Increasing Liberalization: A Time Series Analysis of the Public’s Mood toward Drugs

Abstract Previous research suggests that American drug sentiment is becoming more liberal. However, the absence of a reliable and valid over time measure limits our understanding of changes in drug attitudes. This project utilizes the dyad ratios algorithm and 298 administrations of 66 unique survey indicators to develop a measure of public mood toward drugs from 1969 to 2021. I find that drug mood has trended more liberal since the late 2000s. I then test for the predictors and consequences of drug mood empirically using ARMAX modeling. Results suggest that the violent crime rate, presidential rhetoric on drugs, and college attendance are not significant predictors of drug mood, but punitiveness is significant and negative. Moreover, only drug mood emerges as a significant and negative predictor of punitiveness. Granger causality tests indicate that drug mood Granger causes changes in punitiveness. These results elucidate the socio-political dynamics regarding public opinion toward drug policy.


Introduction
On 3 November 2020, Oregon voters passed Measure 110, a law which not only decriminalized low-level possession of most illicit drugs including cocaine, methamphetamine, and heroin, but also diverted hundreds of millions of dollars in marijuana tax revenue to drug treatment and recovery services (Netherland et al., 2022).On the face of it, this law is one of many progressive drug policies recently passed illustrating the role voters play in shaping future drug policy in America.However, I argue this law reflects something more fundamental about the role of public opinion in American drug policy.Prior research suggests that changing public sentiment, otherwise known as "policy mood," is a powerful, multidimensional force in shaping future policies (Pickett, 2019;Stimson, 1999Stimson, , 2004)).Notwithstanding its importance for criminal justice policy (Enns, 2014(Enns, , 2016;;Ramirez, 2013), research has yet to explore policy mood applied to drugs specifically.To fill this gap of knowledge, the current paper introduces the concept of drug mood, broadly defined as public sentiment toward drug policies.Liberal drug mood is support for policies that try to integrate drugs and drug users into mainstream society.
Little attention has been brought to bear on how the public's mood toward drugs evolves over time, which is problematic for three reasons.First, prior research suggests that public opinion of drugs fluctuates "quickly and dramatically" (Beckett, 1997, p. 16;Beckett & Sasson, 2004), implying there are no coherent long-term trends in public opinion toward drug policy.Creating an over time drug mood measure can reveal if this is true, and whether drug mood differs from other sentiments relevant to the criminal justice system (CJS), such as punitive mood (Enns, 2014(Enns, , 2016;;Ramirez, 2013).Second, public opinion literature on drug policy has drawn primarily from cross-sectional survey questions asking individuals their support for specific drug policies or punitive preferences against drug offenders (Kuettel, 2022;Lee & Rasinski, 2006;Lock et al., 2002;Nielsen, 2010).Yet without a consistent over time drug mood measure, it is difficult to know if changes in drug attitudes reflect question-specific measurement error, or an aggregate latent dimension of public opinion.Finally, when considering research showing that public opinion influences policy outcomes (Nicholson-Crotty et al., 2009;Pickett, 2019;Stimson, 2004), changes in drug mood likely play a major role in drug policy reform.As Pickett (2019) explains, it is dynamic and generalized public opinion toward an issue that most strongly influences policy.Examining how drug mood changes over time can offer a clearer picture of how public opinion shapes drug policy.
An aggregate drug mood measure provides a means for answering descriptive questions about public opinion toward drug policy.For example, how has drug mood changed over time?Do certain historical periods shape the level of support for liberal drug mood?Do changes in drug mood fluctuate quickly, as research using non-policy attitudes suggests (Beckett, 1997), or has it followed a relatively stable trend over time?The answers to these questions are important from a theoretical perspective.For instance, a liberal drug mood trend would align with research showing an increase in tolerance toward traditionally deviant activities (ie same-sex marriage) that have recently gained public support (Doble, 2002;Hout, 2021;Schnabel & Sevell, 2017).
A dynamic measure of drug mood also allows for the testing of theories about its sources.Both theoretical and empirical research suggests that rates of violent crime (Beckett, 1997;Beckett & Sasson, 2004;Blumstein & Rosenfeld, 1998), punitiveness (Enns, 2014;Felson et al., 2019;Ramirez, 2013), presidential rhetoric (Hill et al., 2012;Oliver et al., 2011;Stringer & Maggard, 2021), and young Americans enrolled in college (Derefinko et al., 2016;Farnworth et al., 1998;Garland et al., 2012) should be associated with changes in drug sentiment.An over time drug mood measure offers a way to test these theories empirically.
This article first details the history and theoretical rationale for drug mood.An annual drug mood measure is developed for the years 1969 to 2021 using the dyad ratios algorithm and 298 administrations of 66 unique survey indicators.I conduct a series of unit-root tests for stationarity of the drug mood measure and predictor variables of interest.ARMAX modeling and Granger causality tests are used to empirically test theories explaining changes in drug mood.I find that the violent crime rate, presidential rhetoric on drugs, and the percentage of young adults enrolled in college are not significantly associated with drug sentiment.Only punitiveness is significantly (and negatively) associated with drug mood.I conclude by discussing the theoretical and policy implications of these results and provide suggestions for future research.

Conceptualizing Drug Mood
Past research using cross-sectional surveys asking about levels of drug policy support provides a limited understanding of public opinion toward drugs (Bobo & Johnson, 2004;Garland et al., 2012;Lock et al., 2002).This approach is limited because it reflects question-specific responses at a single period, and comparing levels across questions can lead to divergent (and even erroneous) conclusions (Page & Shapiro, 1992).For example, in 1984 public support for legalizing marijuana was low (24% "should legalize"), while support for increasing spending for drug treatment was high (94% "too little spending") (National Opinion Research Center, 1984).One would come to different conclusions about drug policy preferences depending on which question was examined.Instead of focusing on question levels, examining their over time trends provides a clearer sense of public drug preferences, as question trends should be similar if they capture the same phenomenon (Stimson, 1999(Stimson, , 2004)).Stated differently, related policy questions should change similarly over time despite any differences in their absolute levels, and one would get a consistent story about public preferences no matter how a drug policy question was asked to respondents.
It is important to study trends in drug policy attitudes because policy decisions reflect aggregate changes in public preferences that trend one way or another (Pickett, 2019).Policy attitudes are stable by default (Page & Shapiro, 1992;Pickett, 2019), but they are not immovable (Erikson et al., 2002).Broad changes in policy attitudes happen coherently, for a reason, and are substantial in their consequences (Stimson, 1999(Stimson, , 2004)).The theory of dynamic representation says that politicians, policymakers, and elected officials (eg judges, prosecutors) respond to broad changes in policy attitudes because of fear of electoral consequences (Erikson et al., 2002;Soroka & Wlezien, 2010;Stimson, 2004).To increase their chance of reelection, public officials ride the wave of public sentiment and propose policies they believe will be popular in the future.
Generalized mood matters because it symbolizes the collective preferences of the public.Scholars have long noted that public opinion can look disorderly (Drakulich & Kirk, 2016;Zaller & Feldman, 1992), with some claiming that public opinion matters little for drug policy (Beckett, 1997;Beckett & Sasson, 2004).Yet "except perhaps for hermits, there exist no atomistic individuals whose ideas or context of ideas are wholly their own" (Stimson, 1999, p. 2), which implies there is "structure in aggregates" (Stimson, 1999, p. 3).On this point, Stimson (1999) suggests aggregating over issues (ie survey questions) as well as individuals.If responses to different questions all capture the same underlying phenomenon, the percentage in support of (or opposition to) each question should move in tandem, revealing a shared dimension of public opinion known as policy mood (Stimson, 1999(Stimson, , 2004)).Drug mood represents the aggregate signal of public drug preferences through the noise of individuals and survey question items.
Scholars have formerly described a changing national drug mood in America despite it being an approximation for a more complex set of beliefs.Nielsen (2010, p. 483) argues that drug attitudes "clearly have not remained stable over time and change has not been monotonic," finding the public was less liberal toward drugs during President Reagan's and H. W. Bush's administrations relative to other periods.Lock et al. (2002) shows the public has come to prefer drug treatment strategies over criminal justice approaches, a finding that holds across all sociodemographic groups and is consistent with "parallel publics" regarding punitiveness (Duxbury, 2021;Page & Shapiro, 1992;Ramirez, 2013;Shi et al., 2020).Stringer and Maggard (2021, p. 14) argue that "some aggregate portions of society have begun to question the legitimacy of the rhetoric about fear of crime and drugs at least with regard to marijuana."All of these studies suggest the existence of an evolving aggregate public sentiment toward drugs.

Measuring Drug Mood
Researchers have previously relied on questions asking what Americans identify as the Most Important Problem (MIP) to measure the public's drug preferences over time (Beckett, 1997;Beckett & Sasson, 2004;Gonzenbach, 1992).However, as other scholars have noted (Jennings & Wlezien, 2015;Page & Shapiro, 1992;Pickett, 2019), MIP questions "do not necessarily tell us anything, one way or the other, about the stability or instability of policy preferences" (Page & Shapiro, 1992, p. 40).MIP questions reveal what issue is most salient to the public, not their policy preferences or how they change over time.Using MIP questions can even paint a misleading picture of American drug preferences.If an economic recession captures the public's attention, drugs will not be the MIP no matter the rate of use.The public could even be trending in a more liberal direction when few people say drugs are the countries' MIP.It is equally limiting to use one or a few questions to measure public drug preferences (Felson et al., 2019;Nielsen, 2010;Nielsen et al., 2010).Wording differences or other idiosyncrasies unique to those question(s) can suggest artificial trends and bias resulting inferences.
Combining multiple and repeating survey questions together into a single over time measure will provide a more accurate depiction of generalized drug mood.However, a moving average of these items does not account for differences across questions that result from survey effects, such as changes in question wording or missing data due to infrequent administration of each question (Schuman & Presser, 1981;Stimson, 1999, p. 48).Averaging these items together would introduce significant measurement error that could skew resulting inferences.Fortunately, the dyad ratios algorithm (DRA) is a technique that allows for related questions to be aggregated together into a single measure by using changes in survey item ratios over time (Stimson, 1999).Relatedly, if these drug policy questions capture the same phenomenon, we should expect them to move together over time despite any differences in their absolute levels.
Scholars have previously conceptualized measures of policy mood relevant to the CJS, such as punitive mood (Duxbury, 2021;Enns, 2014Enns, , 2016;;Ramirez, 2013;Weaver, 2007).Using 11 survey items for the years 1950 to 1990, Weaver (2007) was the first to generate a punitive mood measure, finding a dynamic public punitiveness that began increasing in the mid-1960s until the 1990s.More recently, Enns (2016) uses 33 survey items to create a punitive mood measure from 1953 to 2012, confirming previous findings that punitiveness increased in the mid-1960s and then declined in the mid-1990s.This body of literature consistently finds an evolving measure of public punitiveness, and that increases in punitiveness played a key role in the adoption of harsh criminal justice policies (Enns, 2014(Enns, , 2016;;Nicholson-Crotty et al., 2009).Yet to the authors' knowledge, this method has not been applied to measuring drug mood.

Predictors of Drug Mood
Past research points to several factors that likely explain changes in drug mood.Drug mood should respond to rates of violent crime.Broadly speaking, drugs and crime (particularly violent crime) are strongly linked in the minds of the public (Beckett, 1997;Beckett & Sasson, 2004;Taylor, 2008).Gang-related violence over drug market territory during the crack epidemic has also been cited as a key factor for the rise in the violent crime rate (Blumstein & Rosenfeld, 1998;Goldstein et al., 1992;Reuter, 2009).Together, this work suggests that as the violent crime rate increases, the public should become more fearful about drug-related violent crime and thus less liberal about drugs.I hypothesize that H1: Greater rates of violent crime are negatively associated with liberal drug mood.
The public's overall level of punitiveness should be associated with changes in drug mood (Felson et al., 2019;Schnabel & Sevell, 2017;Stringer & Maggard, 2021).Public punitiveness represents moral attachment to traditional group rules and reflects an intolerance of individuality and behaviors viewed as incompatible with mainstream society (Durkheim, 1984).Research suggests that recent decreases in punitiveness has led to a more tolerant society (Doble, 2002;Hout, 2021;Schnabel & Sevell, 2017).Schnabel and Sevell (2017) argue that public outcry over maximizing individual freedom and criticisms over government overregulation may have increased marijuana legalization support, while Tonry (2004) believes the public viewing the CJS as too harsh may have led to more tolerant drug attitudes.However, Felson et al. (2019, p. 25) argues that "attitudes about cannabis became friendly before Americans began to feel that the CJS was too harsh," and finds that the public believing the CJS is too punitive only accounts for 14% of the change in marijuana legalization attitudes.Given this research, I hypothesize that H2: Increases in punitiveness are negatively associated with liberal drug mood.
Presidential attention to drugs should partly explain changes in drug sentiment.Research finds that presidential administrations play a key role in shaping media coverage and public opinion on drugs (Hawdon, 2001;Hill et al., 2012;Oliver et al., 2011;Stringer & Maggard, 2021).Using time series analysis, Oliver et al. (2011) show that presidential State of the Union addresses influence public concern about drugs, while Hill et al. (2012) find that presidents do not influence public opinion directly, but only indirectly through media coverage.Unfortunately, these studies use MIP questions to measure public opinion toward drugs (Hill et al., 2012;Oliver et al., 2011), which as mentioned earlier, is problematic since MIP questions are relative and reflect attention to many competing issues (Pickett, 2019).When considering this, I hypothesize that H3: Greater political rhetoric on drugs is negatively associated with liberal drug mood.
Finally, the percentage of young adults enrolled in college should be associated with changing drug sentiment.College is an important developmental period for young adults, exposing them to new drug use experiences and perspectives (Derefinko et al., 2016).These experiences should liberalize attitudes and views about drugs (Farnworth et al., 1998;Garland et al., 2012;Schulenberg et al., 2020).Many individuals will carry these liberal drug attitudes with them throughout their life and are likely to be more liberal toward drugs compared to previous generations.As new generations of students have their own experiences with drugs and develop liberal attitudes, this cycle repeats itself each generation.These over time dynamics suggest that increases in college attendance will create more liberal drug attitudes in society.I hypothesize that H4: More young adults in college will be positively associated with liberal drug mood.

The Current Study
Combining similar survey questions can provide a more accurate measure of how public drug preferences change over time.Using the DRA to create an aggregate measure of drug mood circumvents the problem of inconsistent survey measures that has previously limited scholars studying public opinion from longitudinal analysis.This drug mood measure can be descriptively analyzed to understand its pattern throughout American history.An empirical testing of the predictors of drug mood can provide further insight into the dynamics responsible for changes in drug mood.This paper seeks to build upon prior work studying American public opinion toward drugs by generating an aggregate drug mood measure, while also testing the socio-political factors responsible for changes in this drug mood measure.

Generating a Measure of Drug Mood
Generating an over time measure of drug mood begins by exploring available survey questions about drugs and narcotics.For instance, the National Opinion Research Center (NORC) asked Americans, "Do you think the use of marijuana should be made legal or not?" a combined number of 27 times since 1973.Yet this question only captures one dimension of drug mood-support for legalizing marijuana, and this question has been asked sporadically without any discernable pattern in survey administration, limiting its use as a systematic measure of drug mood.More importantly, other aspects of drug mood likely exist, such as support for funding drug treatment.A reliable and valid drug mood measure would include all possible questions.
Searching the Roper Center Public Opinion Archive produces 66 unique survey questions about drugs asked two or more times between the years 1969 and 2021. 1uestion topics range from support for drug treatment spending, legalizing marijuana, if respondents think it's okay for adults to sell cocaine to their friends, and more.These questions reflect the majority view of what topics were important enough for pollsters to ask Americans about during this period.I use information from 298 administrations of 66 survey items relating to public support for various drug policies from the period of 1969 to 2021.These items reflect raw data of public support for a broad spectrum of drug policies, and when combined together offer a more complete representation of drug sentiment that is more reliable than any drug policy preference by itself.
The next step is to transform this raw data into survey marginals.By transforming the data into survey marginals, the percentage favoring a liberal drug policy choice is now represented as a ratio of change for each item.Ratios of change are comparable both within and across items because ratios are scaled to a common metric, allowing us to meaningfully compare and combine items (Stimson, 1999).Survey marginals are created using the following formula: Here, the percent of responses in support of liberal drug policies is divided by the percent of responses in support, plus the percent of responses not in support, which is then multiplied by 100.The subscript it references item i at time t.I code the survey marginals so that the numerator represents the percentage of responses in favor of any liberal drug policy category or attitude toward drugs.Responses that are not favorable toward drug liberalization (ie support for punishing drug sellers) are only included in the denominator.Neutral or missing question categories do not enter the equation since the goal is to estimate a ratio expressing changes in support for a given policy (Stimson, 1999).After aggregation into T regular time periods, we have a matrix (ie rows × columns) of N items for T periods represented as survey marginals, x it where i indicates variables and t indicates period.This survey marginal data (represented as ratios) can now be used to create a drug mood measure.
I use DRA to create the drug mood measure.DRA is similar to dynamic factor analysis but specifically designed to handle large amounts of missing survey data (Stimson, 1999(Stimson, , 2018b)).This approach uses recursive estimation (forwards and backwards) to provide weighted measurements for years where surveys have inconsistent coverage by assigning greater weights to survey items with larger samples (Stimson, 2018b).This means that missing data within a survey item is imputed based on the weighted average of observable ratios in the data.The DRA then uses a smoothing technique for the final measure which minimizes within sample forecast error.DRA is essentially an exercise of taking large numbers of empirically observed ratios within and between items, and then averaging over all of them (Stimson, 2018b, p. 14).
The drug mood measure is generated using the extract function in R (Stimson, 2018a). 2 This yields a final weighted average metric (mean = 50.27;std.dev = 3.82) that explains 76.2% of the variance in the latent measure.As indicated by factor loadings of .8 or greater, a majority of the items load strongly onto the latent drug mood measure.Even support for drug treatment, an opinion that might relate more to political ideology than drug attitudes, shows a commonality with the drug mood measure at .72.Descriptive statistics for each of the 66 items is in the online appendix, which is available with the online version of this publication.
Figure 1(a) plots seven survey item questions from the larger number of questions used to generate the drug mood measure in their natural metric form, including support for increasing spending on drug treatment and support for legalizing marijuana.Higher values indicate greater support for the liberal drug policy choice.I report these questions because they reflect distinct yet related drug questions that were asked during the longest time span available.To make the series more comparable visually, Figure 1(b) mean-centers each item to standardize them onto a common scale.I also add the standardized drug mood measure (as indicated by the thick black line).With minor fluctuations between them, each survey item follows an increasing trend over time.This is what we would expect if these items reflect a common measure of policy mood.
Figure 2 presents just the drug mood measure, with higher values signifying greater support for drug liberalization.I include horizontal lines and shaded areas indicating major drug legislation and events, all of which are labeled.The dynamic nature of drug mood is striking and notably different from other mood measures such as punitiveness (Enns, 2014(Enns, , 2016;;Ramirez, 2013).It also does not fluctuate quickly and dramatically as prior work suggests (Beckett, 1997).
Drug mood trends in a liberal direction at the end of the 1960s, a period described as a "counterculture movement" where young adults experimented with drugs (Aikins, 2015).The 1970s marked a decline in drug liberalization, most notably after President Nixon passed the Controlled Substances Act in 1971.In 1975, drug mood reached its historic low before trending more liberal slightly thereafter.Drug mood then became less liberal throughout the 1980s, as President Reagan ratcheted up the drug war to signal an outright condemnation of drug use (Beckett, 1997).After matching its historic low in 1990, drug mood rebounded during the mid-1990s when California passed medical marijuana.In 1999, the first wave of the opioid epidemic began (Centers for Disease Control and Prevention, 2022), and drug mood started to trend more liberal around 2009.Drug mood continued to trend liberal as Colorado and Washington passed recreational marijuana in 2012.In 2020, Oregon decriminalized possession of small amounts of drugs (Netherland et al., 2022), and drug mood reached its liberal peak in 2021.A time series analysis can offer further insight into the dynamics responsible for changes in drug mood.

Time Series Analysis: Methods
Detailed descriptions of the predictor variables are located in the online appendix.As discussed in detail above, drug mood should theoretically respond to rates of violent crime in America.To measure violent crime, I use the violent crime rate per 100,000 persons as collected by the Uniform Crime Report (Federal Bureau of Investigation, 2020).The Uniform Crime Report (UCR) is a commonly used source for capturing the violent crime rate in America (Anderson et al., 2017;Shi et al., 2020).I elect to use the violent crime rate over the homicide rate because the former measure is better suited to capture the full extent of drug-related violent crime.Unlike the homicide rate, the violent crime rate captures occurrences of drug-related violence which did not end in a fatality.Nevertheless, both the violent crime rate and homicide rate move in very similar patterns over time.Notwithstanding issues using UCR data (Lauritsen et al., 2016), I elect to use the UCR over the National Crime Victimization Survey (NCVS) because the NCVS only began data collection in 1973 after the start of the drug war, limiting its benefit over the UCR for the current analysis.
To capture public punitiveness, one possible indicator comes from Ramirez (2013), who uses the DRA to create a punitive mood measure using questions like death penalty support.Enns (2014) also generates a punitive mood measure by adding questions about faith in police and court institutions.Unfortunately, neither of these punitive mood measures extend beyond the year 2010 when drug mood begins its upward liberal trend.However, support for the death penalty is the main set of questions used to measure generalized public punitiveness (Anderson et al., 2017;Enns, 2014;Pickett, 2019;Ramirez, 2013), and is distinct from drug attitudes (Bobo & Johnson, 2004).Research also shows that individuals who support the death penalty are more retributive and likely to support other punitive criminal justice policies (Anderson et al., 2017;Baumgartner et al., 2018).Thus, to measure punitiveness I use annual data on support for the death penalty administered by Gallup survey house.Higher values indicate greater punitiveness.
Scholars have shown that drug attitudes reflect frames constructed by political elites (Beckett, 1997;Beckett & Sasson, 2004).Greater political rhetoric and attention paid to the "drug problem" should be associated with less liberal drug mood.To measure political rhetoric, I use a measure of presidents' State of the Union (SOTU) addresses taken from Shi et al. (2020).This measure captures the number of times the president mentions the words "drug" or "crime" in a SOTU speech, with higher counts indicating greater rhetoric on drugs.I include mentions of "crime" because presidents often associate drugs with crime (Beckett, 1997, p. 44).Research shows that presidents are one of the most visible political figures in the country (Manza & Cook, 2002), and SOTU speeches are widely viewed by the American public, making them a major force in shifting national priorities toward issues like drugs (Oliver et al., 2011).
Lastly, drug mood should trend liberal as college enrollment increases.College offers people many opportunities to be exposed to liberal viewpoints and situations of drug use.This should increase liberal drug attitudes in the population over time, and the percent of young adults enrolled in college is a valid indicator of this phenomenon.I use 2020 data from the National Center for Education Statistics (NCES) capturing the annual percentage of those 18-24 enrolled in college. 3The NCES is the main federal entity for data related to education in the United States (National Center for Education Statistics, 2020).Higher values indicate greater enrollment.

Time Series Analysis: Estimation and Results
I follow the ARDL-bounds procedure in Philips (2018, p. 233) to determine the appropriate method for the time series analyses. 4I use Augmented Dickey-Fuller, Dickey-Fuller GLS, Phillips-Perron, Elliot-Rothenberg & Stock, and Kwiatkowski-Phillips-Schmidt-Shin unit-root tests for stationarity.Results show the dependent and independent variables are either stationary or nonstationary of a first order (I[1]), so the ARDL-bounds test for cointegration can proceed (Philips, 2018).The ARDL bounds test reveals the nonstationary regressand and regressors are not cointegrated, meaning the series must be analyzed in first differences.While a supplementary Johansen test suggests there is up to 1 cointegrating relationship, this result is likely due to the low power and type-1 error when testing relatively short series (Philips, 2018).Moreover, the Johansen test is only appropriate when all series are nonstationary (Enders, 2014), which is not the case here given the unit-root test results suggest key regressors are stationary.I thus conclude there is no cointegration.
Examining the errors from the first difference of the regressand, using the autocorrelation function and partial autocorrelation function, indicates that drug mood follows a first-order autoregressive process.To account for the AR(1) data-generating process, I use the "arima" command in Stata to estimate ARMAX (1, 0, 0) models in first differences.This produces white noise residuals. 5With 51 observations and 4 regressors in the fully specified ARMAX model in first differences, this aligns with prior recommendations and results from simulation studies for achieving adequate statistical power (Krone et al., 2017, p. 13;McCleary et al., 1980, p. 20;Warner, 1998, p. 2) while also avoiding concerns of overfitting (Philips, 2018).
Table 1 shows the results of three ARMAX (1, 0, 0) models predicting drug mood.Model 1 only includes the rate of violent crime, model 2 adds the punitiveness measure, and model 3 includes political rhetoric and percent of young Americans enrolled in college.These models examine the period from 1970 to 2020.
Looking at model 1, there are some noteworthy findings that warrant scrutiny.The rate of violent crime is negative and not significantly associated (b = −.009,p = .099)with changes in drug mood.This insignificant finding is surprising, given the strong association over the past several decades between drugs and crime in the public consciousness (Beckett, 1997;Beckett & Sasson, 2004;Taylor, 2008).It is possible that attitudes toward drugs and crime have become separated in the minds of the public in recent decades, driven by contexts or frames portraying drug use in a more liberal manner.Turning our attention to model 2, I add punitiveness while retaining the violent crime rate in the model.Violent crime is insignificant (b = −.008,p = .181),but punitiveness (b = −.113,p = .005)is significant and negatively associated with 5 These models are estimated using maximum likelihood with the Kalman filter and include one lag of the outcome variable (Δ drug mood).I also use robust standard errors to account for heteroskedasticity of the errors.
Table 1.Period-level predictors of drug mood: arMaX (1, 0, 0) models in first differences.changes in drug mood.It seems that as punitiveness toward criminals declines, drug mood trends liberal.
In model 3, I add political rhetoric and the percentage of young adults enrolled in college, while retaining rates of violent crime and punitiveness.Interestingly, neither political rhetoric (b = .024p = .205)nor college attendance (b = −.238p = .111)are statistically significant.This finding is especially surprising when considering the president has historically played a large role in shaping the national agenda and priorities with respect to drugs (Hawdon, 2001;Hill et al., 2012;Oliver et al., 2011;Stringer & Maggard, 2021).These findings suggest that neither the rate violent crime, presidential rhetoric, and the percent of young adults enrolled in college significantly predict changes in drug mood.Yet punitiveness remains significant and negatively associated (b = −.131p = .002)with changes in drug mood when included in the full model.
These results lead me to reject null hypothesis two and fail to reject null hypotheses one, three, and four.The main implication is that increasingly liberal drug attitudes may be the result of a less punitive public.This finding makes sense when reflecting on the bi-partisan push to reform harsh criminal justice drug policy and reverse mass incarceration.Overall, the results suggest that society is more tolerant of drugs in general as they become less punitive.
It is possible there exists reciprocal relationships among the series, such as punitiveness influencing drug mood and vice versa.To evaluate this idea, I conduct a Granger causality test of drug mood and punitiveness.Granger causality tests are the standard statistical approach to address the question of causal ordering of variables in time series data (Freeman, 1983;Nicholson-Crotty et al., 2009;Stock & Watson, 2001).If the relationship between two variables is statistically significant at the .05level, we can infer that one variable Granger causes another, and that variable X helps us predict variable Y above and beyond past values of series Y alone.
To see if punitiveness Granger causes drug mood, I regress lagged values of drug mood onto punitiveness while including lagged values of drug mood, with both series first differenced.Interestingly, punitiveness does not Granger cause drug mood (p = .712).Next, to determine if drug mood Granger causes punitiveness, I regress lagged values of punitiveness onto drug mood while including lagged values of punitiveness, with both series first differenced.I find that drug mood does in fact Granger cause punitiveness (p = .037),which suggests the direction of causality runs from drug mood to punitiveness.The main implication of this finding is that the public has become less punitive because public opinion about drugs has trended more liberal.

Supplemental Analyses
I estimate several supplementary models to test the robustness of my main findings. 6irst, I replace the UCR violent crime rate with the homicide rate.The conclusions are the same.Punitiveness is significant at the .001alpha level (b = −.143,p = .000)when homicide rate is included in the full model.Second, to account for the possi-bility that drug attitudes are driven by the link between property crimes and drug addiction, I replace the UCR violent crime rate with the UCR property crime rate (per 100,000) and examine its association with drug mood.Again, the conclusions are the same; punitiveness remains significant (b = −.137,p = .001)when the property crime rate is included in the full model.Next, I replace the president SOTU measure with a TV drug coverage measure.This measure comes from the Vanderbilt Television News Archive and is equal to the average number of stories mentioning the word "drug war" or "drug abuse" in ABC, CBS, and NBC evening news from 1969 to 2020 (Cronbach reliability coefficient = .90).The conclusions are unchanged, with punitiveness remaining significant (b = −.116,p = .008)when the TV drug measure is added to the full model.I also replace the college enrollment measure with another measure capturing the percentage of the total U.S. population enrolled in college.Again, only punitiveness emerges as significant (b = −.129,p = .001).
It is possible that rates of drug use predict changes in drug mood.However, to the author's knowledge, there exists no consistent measure of drug use for the entire period under study .For example, data from the National Survey on Drug Use and Health cannot be validly compared across years because of changes to its methodology.7Furthermore, Monitoring the Future (MTF) data is only available from 1976 until 2020 and captures rates of drug use among highschoolers which is not necessarily representative of the adult population.Despite these limitations, I estimate a bivariate ARMAX model (1, 0, 0) predicting drug mood using a MTF measure capturing the percentage of 12th graders reporting any illicit drug use in the previous year.The MTF coefficient is in the expected direction and approaches statistical significance (b = .108,p = .063).This result supports the face validity of drug mood as a valid measure of public opinion toward drugs and implies that rates of drug use may help predict drug mood, but more research using other drug use data is needed before drawing firm conclusions.Moreover, fear of crime may act as a mediator between crime rates and drug mood.To test this, I estimate another bivariate ARMAX model (1, 0, 0) predicting drug mood using a fear of crime measure from the General Social Survey.Results show that fear of crime is not a significant (b = −.024,p = .754)predictor of drug mood.
My main findings show a relationship between punitiveness and drug mood, and the Granger causality tests suggest the direction of causality runs from drug mood to punitiveness.To confirm this finding, I include a supplemental analysis estimating an ARMAX (1, 0, 0) model predicting punitiveness, with drug mood and the other regressors included. 8able 2 presents an ARMAX (1, 0, 0) model predicting punitiveness with rates of violent crime, drug mood, presidential rhetoric on crime and drugs, and college enrollment as predictor variables.The findings confirm that drug mood is the only significant predictor variable of punitiveness (b = −1.26p = .002)when included with the other independent variables.This supplementary analysis reveals that drug mood significantly predicts punitiveness.When combined with the results from the earlier Granger causality tests, this supplementary analysis confirms that the direction of causality runs from drug mood to punitiveness.

Discussion and Conclusion
Since the beginning of the war on drugs nearly 50 years ago, public opinion of drug use has changed slowly but drastically, with recent polls suggesting that 83% of American voters view the drug war as a failure (Franklin, 2021).It is vital to study changes in drug attitudes because policymakers react to broad shifts in public sentiment (Erikson et al., 2002;Stimson, 1999Stimson, , 2004)), which not only has crucial ramifications for future criminal justice policy, but for society in general in the form of mass incarceration (Duxbury, 2021;Enns, 2014Enns, , 2016;;Pickett, 2019;Ramirez, 2013;Weaver, 2007).As drug attitudes continue to liberalize, more states will likely decriminalize drugs and unravel punitive drug war policies.These changes are part of a larger reform to the CJS and highlight the value of studying changes in drug mood.This study reveals three key findings which improve our understanding of American drug sentiment.
The first key finding is the presence of a dynamic drug mood measure that has become more liberal in recent decades.Prior literature using MIP drug questions depict a trajectory that fluctuates quickly and dramatically (Beckett, 1997;Beckett & Sasson, 2004).Instead, the current findings show that drug mood ebbed and flowed gradually over the years, then began trending more liberal starting in the late 2000s.Relatedly, prior work suggesting drug attitudes are dynamic has typically only used a single or a few drug policy questions (Felson et al., 2019;Nielsen, 2010;Nielsen et al., 2010), limiting its generalizability to drug attitudes in general.This study confirms that a dynamic generalized drug mood exists and changes coherently over time.
The second major finding is that existing theory is unable to account for changes in drug mood.No evidence was found to suggest that crime rates, presidential rhetoric, and college attendance significantly predict changes in drug mood.Greater theoretical development on the sources of drug mood is needed.Future research should explore the role of individual rights and bodily autonomy, as support for same-sex marriage, gambling, and liberal drug mood may all stem from a broader socio-political movement among electoral democracies emphasizing greater individual rights and self-expression (Euchner et al., 2013;Ferraiolo, 2014;Schnabel & Sevell, 2017).Future work may also consider Hawdon (1996), who argues that rates of social mobility is related to changing drug sentiment.As rates of social mobility (ie immigration, occupational mobility, etc.) in society increase, the potential groups one may belong to also increase, which in turn dissolves moral boundaries between groups, along with ideas such as drug use being defined as right or wrong.This argument dovetails with Gerber and Jackson (2016) who find that punitiveness reflects a strong preference for authority and the need for a delineated moral order.The third major finding is that liberal drug attitudes appear to help explain why the public has become less punitive in recent decades.Stated differently, the well-documented decline in public support for punitive criminal justice policies over the last 25 years (Enns, 2014(Enns, , 2016;;Ramirez, 2013) appears to be a function, in part, of the public's increasingly liberal attitudes toward drugs.This makes sense because many harsh criminal justice policies were sold to the public as part of the war on drugs and to deal with drug-related crime.Liberal drug attitudes likely underpin the public's view that the drug war is a failure and that harsh crime policies are counterproductive if not inhumane.This finding is consistent with Felson et al. (2019), who argue that "tolerant views of cannabis may have led to a preference for a more lenient criminal justice system that was heavily focused on cannabis possession." Future research should also investigate whether drug mood impacts criminal justice policy.Increasingly liberal drug attitudes has likely motivated criminal justice policy reform over the past few decades.Researchers should explore if liberal shifts in drug mood galvanized the adoption of progressive drug policies such as harm reduction programs (McGinty et al., 2018).Similarly, the responsiveness of criminal justice policymakers to public opinion might vary in strength over time (Pickett, 2019).Some evidence suggests that public officials' responsiveness to public opinion has been stronger over certain periods and may have declined in recent decades (Jacobs & Shapiro, 2000;Lax & Phillips, 2009;Pickett, 2019).Future work should explore the responsiveness of criminal justice policymakers to changes in drug mood.
There are limitations in this study that need to be acknowledged.The drug mood measure only uses survey items that were available, but there is no guarantee that these questions in particular give us a "true" representation of Americans' drug mood.This is not a major concern, however, since most items move in similar patterns over time, and thus likely capture the same phenomenon.Another limitation is the time series measures.Punitiveness only includes a single item asking about death penalty support.Yet, death penalty attitudes contribute a significant amount of variance and are strongly correlated to previous punitive mood measures (Enns, 2014(Enns, , 2016;;Pickett, 2019), which implies these findings are robust to other measures of punitiveness.Moreover, the presidential rhetoric variable counts the number of times "crime" or "drug" is mentioned, which may not capture the nuance of the message.An improved political rhetoric measure would capture the message content using sentiment analysis or a similar technique.Future studies should replicate these findings with other punitive and political rhetoric measures.
This study provides a novel drug mood measure that can be used to better understand the complex realities of American drug policy preferences and their socio-political consequences.Drug mood plays a key role in shaping punitive attitudes and offers a new explanation as to why support for "tough-on-crime" policies has declined in recent years.Liberalizing drug attitudes is one approach policymakers may consider to grow support for nonpunitive policies.While prior research has indicated that declining crime rates have made the public less punitive (Enns, 2014(Enns, , 2016)), it seems that liberal drug mood warrants a closer inspection.

Figure 1 .
Figure 1.seven indicators of public support for drug policies and drug mood, 1969-2021.