critics' hypothesis about the differential framing of the two substance-use epidemics: articles on the opioid epidemic are more likely to use medical or public health terminology—and evince concern about people. By contrast, articles on the crack cocaine epidemic utilize criminal justice terms and evoke concerns about drugs more frequently. To better understand the role that type of substance and race of users associated with the epidemic plays, we also examined articles about the methamphetamine epidemic from 1992–93, as this is an epidemic that has been associated with white users (as with opioids) but focuses on a stimulant (similar to crack cocaine). To better trace shifts in narratives, we pulled articles focusing specifically on heroin use from 1988–89, 1992–93, and 2016–17. The framing of the two epidemics, especially in contrast to articles written during these two periods about heroin, suggests that the evolution of understanding substance use disorders is not the sole explanation for the difference in media narratives. Instead, another factor, such as street drug status or race, may play a role in the public policy response and outreach to substance-use epidemics.
Public Policy, Substance Use, and Media Narratives
The media influences public policy agendas by prioritizing some news topics over others and therefore influencing which issues the public consider important (McCombs 1993). The media also has the ability to frame issues by influencing the perceptions the public and key policy makers have on topics as well as suggesting appropriate responses and solutions (Bennett 1990, 2016; Busby, Flynn, and Druckman 2018; Scheufele and Tewksbury 2007). This influence may be particularly strong on complicated public policy issues, such as poverty and welfare programs. For example, there is evidence that media framing connecting social welfare programs to race can shift public opinion and, in turn, public policies on the best way to address poverty alleviation (Winter 2006).
The media's effectiveness in setting agendas and framing issues is well documented in public responses to health issues (Brodie 2003). The ability of mass media to influence the perceptions regarding substance users and drive policy is no exception. As early as 1963, Howard Becker illustrated the role that media played in shaping the perception of drug users—in his case marijuana users—as deviant (Becker 1963). Criminal justice historians have documented that politicians and media often construct a narrative of drug use that presents substance use as a criminal act rather than one that arises from a medical disorder. Although the “just say no” era of the 1980s and 1990s is the most popular example of the criminalization model of substance use, it is prevalent throughout US political history. For example, the enactment of harsh mandatory minimum sentencing laws for marijuana and heroin occurred in the 1950s, when the dominant narrative around substance use was that foreign pushers were inflicting substances on defenseless Americans (Lassiter 2015).
The model used to frame a substance use epidemic is crucial because it not only shapes public perception of the epidemic but also public policy responses. By linking substance use and crime, the criminalization model drives draconian and drastic solutions to the use of illicit substances. Examples of these public policy interventions include mandatory sentencing, “three strikes and you're out” policies, and even the use of the death penalty in some substance use cases (Krisberg 2015). These policies are often promoted under a “get tough on crime” campaign that can also drive the rise of a police state, with more funding to law enforcement, more criminal prosecutions, and higher rates of incarceration (Messner and Rosenfeld 2007). Unfortunately, the public policy interventions often justified by the criminalization model have been shown to have little impact on rates of serious crime (Messner and Rosenfeld 2007). They do come with important concerns regarding civil rights and liberties, however. For example, likely due to rhetoric based in the criminalization model of substance use in the 1980s, the percentage of Americans who felt that testing workers for substance use in general would be an unfair invasion of privacy declined from 44 percent in 1986 to 24 percent in 1989 (Beckett 1994).
The medical model of substance use disorder calls for very different interventions. Instead of adopting a “tough on crime” stance, policy makers influenced by the medicalization model respond with offers of help, including resources for treatment and prevention (Netherland and Hansen 2017). Treatment interventions include diverting substance users from prisons to rehabilitation facilities, sometimes through drug courts. Prevention policies include prescription drug monitoring programs to limit the number of opioid prescriptions in a community as well as educational efforts around substance use. There is an understanding under this model that substance users are victims of their own biology. Blame may be shifted from the user to the supplier—either the pharmaceutical industry, physicians who are careless with their prescriptions, or illicit drug sellers (Macy 2018). A telling outgrowth of the focus on the supply side of substance use is the rise of prescription drug monitoring programs, funded by the 2005 National All Schedules Prescription Reporting Act and present in 49 states (Oliva 2018). These programs are intended to reduce the availability of opioids by requiring prescribers to report the number of opioid prescriptions they write for patients, suggesting a perception that overeager providers—and not users seeking out opioids—are the driver of opioid use. Overall, the medicalization model is kinder to substance users in that it encourages policy makers to provide health care resources to support recovery rather than harsher criminal justice interventions.
Race, Substance Use, and Media Narratives
Layered on top of media narratives and frames around substance use in the United States is the role of race. Evidence suggests that the nature of racial content influences which media framing resonates most strongly with the public and becomes the dominant narrative. While some of this response may be motivated by explicitly racist biases, most is likely a result of implicit (Beckett, Nyrop, and Pfingst 2006: 106) and structural biases (Bobo and Thompson 2006). Racial cues can feed into the perception of the severity of a crime or social problem (Beckett, Nyrop, and Pfingst 2006: 107). In the case of substance use, associating addiction with communities of color can persuade the public and policy makers that substance use causes violence, crime, and other problematic behaviors. An association between race and addiction can give media narratives that emphasize the criminalization model of substance use greater resonance among policy makers and the public alike.
Indeed, Michelle Alexander (2012) argues that today's mass incarceration of African Americans originates in a deliberate strategy of the Reagan administration to link substance use, criminality, and race in media narratives in order to drum up public and congressional support for the War on Drugs, announced in 1982. Capitalizing on backlash against the civil rights movement of the 1960s, conservative politicians beginning with Richard Nixon and his Southern Strategy realized the political utility of using racial appeals to attract white voters (see also Mendelberg 2001). When it came into power, the Reagan administration deliberately sought to shape perceptions of drug users and views on drug policy, announcing the War on Drugs at a time when illegal drug use was actually declining (Alexander 2012: 6). The announcement predated the crack epidemic, but provided the pretext for mass imprisonment once crack began to spread in poor black neighborhoods.
The impact of the criminalization model of substance use on the African American community is significant. African Americans are incarcerated at more than five times the rate of whites (NAACP n.d.). While African Americans and whites use illicit drugs at similar rates, imprisonment rates for drug-related offenses are almost six times higher for African Americans than for whites, and while African Americans represent only 12.4% of illicit substance users, they constitute 33% of those incarcerated in state facilities for drug-related offenses (NAACP n.d.). The mass incarceration of African Americans for drug-related offenses also has serious collateral consequences, including exclusion from social supports such as financial aid and housing benefits, and challenges in achieving employment and financial security (National Research Council 2014), what Alexander terms “a new racial caste system” (2012: 3).
Even when white users fall under the criminalization model, they are often protected by their racial privilege. For example, when there was an outbreak of heroin use among white high school and college students in the Dallas area in the 1990s, the law enforcement response was to prosecute the Mexican cartels “preying on this community” (Lassiter 2015: 126). As a result, although whites represent a sizable majority of substance users and drug dealers in the United States, they compose only one-quarter of drug offenders in state prisons (127). Similarly, methamphetamine has been linked to white users, and therefore its users are often portrayed as deserving more sympathy, and less linked to violence than crack cocaine users are (Murakawa 2011). An analysis of media depictions of drug use during the first decade of the 2000s found that prescription opioid users were typically portrayed as white, suburban or rural, and deserving sympathy, while heroin users were usually described as black or Latino, urban, and criminal (Netherland and Hansen 2016).
To assess the extant hypothesis that the opioid and crack cocaine epidemics were differentially framed in the media, we drew articles from major US newspapers, including the Chicago Tribune, Los Angeles Times, Newsday, St. Louis Post-Dispatch, Boston Globe, New York Times, Wall Street Journal, and Washington Post. We had originally intended to use several regional newspapers, including from areas with high crack cocaine or opioid use, but were limited to the set of papers whose digital archives extended back sufficiently.
We narrowed our searches to two time periods, 1988–89 and 2016–17, to maximize similarities between the two epidemics. In 1988, during George H. W. Bush's election year, there was a surge in reporting on crack cocaine that continued into the first year of his administration (Reinarman and Levine 1997: 21). Because the 2016 election of another Republican candidate, Donald Trump, coincided with media attention to the opioid epidemic, we chose the years 2016–17 for comparison. To search for articles we used the search engine Lexis Advance. One search was for articles that included crack cocaine and that were published between January 1, 1988, and December 31, 1989. This search resulted in 2,055 articles from the selected newspapers. The other search was for articles that included opioid and that were published between January 1, 2016, and December 31, 2017. This search resulted in 4,026 articles.
We also pulled articles focusing on methamphetamines and heroin, to better capture a variety of factors and to serve as comparisons to the crack cocaine and opioid epidemics. Methamphetamine was selected as a drug epidemic that focused on a stimulant—similar in that respect to crack cocaine—but also as an epidemic that has often been associated primarily with white users (Murakawa 2011)—akin to the opioid epidemic. To search for articles on methamphetamine, we focused on 1992–93, because those years encompassed a presidential election and the start of a new administration under President William J. Clinton. We separated heroin from our opioid search, even though heroin is an opioid, in order to track the trajectory of a longer running epidemic as well as to compare the narratives around a street drug to a broader category of substances that include prescription medications.1 Our searches for articles that focused on heroin use encompassed 1988–89, in order to serve as a temporal comparison to our crack cocaine search, 1992–93, to serve as a comparison to our methamphetamines search, and 2016–17, to serve as a comparison to our broader opioid search. There were 391 articles for the methamphetamine sample, 3,956 articles for heroin 1988–89, 2,865 for heroin 1992–93, and 4,112 articles for heroin 2016–17.
We compared the word frequencies in our samples using a variety of methods, including looking at relative word frequency and creating a topic model using Latent Dirichlet Allocation (LDA). LDA (Blei, Ng, and Jordan 2003; Pritchard et al. 2000) is a generative statistical model that posits that each document (newspaper article in our case) is a mixture of a number of topics and that each word in the document can be attributed to a particular topic. When using LDA, the distribution of topics is generally assumed to have a sparse Dirichlet prior, reflecting an assumption that documents cover only a small set of topics and that topics use only a small set of words frequently. In our case, such an assumption is reasonable because newspaper articles are generally narrowly focused and use common language. In LDA, topics are not semantically strongly defined and words may co-occur across different topics. A topic model allows us to see how topics vary across the 1988–89, 1992–92, and 2016–17 samples. Before running the model, the samples were cleaned as follows: stopwords (and, the, it, etc.),2 punctuation, and numbers were removed, and all words were converted to lowercase.3
The most frequently occurring words in the 1988–89 crack cocaine sample (see table 1) included language relating to criminal justice such as police, law, enforcement, and crime. The emphasis on criminal justice continued through our bigram analysis of the 1988–89 crack cocaine sample, with law + enforcement, substances + crime, drug + trafficking, drug + dealers, and illegal + drugs constituting half of the top ten. The trend continued in the trigram analysis of this sample, with controlled + substances + crime, special + investigative + forces, law + enforcement + officials, and law + court + tribunals.
By contrast, the most frequently occurring word in the 2016–17 opioids sample (see table 2) was health. Our bigram analysis of this sample also emphasized the rise of medical terminology, with health + care, public + health, and health + departments included in the top ten. The trigram analysis was the most heavily weighted toward medical language, with affordable + care + act, health + care + reform, public + health + administration, health + care + policy, substance + abuse + treatment, and health + care + law comprising the majority of the top ten. Criminal justice terms did linger somewhat, including abuse, law + enforcement, and controlled + substances + crime, but were much less prevalent in the 2016–17 opioid sample than in the earlier crack cocaine sample.
Although the unigram analysis of the 1992–93 methamphetamine sample (table 3) only includes two criminal justice words (police and court), the bigram analysis is largely dominated by criminal justice phrases, including law + enforcement, and pleaded + guilty. The trigram analysis consists entirely of criminal justice phrases, both ones that appear in the 1988–89 crack cocaine sample (law + enforcement + officials) and ones that are unique to this sample (fourth + reich + skinheads). Nevertheless, some key criminal justice words and phrases that dominate the crack cocaine word frequency analysis, such as crime, substance crime, drug trafficking, drug dealers, and illegal drugs, are missing from the methamphetamine word frequency analysis. There was an absence of public health or medical terms in all three analyses used for the methamphetamine sample.
The evolution of language in the samples of heroin articles from 1988–89, 1992–93, and 2016–17 shows elements observed in both the opioid sample and the crack cocaine sample. The unigram analyses of these three time periods (see table 4) include two criminal justice terms in 1988–89, only one criminal justice term in the next two periods, and no public health or medical terms. This stands in contrast to the 2016–17 opioid sample, which included four public health terms and no criminal justice terms (see table 2). The bigram analysis of the 1988–89 heroin sample (see table 5) does include one public health term (drug + treatment) but five criminal justice terms, similar to the 1988–89 crack cocaine bigram analysis (see table 1), which includes five criminal justice terms although no public health language. The 1992–93 and 2016–17 bigram analyses (see table 5) are closer to balanced, with three criminal justice terms and two public health terms. Interestingly, the 1988–89 heroin trigram analysis (see table 6) includes two public health terms, although the majority of the other top ten terms are criminal justice terms, whereas the 1992–93 trigram analysis includes no public health terms and is weighed heavily to criminal justice terms. The 2016–17 trigram analysis breaks down along similar lines to the 1988–89 trigram analysis for heroin. This is in sharp contrast to the trigram analysis of the 2016–17 opioid sample (see table 2), which is majority public health terms.
By comparing the relative frequencies of words in these samples, we can more clearly observe similarities and differences in the media narratives around these substance use epidemics. In the following figures, words that occur with similar frequency in both samples appear closer to the 45-degree line. Words that are more prominent in the first named sample occur above the line, and words that are more prominent in the second named sample occur below the line. High-frequency words in figures 1–3 almost uniformly suggest a criminalization model while high-frequency words more prominent in the opioids sample (below the 45-degree line) suggest a medicalization model.
In the unigram analysis (fig. 1), high frequency words drawn from articles written about the opioid epidemic include arrested and police, while words drawn from articles about crack cocaine include health, care, addiction, medicaid, and fentanyl. The criminalization model is further seen in the bigram analysis (fig. 2), with words such as law enforcement and drug trafficking from the crack cocaine articles, in comparison with the medicalized language of doctors and pain for the opioid sample. In the trigram analysis (fig. 3), expressions such as controlled substances crime stand out as associated with the crack cocaine epidemic as compared to health care professionals, which occurs with higher relative frequency in the opioids sample. Assertions that public discourse on the crack cocaine and opioid epidemics was different are upheld in this analysis of media coverage from each period.
The trend away from criminal justice to public health and medical terms found in the relative frequencies comparisons between the crack cocaine and opioid samples was also found in the comparison of the relative frequencies of words in the 1988–89 and 2016–17 heroin samples (see fig. 4). Although these samples cover the same substance, the terminology used in 2016–17 was much more medical, such as addiction treatment, than the 1988–89 sample, which includes terms such as crime family and court judge. This suggests that the narrative around substance use overall was shifting during this time period. This shift is also suggested by figure 5, which is the relative frequencies of words in the 1988–89 crack cocaine and 1988–89 heroin samples. In this figure, the words are clustered tightly around the dividing line, suggesting relatively little difference in the terminology used. The comparison between the 2016–17 heroin and opioid samples in figure 6 shows a little more divergence, on the other hand, with heroin being associated slightly more with criminal justice terminology and opioids being associated with health care terms.
Another way to see the divergence in rhetoric between the 1988–89 and 2016–17 samples is to use a topic model. As stated above, a topic model posits that each article is a mixture of a number of topics and that each word in the document can be attributed to one of the article's topics. Figures 7–10 show the results of topic models generated using LDA on the two samples. After repeated tests we chose to set the model to find three topics. This choice is justified by analyzing the Bayesian Information Criterion (BIC) associated with each model, which is an indicator of the trade-off between complexity and parsimony (Soleimani and Miller (2014). The BIC was minimized by either a two-topic model (for the crack cocaine 1988–89, opioids 2016–17, and methamphetamine 1992–93) or a three-topic model (for heroin 1988–89 and heroin 2019–17). To facilitate comparability, three-topic models were used for all, with the understanding that the third topic will vary in its coherence, often resembling a “remainder” category. The values in figures 7–10 are betas, which reflect the concentration of words in the topic.
In figure 7 (1988–89 crack cocaine sample) the three topics include two that are relatively coherent—topic 1 is a clear law-and-order topic with words such as law, enforcement, crime, and arrested. Topic 2 is centered around communities and families but includes the interesting additions of black and white, which are likely to touch on race. Topic 3 is less coherent, but seems to touch on government policy.
Figure 8, which shows the same analysis for the opioid sample (2016–17) has some overlap with the 1988–89 crack cocaine analysis in its topics—topic 1 is a community and home category (family, home, children) and topic 2 is still a political (trump, republican, government, policy) but topic 3 is clearly more medical than law enforcement (health, pain, addiction, patients, epidemic).4 This analysis shows that when a three-topic model is considered the two samples differ strongly on topic 3 with the 1988–89 crack cocaine sample including a law enforcement topic and the 2016–17 opioid sample including a medical topic.
Performing the LDA topic model analysis for the 1992–93 methamphetamine sample (see figure 9) resulted in a very clear criminal justice topic (topic 3), a family and community topic that leaned in an interestingly negative direction, including words such as jail (topic 2), and a remainder topic that was relatively incoherent, as can be seen from the low values of beta (topic 1). Therefore, the topic model analysis for the methamphetamine sample was closer to the crack cocaine sample than to the opioid sample, despite the perception of both opioids and methamphetamine as white drugs. A key difference, however, between the topic model analyses for crack cocaine and for methamphetamine was that the criminal justice topic was topic 1 in the crack cocaine model but only topic 3 in the methamphetamine model, indicating that this topic was more prevalent in the framing of crack cocaine coverage.
We performed the same analysis for the 1988–89 heroin sample and the 2016–17 heroin sample to compare them to the crack cocaine and opioid samples and also to compare them to each other to get a sense of any evolution in the media narrative between these two time periods. Figure 10 shows the analysis for the 1988–89 heroin sample. Topic 1 is very clearly a criminal justice topic with terms such as prison, crime, and criminal. Topic 2 again appears to be a community-focused topic, although, similar to the crack cocaine sample and not the opioid sample, it includes the term black, suggesting race. Topic 3 is focused around public health (with both public and health making an appearance) and treatment, which more closely aligns with the 2016–17 opioid analysis than the 1988–89 crack cocaine analysis. Figure 11 performs the sample analysis for the 2016–17 heroin sample. Topic 1 is a public health topic, while topic 3 appears to be a criminal justice topic. Topic 2 suggests a community and family focus with words such as school and mother and children.
The presence of both criminal justice and public health topics for both the 1988–89 and 2016–17 heroin samples is an interesting contrast to the topics for the crack cocaine and opioid samples.
With the heroin and crack cocaine samples, which both deal with illegal drugs purchased outside the medical system, we find criminal justice topics well represented. On the other hand, the opioid topic analysis does not yield a criminal justice topic. Using heroin in the two time periods as a control, we can see that the change from crack cocaine (with a criminal justice but no public health topic) to opioids (with no criminal justice but a public health topic) is not merely due to some secular change in increasing public health framings of drug epidemics. Heroin was around in both eras, and had a public health framing in both eras. But opioids appear to be framed differently than crack, because crack had no public health framing back in 1988–89.
The word frequency patterns in the samples analyzed suggest a shift in the media narrative between the crack epidemic of the late 1980s to the more recent opioid epidemic. The 1988–89 crack cocaine sample demonstrates a strong criminalization focus, while the 2016–17 opioid sample illustrates the rise of the medicalization model. The analysis of the 1992–93 methamphetamine sample suggests that the criminalization narrative around substance use was still strong then. Tracing the narrative around heroin from 1988–89 through 1992–93 to 2016–17 provides an interesting contrast to the narratives utilized in the crack cocaine and opioid epidemics. The analysis of the 1988–89 heroin sample demonstrates that a public health framing was available during that time period and indeed appears in the topic models for 1988–89 heroin, but was not similarly used in the crack cocaine sample. While we observed both criminal justice and public health topics for heroin in both the earliest and latest heroin samples, we did not observe criminal justice or law enforcement topics appearing in the 2016–17 opioid sample. This suggests that the tone and content of media coverage of the opioid epidemic has indeed been different from both that of crack cocaine and that of heroin. Comparing the media narrative around heroin to the media narrative on the broader opioid epidemic in both 2016–17 samples indicates that the opioid epidemic has been consistently represented in more medicalized terms than other substance use epidemics.
While the framing around the epidemics has differed in public health versus law enforcement/criminal justice content, it is difficult to conclusively state whether the difference in narratives between the opioid epidemic and other substance use epidemics stems from the perception that the opioid epidemic is uniquely white. The heavy criminal justice narrative found in the 1992–93 methamphetamine sample, which was also a substance use epidemic associated with Caucasians, suggests otherwise. Furthermore, topic modeling indicates that criminal justice framing was more prevalent for the crack cocaine sample than for the methamphetamine sample, although it was present in both. Additionally, there is a distinction between opioid use and heroin use, which may reflect who we expect to be using each substance or our perception between users of prescription and illegal drugs. It could be that in order for the media narrative to shift, two factors were needed: an overall reframing of substance use as a public health issue that occurred at some point after 1992–93 and a perception that most users of the particular substance are white.
Interestingly, previous work analyzing the portrayal of the opioid epidemic may explain the shift in models between the opioid epidemic and other substance use epidemics. Emma McGinty et al. (2016) conducted an analysis of media coverage of opioid abuse from 1998 to 2012, focusing on whether law enforcement solutions (suggesting a criminalization model) or prevention-oriented solutions (suggesting a medicalization model) were proposed. Prior to 2007, news stories were much more likely to focus on law enforcement solutions to opioid use. The gap between the two types of solutions narrowed in 2007–09, and in 2010–12 the two types of solutions were represented with near equal frequency. We know that between 1993 and 2009 prescription opioid overdose admissions for whites increased 7.5 times, outstripping the rates of increase for African Americans (3.3) and Hispanics (3.2), and that since 2008, heroin-related overdose hospitalization rates for whites exceeded that of African Americans (Unick et al. 2013). While the shift in media narratives documented from 1998 to 2012 by McGinty et al. (2016) and then suggested for 2016–17 in the present article may be attributable to a variety of causes, it is likely that the changed demographics of addiction during this time period helped shift the trend away from law enforcement solutions to prevention-oriented solutions.
Media narratives matter because they shape and are bellwethers of solutions to public policy problems. A dominant narrative that substance use is a criminal justice issue is problematic because it can escalate law enforcement interventions that are ineffective and that raise serious civil rights concerns. Furthermore, a criminalization model of substance use reinforced by racial bias can contribute to the high rates of incarceration of people of color, especially African Americans. By contrast, a medicalization model of substance use promotes more effective public health interventions. In the samples we reviewed, the contrast of models was noticeable and lends credence to the popular hypothesis that the opioid epidemic is perceived and framed differently because of the demographic groups it impacts. This is especially notable when contrasted to previous works that documented a shift from criminalization to medicalization in the opioid epidemic media narratives at the same time that the rates of use among white Americans began to skyrocket. These findings, reinforcing the connection between race and public policy responses, may also have implications for other complex public policy issues that involve structural factors and are framed differently depending on the racial group most impacted, such as education policies including busing and welfare programs to address poverty.
We would like to thank the attendees of the Politics of the Opioid Epidemic conference, hosted at Brown University in collaboration with JHPPL, for their great feedback on our initial presentation of this work. Similarly, we would like to thank Eric Patashnik and Susan Moffitt for organizing the conference and for this opportunity. Lastly, we would like to thank our reviewer for his or her extremely helpful, thorough, and thoughtful comments.
One might worry that separating heroin from opioids would bias the results in favor of a medical frame. At the suggestion of a helpful anonymous reviewer, we conducted an analysis of the combined sample of opioid and heroin articles for 2016–17. While the language in this combined sample does shift slightly toward a law enforcement frame, the medical theme continues to dominate as can be seen from the results in appendix A.
We developed a custom list of stopwords that included words associated with newspaper formatting (e.g., section, byline, graphic) in addition to three common libraries of stopwords (onix, SMART, and snowball) found in the R package “stopwords” (Benoit, Muhr, and Watanabe 2017).
We chose not to “stem” or “lemmatize” the words (e.g., substance and substances could be reduced to substanc) for ease of interpretation and integrity to the original texts, especially for bigrams and trigrams.
This is not to say that law enforcement is entirely absent, merely that it is a less prominent topic than health or politics. Adding a fourth topic to the model reveals a law enforcement–focused topic but with relatively low values of beta (i.e., words that occur less frequently) and at the expense of a higher (less desirable) BIC.