Drug Crisis and Voting Behavior

Major changes have reshaped American politics over the last 20 years. A political realignment across education levels and social classes has moved the Democratic Party closer to upper-middle-class, college-educated voters (Goidel, Moreira, and Armstrong Reference Goidel, Moreira and Armstrong2024; Kitschelt and Rehm Reference Kitschelt and Rehm2019; Zingher Reference Zingher2022). In contrast, Republicans gained support among lower-income voters and consolidated a hegemony in rural areas (Aistrup, Mahato, and Morris Reference Aistrup, Mahato and Morris2023). Another major transformation took place during this same period: the use of opioids dramatically increased, causing the deaths of approximately 806,000 people between 1999 and 2023 (CDC 2025). Some of the localities most affected by opioids are also areas where the Republican Party has expanded its support (Goodwin et al. Reference Goodwin, Kuo, Brown, Juurlink and Raji2018).

In this paper, we contribute to the growing literature on recent changes in American politics by showing how overdose death rates affect election results and which voters are more likely to change their voting behavior in response to this increasingly grave situation. Traditionally, research on the opioid crisis in both political science and public health has been of three veins: (1) observational research on factors that correlate with the drug crisis (Goodwin et al. Reference Goodwin, Kuo, Brown, Juurlink and Raji2018; Stokes et al. Reference Stokes, Purtle, Meisel and Agarwal2021; Sun et al. Reference Sun, Graham, Feldmeyer, Cullen and Kulig2023; Weiss and Zoorob Reference Weiss and Zoorob2021), (2) experimental work that focuses on the framing of the crisis and its victims (Benedictis-Kessner and Hankinson Reference Benedictis-Kessner and Hankinson2019; Reference Benedictis-Kessner and Hankinson2024; Gollust and Miller Reference Gollust and Miller2020; Raychaudhuri, Mendelberg, and McDonough Reference Raychaudhuri, Mendelberg and McDonough2023; Sobotka and Stewart Reference Sobotka and Stewart2020), and (3) theoretically driven observational work unrelated to vote choice (Brown and Zoorob Reference Brown and Zoorob2022; Feldmeyer et al. Reference Feldmeyer, Sun, Harris and Cullen2022; Shepherd Reference Shepherd2022). We expand this literature with a theoretical grounding that links the overdose crisis with voting behavior. Based on this theory, we predict that the worsening of the opioid crisis has advantaged Republicans in elections.

We expect that Republicans electorally benefit from the opioid crisis as a result of their unique ownership of this public health issue—an arena usually dominated by Democrats (Clifford Reference Clifford2022; Fagan Reference Fagan2021).Footnote ⁴ Most voters associate the Democratic Party with issue ownership of public health, social security, and healthcare. In effect, Democrats typically deal with the drug crisis as a health issue, arguing for drug decriminalization and supporting health programs to treat addiction (Silwal and Dayton Reference Silwal and Dayton2025; Weiss and Zoorob Reference Weiss and Zoorob2021). Yet Republicans are equally or more likely to discuss the opioid crisis. However, they frame this health problem to fit political issues that the party advocates for, such as crime and border security.

Oregon serves as an extreme example of Democratic decriminalization efforts gone wrong. In 2020, 58.5% of Oregon voters backed a ballot initiative that would decriminalize the personal possession of most drugs. The goal of the broader movement was to reclassify drug addiction as a mental health issue, and not a criminal issue. The initiative mandated the “establishment/funding of ‘addiction recovery centers’ (centers) within each existing coordinated care organization service area by October 1, 2021,” and the redirecting of tax revenues from the legal sale of marijuana to drug treatment facilities. Oregon’s decriminalization effort was followed by a 14%–25% spike in overdose deaths within the state (Spencer Reference Spencer2023). In the meantime, the state did not effectively implement addiction recovery centers. Before decriminalization, Joe Biden won Oregon by 16 percentage points in the 2020 presidential election. After decriminalization, in a gubernatorial election where Republican candidate Christine Drazan made opposition to decriminalization central to her campaign, Democratic candidate Tina Kotek won by only three percentage points (Oregon Public Broadcasting 2022; VanderHart Reference VanderHart2022). Whether the spike in overdose deaths was caused by decriminalization or the inadequate funding of rehabilitation centers is irrelevant; support for decriminalization was a political liability.Footnote ⁵ The Democratic-controlled Oregon State Legislature reversed the voters’ decision within four years and recriminalized possession of hard drugs on April 1, 2024.

Republicans, in turn, have disproportionately campaigned on the opioid crisis. Panels A and B in figure 1 show the distribution of congressional emails containing the words “overdose” and “fentanyl” by the partisanship of the congressperson sending the email.Footnote ⁶ In all periods since 2012, there are more mentions of “overdose” and “fentanyl” by Republican congresspeople than Democratic ones. Not only do Republicans talk about opioids more, but when they talk about opioids, they are more likely to emphasize the need for a more punitive policy. Looking at congressional floor speeches, Weiss and Zoorob (Reference Weiss and Zoorob2021) find that more conservative members of Congress emphasize law enforcement in their speeches about opioids, while more liberal members advocate for better public health.Footnote ⁷ When state legislators tweet about the crisis, Democrats are more likely to discuss funding of health-driven solutions, treatment and recovery, alternative pain-management solutions, and pharmaceutical company accountability, while Republicans are more likely to discuss funding local police initiatives, legislative work on the issue, and the drug trade (Stokes et al. Reference Stokes, Purtle, Meisel and Agarwal2021). At the national level, Republicans and Democrats replicate this policy solution bifurcation in the halls of Congress through floor speeches and public statements (Weiss and Zoorob Reference Weiss and Zoorob2021). Ideologically liberal members are more likely to focus on healthcare; ideologically conservative members, in turn, are more likely to focus on security solutions when they speak on the crisis. In a nutshell, representatives are framing the drug crisis in terms of issues their parties own: healthcare versus security.

Figure 1

Congressional Emails

Source: These data are from congressional e-newsletters, which are compiled by Cormack’s (Reference Cormack2017) DCinbox. Dates in panels A and B are generated by the DCinbox website (https://www.dcinbox.com).

One might initially expect Republicans, despite their best messaging efforts, to be disadvantaged by the opioid crisis, given accusations that their deregulation policies have exacerbated pharmaceutical misuse. However, issue ownership is less about historical culpability than about how effectively parties frame themselves as credible responders to a crisis. Despite potential blame for deregulation, Republicans have strategically reframed the opioid epidemic from a healthcare crisis into an issue of law enforcement and border security (Silwal and Dayton Reference Silwal and Dayton2025; Weiss and Zoorob Reference Weiss and Zoorob2021).Footnote ⁸ This strategic framing is reflected clearly in public perceptions. For example, in August 2023, 56% of Americans identified foreign entities as being most responsible for the opioid crisis (Owens Reference Owens2023). Additionally, as of June 2024, 39% of Americans approved of President Donald Trump’s handling of drug abuse during his first term, while only 25% approved of President Biden’s handling of the same issue (Ballard Reference Ballard2024). This gap underscores Republicans’ comparative advantage in public perception—rooted in their more punitive, security-focused narrative—despite ongoing debate about the efficacy of, or the historical responsibility for, these policies.

Beyond elected officials, these ideological differences are reflected in the electorate. While partisans below the elite level tend to be less consistent in terms of ideological preferences, they do appropriately identify the priorities of each party (Craig and Cossette Reference Craig and Cossette2020; Goggin, Henderson, and Theodoridis Reference Goggin, Henderson and Theodoridis2020), and, in turn, favor the opioid crisis solution proposed by their preferred party. Voters who identify as Republican are less likely to support policies that are in line with the healthcare approach of Democratic elites, even when the policy may help them or someone they know (Benedictis-Kessner and Hankinson Reference Benedictis-Kessner and Hankinson2019).

Beyond the staunch partisans, healthcare policies designed to address the drug crisis may resonate with voters as complex, long-term, and ineffective strategies to combat an urgent situation. Further, incomplete or underfunded policies that aggravate the crisis, such as Oregon’s experience, tend to disseminate the perception that these healthcare strategies are fated to fail. Policies and appeals that resort to the criminalization of drug use, in turn, are easy to understand and catchy. They typically offer voters a simple solution: patrol and imprisonment to fight against drugs. Even though the effectiveness of this strategy is dubious (Baum Reference Baum1996; Cooper Reference Cooper2015; Werb et al. Reference Werb, Rowell, Guyatt, Kerr, Montaner and Wood2011), it spreads the promise of law and order.

Importantly, when ownership of an issue is clear, it can also be persuasive and mobilizing, but only for some portions of the public. Independents are more likely to be mobilized in favor of a partisan stance when ownership is clear and strong (Wright, Clifford, and Simas Reference Wright, Clifford and Simas2022), and previous work has established that Republican messaging on crime and punishment is more consistent across both political and news platforms than equivalent health-policy framing among Democrats (Ash and Poyker Reference Ash and Poyker2024; Porter Reference Porter2022).Footnote ⁹ Hence, most moderate voters are likely to support the “simple” approach provided by Republicans to facing the drug crisis (Kaufman and Hersh Reference Kaufman and Hersh2020; Shepherd Reference Shepherd2022; Sun et al. Reference Sun, Graham, Feldmeyer, Cullen and Kulig2023). As a result, increasing overdose death rates—a proxy for the worsening drug crisis in communities—are expected to increase the Republican vote, especially among independents who are more likely to switch parties in response to the salience of specific policy issues.Footnote ¹⁰

In addition to influencing the vote choice of moderate voters, the opioid crisis may also have a mobilizing effect. Existent research on anger and perceived threats suggests that they should mobilize the members of the group exposed to them (Valentino and Neuner Reference Valentino and Neuner2017; Webster Reference Webster2020). This research is particularly relevant when we consider how voters may react to rising opioid deaths in their communities. Not only are Americans aware of the problem, but certain subsets of Americans also perceive the crisis as a critical threat to their group. Conservative and moderate voters perceive white citizens to be the losers in the opioid crisis (Gollust and Miller Reference Gollust and Miller2020), and, relatedly, white voters are more supportive of action on the opioid crisis when they perceive it to be a crisis particularly damaging to their racial group (Raychaudhuri, Mendelberg, and McDonough Reference Raychaudhuri, Mendelberg and McDonough2023). Given the rising death toll among nonwhite, Hispanic, and Black voters, it is not unreasonable to think they may also view the opioid crisis as a group threat (Drake et al. Reference Drake, Charles, Bourgeois, Daniel and Kwende2020).

As Americans see their community face a worsening rather than improving crisis—despite many political promises otherwise—there may also be rising anger in the population. If they are angry as a result of this increasing exposure to a community-level threat, then they may not just mobilize, but mobilize in favor of more punitive policies as has been observed in other contexts (Gutierrez et al. Reference Gutierrez, Ocampo, Barreto and Segura2019). Given that Republicans have ownership of punitiveness and security in the US, it is likely that this mobilization benefits Republicans. Even among voters already interested in politics, threat perception and anger may also make Republican candidates more attractive. In particular, the overdose epidemic should mobilize conservative voters. As the crisis escalates, those who are sensitive to the issue and lean toward the party that owns it will become more engaged in the election.

While both parties may claim ownership of the opioid crisis, they offer competing frames and policy solutions that vary in electoral effectiveness. Democrats tend to frame the crisis as a public health issue requiring treatment, prevention, and systemic healthcare investments. Republicans, by contrast, frame it as a law-and-order problem, emphasizing criminal justice and border control. We follow recent extensions of issue ownership theory that show voters evaluate not just parties’ competence on shared goals, but also the means they propose to achieve them (Bélanger and Meguid Reference Bélanger and Meguid2008; Van der Brug Reference van der Brug2004). In this context, the Republican frame may be more electorally effective because it offers a simpler, more immediate, and more intuitive response to voters, particularly in high-overdose areas where fear and frustration are elevated.

Relatedly, as Wagner and Meyer (Reference Wagner and Meyer2015) argue, negative ownership can disadvantage or hurt a party. Democrats, even if they own the healthcare approach, may be viewed as less willing or less capable of confronting the crisis through decisive action, because Democrats are seen as ill-suited to respond to what people perceive as a criminal or security threat. This asymmetry is also reflected in elite behavior. As Poljak and Seeberg (Reference Poljak and Seeberg2024) show, parties are more likely to emphasize issues when both ownership and salience align. Republicans appear more willing to actively emphasize the opioid crisis in campaign messaging (Porter Reference Porter2022; Weiss and Zoorob Reference Weiss and Zoorob2021), whereas Democratic candidates often sideline it. The congruence between Republican issue ownership and salience, coupled with the willingness of Republicans to campaign on the issue, motivates our expectation that Republicans benefit electorally from a worsening crisis.

From this theoretical discussion, we expect that Republican candidates will electorally benefit from overdose deaths. Since the Republican Party frames the crisis as a security issue, we predict that the drug crisis mobilizes Republican supporters and makes independents more inclined to cast a ballot for a Republican candidate. At the county level, this expectation implies that drug death rates should increase the percentage of votes for the Republican Party at the expense of both the Democratic Party—which loses the support among independents—and the share of voters who would abstain. Therefore, our first hypothesis predicts the following:

The aggregate nature of election results does not allow us to investigate whether independents and Republican supporters are more likely to change their voting behavior in response to overdose deaths. To overcome this limitation and test which voters are more sensitive to the drug crisis, we combine death rates at the county level with survey data. At the individual level, we expect to observe that drug deaths mobilize Republican supporters:

Finally, our theory also predicts that the drug epidemic makes independent voters more likely to vote Republican. Independents, by definition, are less aligned with either major party and thus are more susceptible to persuasive, issue-based appeals (Bafumi and Shapiro Reference Bafumi and Shapiro2009; Zaller Reference Zaller1992). In contrast, partisans tend to engage in motivated reasoning, filtering new information through existing party loyalties and discounting opposing frames (Bartels Reference Bartels2002; Taber and Lodge Reference Taber and Lodge2006). When an issue becomes salient—especially one as locally visible and emotionally charged as the opioid epidemic—party alignment can condition the type of policy solutions that resonate. Because independents lack strong partisan priors, they are more likely to respond to the party that most actively frames the issue in compelling terms (i.e., the party that “owns” the issue).

In the case of opioids, Republicans have long held ownership over law-and-order issues (Petrocik Reference Petrocik1996), and have consistently framed the drug crisis as a matter of crime and border security rather than healthcare (Weiss and Zoorob Reference Weiss and Zoorob2021). This security-focused framing aligns with broader Republican messaging and may appeal to independents looking for immediate, tangible responses to community threats. Even if these voters do not identify with the Republican Party, they may nonetheless find its framing of the drug crisis more persuasive, particularly in high-exposure contexts. As such, we expect the following:

The next sections discuss the data and empirical strategies to test these expectations.

County-Level Overdose Death Rates over Time

To operationalize the explanatory variable in our empirical models, we use the county-level overdose death rate.Footnote ¹¹ This measure comes from the National Center for Health Statistics (Rossen et al. Reference Rossen, Bastian, Warner, Khan and Chong2022), administered by the Centers for Disease Control and Prevention (CDC). The CDC calculates overdose death rates as the number of overdose deaths per 100,000 residents.Footnote ¹² Overdose deaths are broadly defined to include unintentional or undetermined deaths, suicides, and homicides. This measure includes all drug poisonings. It is estimated, however, that opioids account for approximately 76% of overdose deaths in the US in 2023 CDC (2025).Footnote ¹³ These data span from 2003 to 2021. Since we are interested in the relationship between overdose death rates and vote share or vote choice, we focus on presidential election years: 2004, 2008, 2012, 2016, and 2020 (for replication data, see Moreira, Goidel, and Armstrong Reference Moreira, Goidel and Armstrong2025).

Although the drug crisis was not declared a public health emergency until 2017, overdose deaths began to rise in the late 1990s and early 2000s due to the widespread prescription of pharmaceutical opioids (Maclean et al. Reference Maclean, Mallatt, Ruhm and Simon2020). As opioid addiction and overdose increased, the government cracked down on prescription opioids. Unintentionally, the crackdown on prescription opioids and the emergence of an abuse-deterrent formulation of OxyContin led to a sharp rise in overdose deaths in the 2010s (Alpert, Powell, and Pacula Reference Alpert, Powell and Pacula2018; Bonnie, Ford, and Phillips Reference Bonnie, Ford, Phillips, Bonnie, Ford and Phillips2017; Cicero, Ellis, and Harney Reference Cicero, Ellis and Harney2015).

Panels A and B in figure 2 illustrate the drastic increase in drug overdose deaths experienced by a vast number of US counties.Footnote ¹⁴ From 2004 to 2020, overdose death rates across counties increased on average by 16.5 deaths per 100,000 residents. In 2004, the drug death rate ranged between 1.44 and 47.4, with a mean equal to 8.18 and a standard deviation of 4.64. In 2020, the drug death rate across counties ranged from 4.3 to 142.6, with an average value of 24.6 and a standard deviation equal to 14. What was mostly a regional crisis in the early 2000s—limited to West Virginia, Kentucky, and New Mexico—now plagues a majority of the country, with much of the national increase in drug overdose deaths attributed to the rise of fentanyl in the US (Zoorob Reference Zoorob2019).

Figure 2

County-Level Drug Overdose Death Rates per 100,000 Population

Source: Rossen et al. (Reference Rossen, Bastian, Warner, Khan and Chong2022).

Given the observational nature of our study, it is important to explicitly recognize the potential for endogeneity between opioid overdose deaths and increasing Republican vote share. It is plausible, and even likely, that the relationship between these phenomena is reciprocal or driven by underlying, unobserved factors simultaneously influencing both. For instance, counties predisposed to supporting Republican candidates could differ systematically in ways that also predispose them to higher overdose rates, such as weaker healthcare infrastructure, declining economic conditions, or particular cultural attitudes toward substance use and healthcare. Our models attempt to mitigate this by including extensive county-level demographic and economic controls, lagged dependent variables, rural-urban codes, a time-trend variable, state fixed effects, and robustness checks, including unit fixed-effect models and survey analyses. Although we acknowledge these strategies cannot entirely eliminate concerns of endogeneity, our findings provide strong support for our predictions that an increasing overdose death rate benefits the Republican Party.

Aggregate-Level Analysis: Death Rates and Republican Vote

A quick glance at the data before estimating our models is suggestive of the expected relationship between overdose deaths and Republican vote. Table 1 shows the change in Republican vote share from 2004 to 2020 in the 10 counties that witnessed the smallest increase in opioid overdose rates (panel A) and the 10 counties that witnessed the largest (panel B). In counties that were less exposed to the opioid crisis—where overdoses per 100,000 residents moved from 1.81 to 5.46 on average—Republican vote share rose by 5% on average. In contrast, where the drug crisis was more severe—where the average overdose rate soared from 37.0 to 111.4 per 100,000 residents—Republican vote share rose by 14.4%. In general, Republican vote share increased in US counties from 2004 to 2020 due to the geographical sorting of rural Americans into the Republican Party. However, none of the 10 counties that witnessed the largest increase in overdose deaths became less Republican between 2004 and 2020.

Table 1

Republican Vote Share in Counties with the Smallest and Largest Increases in Overdose Death Rates from 2004 to 2020

Sources: Overdose death rates are available from the CDC (Rossen et al. Reference Rossen, Bastian, Warner, Khan and Chong2022) and the Republican vote is from Leip (Reference Leipn.d.).

We expect that overdose deaths benefit the Republican Party at the expense of both the Democratic vote and the percentage of abstainers. To test our expectation at the aggregate level, we estimate a seemingly unrelated regressions (SUR) strategy in which we split a county electorate into three categories: the percentage of voters who voted for the Republican Party in a presidential election, the share of those who cast a Democratic ballot, and the percentage of abstainers. Following the approach developed by Aitchison (Reference Aitchison1982) to modeling compositional data and implemented by Tomz, Tucker, and Wittenberg (Reference Tomz, Tucker and Wittenberg2002) to study election results in multiparty systems, we define the outcome variables of our SUR model as the $ J-1 $ log-transformed ratios between the percentage of voters in each $ j $ complement of the constituency—votes for the Democratic candidate, for the Republican contender, and abstention—to a baseline category ( $ j=1 $ ).Footnote ¹⁵ Therefore, for county $ i $ in election $ t $ , our outcome variables are specified as follows:

(1) $$ {s}_{jit}=\mathit{\log}\left(\frac{y_{jit}}{y_{1 it}}\right)\forall j\ne 1 $$

where $ {y}_{jit} $ is the share of voters in category $ j $ . We compute the log ratios $ {s}_{jit} $ with data on presidential election results, turnout, and voting-age population by county from 2004 to 2020.Footnote ¹⁶ Abstention is calculated as the county voting-age population minus the total number of votes cast for the Democratic and Republican presidential candidates. As such, we use a measure of effective abstention, which combines votes for third parties and the number of voters who abstained (Moreira Reference Moreira2025). We do so because third-party votes in US presidential elections are wasted votes. Yet casting a ballot for a third party can still express voters’ dissatisfaction with the two major options, the Republican and Democratic candidates. In this sense, abstention and voting for a third party are two options that may represent voters’ indifference to the electoral competition.Footnote ¹⁷

Modeling abstention as an option available to voters is crucial to understanding voting behavior in American elections (Horiuchi and Kang Reference Horiuchi and Kang2018; Moreira Reference Moreira2025; Weschle Reference Weschle2014). For instance, Brown and Zoorob (Reference Brown and Zoorob2022) find that an increase in exposure to disorder (homelessness, crime, and drug use) leads to increases in voter turnout. Yet previous research estimates the effect of opioid overdose deaths on the Republican or Democratic share of the two-party vote (Arteaga and Barone Reference Arteaga and Barone2023; Shepherd Reference Shepherd2022; Siegal Reference Siegal2023; Xiang et al. Reference Xiang, Hou, Rashidian, Rosenthal, Abell-Hart, Zhao and Wang2021). We see abstention, and the trade-off between abstention and party vote shares, as integral to explaining the relationship between the drug crisis and voting behavior. Our theory predicts that overdose death rates mobilize Republican voters and independents to express their preferences for the security approach to addressing the drug crisis proposed by the Republican Party. At the county level, this prediction implies that a set of voters decides to vote for the Republican candidate instead of abstaining in response to higher overdose death rates.

Once we calculate log ratios from equation 1 measuring the trade-off between the Democratic and Republican vote shares— $ {s}_{dit}=\mathit{\log}\left(\frac{Democrati{c}_{it}}{Republica{n}_{it}}\right) $ —and between Republican vote share and abstention— $ {s}_{ait}=\mathit{\log}\left(\frac{Abstentio{n}_{it}}{Republica{n}_{it}}\right) $ —we can estimate the SUR model using the following specification:

(2) $$ {s}_{jit}={\displaystyle \begin{array}{l}{\beta}_{1j} DeathRat{e}_{it}+{\delta}_{kj}{x}_{kit}+{\alpha}_{\boldsymbol{j}}\boldsymbol{Stat}{\boldsymbol{e}}_i\\ {}+\hskip2px {\rho}_j TimeTren{d}_t+{\beta}_{2j} RepPr{e}_t+{\beta}_{3j} IncCan{d}_t\\ {}+\hskip2px {\beta}_{4j} CloseEle{c}_{it}+{\phi}_j{s}_{jit-1}+{\varepsilon}_{jit}\end{array}} $$

where $ DeathRat{e}_{it} $ is the overdose death ratio in county $ i $ in election year $ t $ , $ {x}_{it} $ is a vector of $ K $ economic and demographic variables measured at the county level, $ \boldsymbol{Stat}{\boldsymbol{e}}_i $ is a vector of fixed effects indicating the state where county $ i $ is located, and $ TimeTren{d}_t $ is a continuous variable that counts election years from 2004 to 2020 to control for unobserved factors that might trend over time.Footnote ¹⁸ $ CloseEle{c}_{it} $ captures the distance in vote share between the two major presidential candidates in the state where the county $ i $ is located, $ RepPr{e}_t $ is a dummy variable indicating whether the Republican Party held the presidency at $ t $ , $ IncCan{d}_t $ is an indicator of whether the sitting president was running for reelection at $ t $ , and $ {s}_{jit-1} $ is the temporally lagged dependent variable.Footnote ¹⁹ The lagged dependent variable controls for persistent features of the county that influence voting behavior but cannot be observed (Keele and Kelly Reference Keele and Kelly2006).Footnote ²⁰ The trade-off between categories in the last election can also address concerns about factors that may be gradually changing over time—such as turnout—and could be associated with the uptrend of overdose deaths. Finally, $ {\varepsilon}_{jit} $ is an error term that can correlate across the $ J-1 $ equations being estimated, and the other Greek letters are parameters we estimate.

The county-level controls ( $ {x}_{kit} $ ) in our models include the unemployment rate, the average per capita income, a rural-urban code, the county population, and the following percentages: Black voters, Hispanic voters, Asian voters, young voters (less than 30 years old), the elderly population (more than 70 years old), people with a college degree, and workers in the manufacturing sector.Footnote ²¹ Hence, our models control for major potential confounders in the relationship between drug use and vote choice, such as economic conditions, educational attainment,Footnote ²² and urban-rural characteristics.Footnote ²³

The parameterization of the SUR model is similar to a multinomial logistic regression. For each explanatory variable, the SUR model estimates $ J-1 $ coefficients. Each of these coefficients is the estimated impact of a one-unit increase in the explanatory variable on one of the $ {s}_{jit} $ outcomes, which is in the log-ratio form. Since Republican vote share is the baseline category in our models (the denominator in equation 1), a positive coefficient indicates that its predictor harms the Republican Party in favor of the category in the numerator of the logged ratio (the Democratic Party or abstention). By symmetry, negative coefficients have the opposite interpretation: they benefit the reference category (Republican Party) at the expense of the numerator (Democratic Party or abstention). Although we can use these coefficients in the log-ratio form to make statistical inferences about the direction of the effect, the substantive interpretation of SUR results requires “translating quantities of interest back into their original composition structures” (Philips, Rutherford, and Whitten Reference Philips, Rutherford and Whitten2016, 271); in our case, this involves transforming predicted log ratios, $ {\hat{s}}_{jit} $ , back into predicted percentages of voters, $ {\hat{y}}_{jit} $ . To do so, we use bootstrapping to estimate the impact of a shock in overdose death rates on each category of our compositional outcome—Republican vote share, Democratic vote share, and abstention.

County-Level Findings

Results from our county-level models and subsequent simulations are shown in figure 3.Footnote ²⁴ Panel A shows coefficients for death rates across log-ratio equations. In both equations, the effects are negative, indicating that higher overdose death rates benefit Republican presidential candidates at the expense of both Democratic vote share and abstention. Although the coefficients in panel A indicate directionality and statistical significance, interpreting magnitude is difficult in log-ratio form.Footnote ²⁵

Figure 3

The Aggregate Effect of Overdose Death Rates on Voting Behavior

To overcome this limitation, we use a nonparametric bootstrapping strategy in which we randomly sample with replacement from the data a thousand times and, for each trial, we estimate the difference in predicted vote share for each category—Republican vote, Democratic vote, and abstention—across a range of overdose death rates.Footnote ²⁶ Panel B of figure 3 shows the simulated effects of incremental shocks of 20, 40, 60, 80, 100, and 120 overdose deaths per 100,000 residents.Footnote ²⁷ We acknowledge that even the smallest scenario—a 20-death increase—is substantial, equating to a nearly two standard-deviation increase in our panel data. However, it is important to contextualize the size of these shocks within the real-world experience of American counties. From 2004 to 2020, the average county-level increase in overdose deaths was approximately 16 deaths per 100,000 residents. Specifically, deaths per 100,000 increased by more than 20 in 27.1% of counties (843) between 2004 and 2020. Thus, while a shock of 20 deaths appears large when measured purely by standard deviations, it closely reflects the actual overdose death experience of the average US county in recent elections.

Consistent with our expectations, each incremental increase in overdose deaths is associated with an incremental rise in the Republican vote share. Specifically, an increase of 60 overdose deaths per 100,000 residents corresponds to roughly a 1.5 percentage-point gain in Republican votes.Footnote ²⁸ An increase in county-level overdose death rates by 80 per 100,000 people, in turn, corresponds to a two percentage-point gain in Republican vote share.Footnote ²⁹ It is worth noting that robustness tests in table A9 and figure A2 of the online appendix show that, when a quadratic term is included in the model, there appears to be some plateauing in the effect of overdose deaths on Republican vote share. Specifically, even at very high overdose death rates (e.g., 100 or 120 deaths per 100,000 residents), the Republican vote-share effect seems to level off at around two percentage points. As such, we caution readers not to overinterpret the larger effect sizes at the extremely high shocks in overdose rates.

The statistically significant decline in abstention reveals that there is a mobilizing effect of the worsening opioid crisis. Under an increase of 60 overdose deaths per 100,000 residents, aggregate-level vote share moves about equally—around 0.75% each—from abstention and the Democratic Party to the Republican Party. This suggests that the growing Republican advantage in these counties is not just due to party switching, but partially due to new voters entering the political fray as well. Building on previous research that finds homelessness, crime, and drug use are associated with higher turnout (Brown and Zoorob Reference Brown and Zoorob2022), our findings suggest that an increase in exposure to drug overdose deaths also leads to an increase in voter turnout.

Individual-Level Analysis: Death Rates and Vote Choice

We also expect that overdose death rates mobilize Republican voters and make independents more likely to vote for the Republican candidate (hypotheses 2 and 3, respectively). To test these expectations, we match respondents in the Cooperative Election Study (CES) Cumulative Common Content to their respective county’s overdose death rate data. The CES survey data spans from 2006 to 2022, giving us coverage for the 2008, 2012, 2016, and 2020 presidential elections. Each wave of the survey has approximately sixty thousand respondents and includes the respondent’s county of residence, so we can look at the relationship between overdose death rates observed in the county and vote choice at the individual level.

Vote choice is taken from the following question on the postelection wave of the CES: “For whom did you vote for President?” Response options include the Democratic and Republican presidential candidates, as well as the third-party and other candidates. In alignment with our aggregate-level analysis, we include third-party and other candidate votes as abstention. Thus, our dependent variable consists of three categories: voted for the Republican, voted for the Democrat, and abstained (or voted for a third-party or write-in candidate).

Before testing our expectations, we first estimate a multinomial logistic regression to investigate the effect of overdose death rates on vote choice with the following model specification:

(3) $$ \Pr \left({y}_{rit}=m\right)={\displaystyle \begin{array}{l}F\Big({\beta}_j DeathRat{e}_{it}+{\delta}_{kj}{x}_{kr}+{\delta}_{kj}{z}_{kit}\\ {}+\hskip2px {\gamma}_{\boldsymbol{jt}} Electio{n}_{rt}+{\varepsilon}_{rit}\Big)\end{array}} $$

where $ m $ is the probability respondent $ r $ in county $ i $ and election $ t $ chooses one of the categories of our nominal variable: Republican vote, Democratic vote, or abstention. $ DeathRat{e}_{it} $ is the overdose death rate used in the county-level analysis, and $ {z}_{kit} $ is the vector of $ k $ economic and demographic variables from equation 2. Along with the same county-level variables that are included in the aggregate-level analysis, our individual-level model includes a set of individual-level controls, $ {x}_{kr} $ . This vector contains dummy variables for race (Black, Hispanic, Asian, or other race, such that white is the reference category), gender (one if female), whether the respondent holds a college degree, party identification, and the respondent’s household income.Footnote ³⁰ The model includes election fixed effects, $ Electio{n}_{rt} $ . Survey weights provided by the CES are included in our individual-level analyses.

Second, we estimate the above model with an interaction between party identification and overdose death rates to explore whether Democrats, independents, or Republicans are affected in different ways by the worsening crisis. Predicted probabilities are presented in the main body of the text, and the full results are in tables A4 and A5 in the online appendix.

Individual-Level Findings

Before showing the effect of the drug crisis on vote choice across party identification, we replicate our county-level results with survey data. The regression results are presented in table A4 in the online appendix. Figure 4 presents the predicted probabilities of each category across county-level overdose death rates. The results from the individual-level model conform with our aggregate-level findings. County-level overdose death rates have a positive and statistically significant effect on the probability of voting for the Republican candidate. The predicted probabilities also reveal the same basic trade-off occurring where the likelihood of voting for the Democratic candidate and the likelihood of abstaining declines as the likelihood of voting Republican increases.

Figure 4

Voting Choice across Overdose Death Rates

In fact, as overdose death rates increase, the probabilities of voting Democratic and voting Republican converge until, at extreme levels of overdose death rates (110 overdose deaths per 100,000 residents), the predicted probability of voting Republican, 0.52, surpasses the predicted probability of voting Democratic, 0.44. However, few counties have such high overdose death rates, and only 0.13% of the respondents in our sample reside in such counties. Moving, more realistically, from 10 to 70 overdose deaths per 100,000 in the county (a range that contains 80% of our sample), we see that the predicted probability of voting Republican goes from 0.37 to 0.47.

Which respondents become more likely to vote Republican in counties with higher drug overdose death rates? We predicted that drug deaths mobilize Republicans (hypothesis 2) and make independents more likely to support the Republican candidate (hypothesis 3). To test those expectations, we estimate heterogeneous effects on voting behavior using an interaction between overdose deaths and party identification. Predicted probabilities from this model are presented in figure 5. The regression coefficients from the interaction model, as well as the alternative specifications of our individual-level models, are included in table A5 in the online appendix.

Figure 5

Overdose Death Rates and Vote Choice by Party Identification

Panel A of figure 5 illustrates that, while Republican identifiers do become more likely to vote for the Republican candidate as overdose death rates increase, the movement among independents is most stark. The predicted probability that independent identifiers vote Republican at the low end of overdose death rates is 0.4; at the high end of overdose death rates it is 0.6. At the same time, independent identifiers become significantly less likely to vote for the Democratic presidential candidate (panel B), and more likely to abstain (panel C). Moving from 10 to 70 overdose deaths per 100,000, the probability an independent would vote for the Democratic Party decreases from 0.43 to 0.29. These results support hypotheses 2 and 3. That independents become more likely to abstain runs counter to our expectations, but the magnitude of these changes is smaller than the impact on support for a Republican candidate among independents. Further, it is worth noting that overdose death rates mobilize partisans: Democratic and Republican identifiers become slightly less likely to abstain as the overdose death rate increases. As true independents are a much smaller portion of the electorate, these results are in line with our county-level findings.Footnote ³¹ Yet we observe that drug deaths mobilize not only Republicans, but also make Democrats more likely to turn out.

As with any observational study, one must interpret these results carefully. Although our model specifications control for most factors that could confound the association between drug crisis and vote choice, we cannot rule out the possibility that some endogeneity is still present. However, across all models and robustness checks, we observe that high death rates are associated with an increase in the Republican vote. The survey analyses in this section suggest that the drug crisis tends to mobilize Republican voters and move independents toward the Republican Party.