Hidden Costs of Ban the Box Laws: Unraveling the Effects on Drug‐Related Deaths

ABSTRACT

Ban the Box (BTB) laws delay criminal background checks until the later stages of the hiring process. This study provides new evidence that BTB laws that apply to both private and public employers have negative spillover effects beyond labor market outcomes. Drawing on data from the National Vital Statistics System (NVSS), I investigate the impact of BTB laws on drug‐related mortality. Two years after adoption, BTB laws are associated with more than a 35 percent increase in drug‐related mortality among Black and Hispanic men. The main mechanism appears to be diminished labor opportunities. BTB adoption reduces wages and the probability of employment among Black and Hispanic men.

1. Introduction

Drug overdose deaths in the United States have risen dramatically over the past decade. In 2010, there were, on average, 105 deaths involving drugs every day. By 2020, that figure had risen to 252, establishing the current drug epidemic as the deadliest in US history (Hedegaard et al. 2021; Rees et al. 2019). Apart from the pain and suffering caused by substance abuse, the economic costs attributable to opioid use disorder alone were estimated to be $1.02 trillion in 2017 (Luo et al. 2021). These economic costs included reduced quality of life, fatal opioid overdoses, and the loss of work productivity.

Prescription drug monitoring programs (PDMPs) and Naloxone Access Laws (NALs) have been adopted by most states in an effort to tackle the ongoing opioid epidemic in the United States. There is some evidence that PDMPs and NALs are effective (Patrick et al. 2016; Rees et al. 2019), but there is less consensus regarding the social and economic determinants of illicit drug use. Based largely on descriptive evidence, Case and Deaton (2017) argue that poor labor market conditions have increased drug overdoses, suicides, and liver disease among White non‐Hispanic men without a college degree. By contrast, Ruhm (2019) and Currie and Schwandt (2021) argue that economic conditions and labor market opportunities are, at best, only weakly predictive of the drug epidemic.

Leveraging the staggered adoption of Ban the Box laws that apply to both private and public employers (hereafter, BTB laws) during the period 2005–2020, this study provides new evidence on the causal relationship between labor market conditions and drug‐related mortality. BTB laws delay criminal background checks until the later stages of the hiring process. Their proponents argue that BTB laws improve the employment prospects of individuals with prior convictions (hereafter, “ex‐offenders”). On the other hand, delaying criminal background checks may increase statistical discrimination against demographic groups who are more likely to have a criminal history (Agan and Starr 2018; Doleac and Hansen 2020).

In theory, the relationship between BTB laws and drug‐related fatalities, at least in part, depends on the impact of BTB laws on labor market outcomes. BTB laws could discourage the use of drugs and reduce drug‐related fatalities among ex‐offenders by improving their job prospects and increasing the opportunity cost of non‐productive, often illicit, activities. On the other hand, if BTB laws diminish employment opportunities for racial and ethnic minority groups, then their adoption could increase drug‐related fatalities due to an increase in cumulative disadvantage, mental distress, and despair among these demographic groups (Case and Deaton 2017, 2022). Moreover, BTB laws may increase drug‐related mortality by lowering the future costs of drug‐related crimes.

Using state‐year level data from the National Vital Statistics System (NVSS) and employing an event study design, I find that one year after adoption, BTB laws are associated with a 22 to 27% increase in drug‐related mortality among Black and Hispanic men (i.e., 2.7 to 4.5 deaths per 100,000 relative to the pre‐treatment means). The association between BTB laws and drug‐related mortality increases in magnitude during the first 2 years after BTB implementation but weakens in magnitude in subsequent years. A possible explanation for this pattern of results might be a change in the pool of individuals susceptible to the effects of BTB laws. In particular, the higher number of drug‐related fatalities in the early stages of policy adoption might have contributed to a decrease in the number of individuals susceptible to the policy in subsequent years.

Additionally, I find that BTB laws are positively associated with both opioid‐related mortality and mortality from stimulants such as methamphetamine and cocaine. Similarly, the results show a positive relationship between BTB laws and drug‐related mortality among both younger (15–34) and older (35–64) Hispanic and Black men. These findings suggest that the impact of BTB laws is not limited to a specific age group or type of substance but affects a broad range of individuals within racial and ethnic minority groups.

To explore channels through which BTB laws may affect drug‐related mortality, I utilize data from the American Community Survey (ACS) for the period 2005–2020. My results confirm those of previous studies, including Doleac and Hansen (2020) and Sabia et al. (2021). By examining the effect of BTB laws on labor market outcomes of individuals without a college degree, I find that BTB decreases employment of Black men by 1.2% points (1.45%) and Hispanic men by 0.73% points (0.8%). Moreover, BTB laws are associated with reduced wages and a lower probability of full‐time employment among low‐skilled Black and Hispanic men. I also find evidence that BTB laws reduce the wages of low‐skilled White males, but the estimated effects are smaller in magnitude and less precise.

Overall, there are two key findings of this study. First, I find that BTB laws decrease job opportunities among low‐skilled racial minorities. Second, as a consequence, BTB laws raise drug‐related mortality among these demographic groups. Together, the findings from this study contribute to the literature on the unintended consequences of BTB laws on labor market outcomes and the effects of labor market outcomes on drug‐related mortality. ¹ The results of this study are broadly consistent with those documented by Pierce and Schott (2020), who exploit trade shocks to investigate the relationship between labor market outcomes and drug‐related mortality during the period 1990–2013. ² Their findings are of a smaller magnitude compared to the findings presented in this study. This might be because the drug epidemic after 2015 was characterized by the abuse of more lethal substances such as fentanyl.

The remainder of the paper is structured as follows. In Section 2, I provide background information on BTB laws and the relationship between labor market outcomes and drug use in the United States. I also discuss the reasons why BTB laws could affect drug use. In Section 3, I describe my data and empirical strategy. Baseline results and robustness checks are provided in Section 4. Section 5 concludes.

2. Background

2.1. Ban the Box Laws

Criminal background checks used to be a common practice during the hiring process in the United States. A 2012 survey conducted by the Society of Human Resource Management found that about 70% of employers in the US were inquiring into applicants' conviction history (Sabia et al. 2021). Since a felony conviction reduces the probability of employment (Holzer et al. 2006; Holzer 2009), opponents of criminal background checks argue that they are highly discriminatory against ex‐offenders.

In the United States, the practice of delaying access to applicants' conviction histories is commonly known as Ban the Box legislation. The first BTB policy in the US was adopted in Hawaii in 1998. This policy prevented employers from inquiring into applicants' criminal records until a conditional offer was made and applied to both the private and public sectors. As of 2020, 37 states and over 150 cities and counties had implemented similar laws in the public sector. ³ BTB laws in 15 states and 22 cities and counties have been extended to private‐sector employment. The timing of when a criminal background check is permitted can vary, with some laws allowing for such checks after the initial job application is submitted, after an interview, or after a conditional job offer has been made. This study focuses on the effect of BTB policies that apply to both public and private employers, but I also investigate the heterogeneity of the effects by type of law. ⁴

Table 1 reports the effective dates of BTB laws that apply to all employers, including those in the private sector. Most states adopted these laws after 2014. ⁵

TABLE 1.

BTB laws applicable to all employers, 1998–2020.

Year effective	State\DC
1998	Hawaii
2010	Massachusetts
2014	District of Columbia
2014	Minnesota
2014	Rhode Island
2015	Illinois
2015	New Jersey
2016	Oregon
2017	Connecticut
2017	Vermont
2018	California
2018	Washington
2019	Colorado
2019	New Mexico
2020	Maryland
2020	Virginia

Note: Based on the National Employment Law Project. Virginia's BTB law only prohibits employers from obtaining information about arrests, criminal charges, or convictions for simple marijuana possession.

2.2. Previous Studies

The main aim of BTB laws is to reduce discrimination against ex‐offenders by giving them a chance to compete for employment opportunities rather than being immediately disqualified based on their conviction history. However, the evidence on the relationship between criminal background checks and employment opportunities for ex‐offenders is mixed. Craigie (2020) and Shoag and Veuger (2021) provide evidence that criminal background checks are negatively associated with labor opportunities for ex‐offenders. On the other hand, Jackson and Zhao (2017b) find that limiting access to applicants' criminal histories has negative effects on the employment of ex‐offenders. ⁶

However, even if BTB policies improve employment opportunities for ex‐offenders, there is a concern over statistical discrimination against demographic groups that are, on average, more likely to have criminal records. Most studies, with the exception of Kaestner and Wang (2024), find that BTB laws increase statistical discrimination against racial and ethnic minorities. By randomly assigning a felony conviction, Agan and Starr (2018) show that BTB policies widen an already existing call‐back gap between White and Black applicants sixfold. Doleac and Hansen (2020) utilize geographic variation in BTB laws and find that these laws are associated with a decrease in the probability of employment for young Black and Hispanic men without a college degree. ⁷ These findings are consistent with the previous literature that examined the employment effects of performing criminal background checks in the early 1990s, as well as other worker screening practices such as drug testing and credit history checks (Holzer 2009; Wozniak 2015; Bartik and Nelson 2016).

This study adds to the literature on the consequences of (delaying) criminal background checks. To the best of my knowledge, it is the first paper to provide evidence on the effects of criminal background checks on risky behaviors, particularly on drug‐related deaths, which is the most severe consequence of drug usage.

The main channel through which BTB policies may affect drug‐related mortality is through their effect on labor opportunities. There has been a long debate on the impact of employment opportunities on health and risky behaviors. The earlier literature shows that recessions are good for health. There is evidence that economic downturns reduce liquor consumption, alcohol‐involved driving, smoking, obesity, and mortality from all causes of fatalities but suicides (Ruhm 1995, 2000, 2005). The author argues that a drop in income during economic downturns outweighs the psychological responses due to stress caused by recessions and results in decreased spending on alcohol and cigarettes. In another related paper, Ruhm (2019) argues that even though mortality from drugs is positively associated with economic downturns, this relationship is rather spurious than causal.

However, there is a growing recent literature that documents a positive association between economic downturns and drug‐related deaths. Case and Deaton (2017) suggest that the worsening of labor market conditions for White non‐Hispanic men without college degrees contributed to an increase in deaths from suicide, alcohol‐related, and drug‐related causes. The authors hypothesize that this is due to a lack of jobs that provide prospects and a sense of belonging (Case and Deaton 2022). These papers provide a valuable hypothesis on the deaths of despair, which is evaluated by several quasi‐experimental studies.

Charles et al. (2019) document that a structural change in the manufacturing sector, which resulted in employment losses of less educated individuals, is associated with an increase in drug‐related deaths and the use of prescription opioids. Similar results are reported by Autor et al. (2019), Pierce and Schott (2020), and Adda and Fawaz (2020), who exploit trade shocks as a natural experiment to examine the effect of employment outcomes on substance abuse and mortality from drugs. Finally, Carpenter et al. (2017) provide mixed evidence on the association between economic conditions and drug use. The authors find that the occurrence of substance use disorders related to marijuana, analgesics, and hallucinogens tends to increase during periods of economic downturns, but the use of cocaine, crack, and LSD is highly pro‐cyclical.

Exploiting the adoption of BTB laws, this paper utilizes a novel natural experiment to provide evidence regarding the relationship between labor market outcomes and drug‐related fatalities. A distinguishing aspect of this study is that BTB laws predominantly affect Black and Hispanic individuals in the labor market. This allows the study to provide evidence regarding the relationship between labor opportunities and drug‐related mortality among racial and ethnic minority groups.

3. Data and Empirical Strategy

3.1. Data

3.1.1. Drug‐Related Mortality Data

To investigate the relationship between BTB laws and drug‐related mortality, I use state‐by‐year public‐use data from the NVSS multiple cause‐of‐death by bridged‐race categories files during the period 2005–2020. ⁸ The sample includes 49 states that adopted BTB laws that apply to both private and public employers before the study period (i.e., all the states except Hawaii) and the District of Columbia. ⁹ Since previous studies have shown that BTB laws lead to statistical discrimination against Black and Hispanic men, I disaggregate data by sex and three racial groups: Black non‐Hispanic, White non‐Hispanic, and Hispanic. The main results are presented for the total population by race and sex to decrease zero values in the drug‐related mortality rate. However, Section 4.3.1 also shows the results by age groups.

As defined by the CDC, drug‐related deaths are identified by using the following codes from the International Classification of Diseases, 10th Revision (ICD–10): X40–X44 (unintentional), X60–X64 (suicide), X85 (homicide), and Y10–Y14 (undetermined). Among these deaths, the definition of opioid‐related deaths is based on the multiple cause‐of‐death codes T40.0 (opium), T40.1 (heroin), T40.2 (other opioids), T40.3 (methadone), T40.4 (other synthetic narcotics), and T40.6 (other or unspecified narcotics) while T40.5 (cocaine) and T43.6 (psychostimulants with abuse potential) indicate deaths from stimulants such as cocaine and amphetamines.

For a long time, White non‐Hispanic Americans have experienced a disproportionate increase in drug overdose deaths compared to other racial groups. However, this trend has changed in recent years. Figure 1a shows the drug‐related mortality rate among men by race and ethnicity from 2005 to 2020. Between 2005 and 2015, the largest increase in the drug‐related mortality rate was among White men. However, since 2015, the drug‐related mortality rate has increased by around 70% among White men, while it has more than tripled among Black men. Furthermore, Figure 1a shows that although the drug‐related mortality rate among Hispanic men remained the lowest throughout the period, it increased more rapidly than that of White men between 2015 and 2020. According to Figure 1b, women had consistently lower drug‐related death rates than men, but the pace of increase by race followed a similar pattern.

FIGURE 1.

Drug‐Related Mortality per 100,000. The data come from the National Vital Statistics System. The number on the y‐axis indicates the drug‐related mortality per 100,000 of the corresponding population.

3.1.2. Labor Market Outcomes Data

To investigate the mechanisms through which BTB laws may affect the drug‐related mortality rate, this study utilizes data from the American Community Survey during the period 2005–2020. The ACS is an annual, repeated cross‐sectional survey conducted by the US Census Bureau. The sample for this study is defined as follows: first, it is restricted to individuals aged 15 and above who are not attending college at the time of the survey. Second, given that the survey does not provide specific information on the reasons for being out of the labor force (i.e., retirement, disability, or other factors), the analysis focuses solely on individuals in the labor force. ¹⁰ Third, the main analysis focuses on individuals without a college degree. This is because I aim to examine the statistical discrimination in the labor market for low‐skilled workers. ¹¹ Finally, the analysis includes workers in both the private and the public sectors.

The outcome variables are employment status, full‐time employment, and the natural logarithm of wages. The employment variable is coded as one if an individual is employed and zero otherwise. Full‐time employment is a binary variable, taking a value of one if an individual's usual work hours per week are 35 or more. The analysis of full‐time employment is restricted to individuals who are employed.

Summary statistics for the treated and control states are presented in Supporting Information S1: Table A.1. States that adopted BTB laws were more likely to adopt medical and recreational marijuana laws, had a lower share of the White population, and a larger share of the Hispanic population. However, this would be a concern only if the treated states were different from the control states in terms of unobservable time‐varying characteristics. This would be evident if the treated states were trending differently from the control states prior to the policy adoption.

Alternatively, to estimate the effects of the BTB laws on labor market outcomes, I use the data from the Current Population Survey (CPS). ¹² Even though the ACS has more observations on the state‐race/ethnicity‐year (Burton and Wasser 2024), it has several limitations. First, these data may underestimate the number of employed individuals with irregular or unstructured jobs (U.S. Census Bureau, 2019). Second, the usual hours of work are reported for the 12‐month reference period, which may not align with the census year and may wrongly indicate the presence of pre‐trends in the event studies. In contrast, the CPS employs a more comprehensive battery of questions to determine respondents' employment status. To compare the estimates between the ACS and CPS, I impose the same sample selection as in the ACS data.

3.2. Empirical Strategy

To evaluate the presence of pre‐trends and to estimate the short‐ and long‐term effects of BTB laws on drug‐related mortality, I estimate the event study of the following Equation (1):

$$Y_{\mathit{rgst}}={\beta }_0+\sum _{j\ne -1}{\pi }_{j}1\left(t-BT{B}_s^{\ast }=j\right)+{\beta }_{1}BT{B}_{\text{public}\text{only}}+{\beta }_2{X}_{st}+{\gamma }_s+{\delta }_t+{ϵ}_{\mathit{rgst}}$$ (1)

I utilize two estimation strategies: the Poisson model and the interaction‐weighted estimator proposed by Sun and Abraham (2021). In the Poisson model, the dependent variable $Y_{\mathit{rgst}}$ represents the number of drug‐related deaths categorized by race $(r)$ and sex $(g)$ in state $s$ and year $t$, with the population of the corresponding group serving as the exposure variable. The year when state $s$ adopted BTB laws in both the private and public sectors is denoted by $BT{B}_s^{\ast }$, and $\left(t-BT{B}_s^{\ast }=j\right)$ stands for the relative years before and after the initial policy adoption. The variable $BT{B}_{\text{public}\text{only}}$ is equal to one if BTB laws apply only to the public sector, which implies that coefficients ${\pi }_j$ should be interpreted as the overall effect of private and public BTB laws. The vector $X_{st}$ includes state‐specific control variables such as the share of four age groups in the corresponding population (15–24, 25–34, 35–64, and over 64), the natural logarithms of beer excise taxes, minimum wage, and income per capita in 2020 US dollars, the natural logarithm of police officers per 10,000 people, and a set of dummy variables for medical and recreational marijuana laws, the availability of corresponding dispensaries, operational prescription drug monitoring programs, Naloxone and pill‐mill laws, and immigration policies such as E‐Verify mandates in the public and private sectors. The detailed explanation of the control variables is shown in Supporting Information S1: Table A.2. Finally, ${\gamma }_s$ is a set of state‐fixed effects that account for state time‐invariant characteristics, and ${\delta }_t$ control for time‐specific shocks.

The Poisson model is particularly suitable in the setting when the dependent variable contains zeros since it can handle zero values and accounts for the discrete nature of the data. The coefficients in the Poisson model can be interpreted as a log point change in the number of deaths of a particular group over the corresponding population.

Given that two‐way fixed effects estimates might be biased if treatment effects are heterogeneous over time and across groups (Goodman‐Bacon 2021; Sun and Abraham 2021), I also employ the interaction‐weighted estimator proposed by Sun and Abraham (2021). ¹³ For this purpose, I estimate a log‐linear model where the dependent variable is the natural logarithm of the drug‐related death rate per 100,000. In this specification, I am weighting the regressions by the corresponding population. ¹⁴

Next, I examine the impact of BTB laws on statistical discrimination in the labor market with the ACS data. To investigate the effect of BTB laws on labor market outcomes, I estimate Equation (2):

$$Y_{\mathit{irgst}}={\alpha }_0+{\alpha }_{1}BT{B}_{st}+{\alpha }_{2}BT{B}_{\text{public}\text{only}}+{\alpha }_3{Z}_{st}+{\alpha }_4{W}_{\mathit{irgst}}+{\gamma }_s+{\delta }_t+{ϵ}_{\mathit{irgst}},$$ (2)

where $Y_{\mathit{irgst}}$ denotes the labor market outcome of an individual $i$ of race $r$ and sex $g$ in state $s$ and year $t$. ¹⁵ $BT{B}_{st}$ is a binary variable that is coded as one if the BTB law is in effect in both sectors. Similarly to Equation (1), $BT{B}_{\text{public}\text{only}}$ denotes BTB laws that apply only to public employers. The vector $Z_{st}$ is a subset of state‐level controls $X_{st}$ that includes the natural logarithms of income per capita and the minimum wage in 2020$, and e‐verify laws in the private and public sectors. The vector $W_{\mathit{irgst}}$ includes individual‐level control variables such as binary variables for marital status, parental status, high school completion, being in a certain age group (25–34, 35–64, and over 64), and citizenship status. ¹⁶ Moreover, I estimate the relationship between BTB laws and labor market outcomes by using an estimator suggested by Borusyak et al. (2024) (hereafter, BJS) and estimate the event study similar to the one used in Equation (1).

4. Results

4.1. BTB Laws and Drug‐Related Mortality

Figures 2 and 3 show the estimated effect of BTB laws on drug‐related mortality obtained from a Poisson regression and the interaction‐weighted estimator proposed by Sun and Abraham (2021). I focus on the estimates obtained with the interaction‐weighted estimator. The standard errors in all the event studies are clustered at the state level.

FIGURE 2.

BTB and Drug‐Related Mortality among Men: SA and Poisson Estimates. Based on data from the National Vital Statistics System. The event‐study estimates are obtained using a two‐way fixed effects Poisson regression and the interaction‐weighted estimator proposed by Sun and Abraham (SA). The dependent variable in the Poisson model is the number of drug‐related deaths, with the corresponding population specified as an exposure variable. The dependent variable in the SA model is the natural logarithm of the drug‐related death rate. The SA regressions are weighted by the corresponding population. See Section 3.2 for a list of controls. The 95% confidence intervals are obtained with standard errors clustered at the state level.

FIGURE 3.

BTB and Drug‐Related Mortality among Women: SA and Poisson Estimates. Based on data from the National Vital Statistics System. The event‐study estimates are obtained using a two‐way fixed effects Poisson regression and the interaction‐weighted estimator proposed by Sun and Abraham (SA). The dependent variable in the Poisson model is the number of drug‐related deaths, with the corresponding population specified as an exposure variable. The dependent variable in the SA model is the natural logarithm of the drug‐related death rate. The SA regressions are weighted by the corresponding population. See Section 3.2 for a list of controls. The 95% confidence intervals are obtained with standard errors clustered at the state level.

I find evidence that BTB laws are positively associated with the drug‐related mortality of racial minorities. I first focus on the estimates for Black men that are presented in Figure 2. The coefficients from both the Poisson model and the interaction‐weighted estimator provide limited evidence of a pre‐trend but suggest a gradual increase in the drug‐related death rate during the initial 2 years after policy adoption. After two years, BTB adoption is associated with a 37% increase $\left(e^{0.316}-1\right)$ in the drug‐related mortality rate. Starting from the third year after the enactment of BTB laws, the association between BTB laws and the drug‐related death rate weakens in both magnitude and precision. One possible explanation for this pattern of results might be a change in the susceptible population. The initial increase in drug‐related fatalities during the early stages of policy adoption might have contributed to a reduction in the number of vulnerable individuals in subsequent years.

Similarly, Figure 2b shows that there is a positive association between BTB laws and drug‐related mortality among Hispanic men. The estimated effect of the policy on the drug‐related mortality rate gradually increases during the first 2 years after policy adoption and remains constant in subsequent years.

Next, Figure 2c presents the estimates for White men. There is a slight increase in the drug‐related death rate for White men following the adoption of BTB laws. However, the association between BTB laws and drug‐related mortality is weaker in both magnitude and precision compared to their Black and Hispanic counterparts. These results are consistent with the findings of Doleac and Hansen (2020) and Sabia et al. (2021), as well as the results in Section 4.2, all of which suggest that White men appear to be less affected by BTB laws compared to Black and Hispanic men.

The estimates for women are presented in Figure 3. Figure 3a shows that following the enactment of BTB laws, the drug‐related mortality rate among Black women begins to rise. Two years after policy adoption, BTB laws are associated with a 43% increase in their drug‐related mortality rate. Four years after the policy adoption, the estimated effect of BTB laws on the drug‐related mortality rate begins to weaken, which is similar to the pattern of results for Black men.

The estimates for Hispanic women (Figure 3b) are ambiguous. Poisson estimates for Hispanic women suggest a change in the trend in the drug‐related mortality rate following the policy adoption. However, the estimates obtained with the interaction‐weighted estimator provide little evidence of a change in the drug‐related mortality rate following the adoption of BTB laws. Finally, BTB laws appear to have no effect on the drug‐related mortality rate of White women since the estimates in Figure 3c are small and not statistically significant.

4.2. Mechanisms

Agan and Starr (2018) and Doleac and Hansen (2020) find that BTB laws increase statistical discrimination against racial and ethnic minorities without a college degree. In Table 2, I show the static two‐way fixed‐effect ordinary least squares (TWFE OLS) estimates of the effect of BTB laws on the labor market outcomes for low‐skilled individuals by using the ACS data. The corresponding BJS estimates are presented in Supporting Information S1: Appendix Table A.3, and the event studies are presented in Supporting Information S1: Appendix Figures A.1 and A.2. ¹⁷

TABLE 2.

BTB and employment outcomes among individuals without a college degree.

	Men			Women
	White	Black	Hispanic	White	Black	Hispanic
Panel A
Employment	−0.0042	−0.0123**	−0.0073*	−0.0052*	−0.0089**	−0.0099**
	(0.0046)	(0.0047)	(0.0041)	(0.0031)	(0.0041)	(0.0046)
Pre‐BTB Mean	0.929	0.847	0.921	0.936	0.870	0.898
$N$	4,573,531	633,740	1,228,691	3,551,655	654,261	852,815
Panel B
Full‐time Hours	−0.0046*	−0.0171***	−0.0116***	0.0014	−0.0135*	−0.0125**
	(0.0024)	(0.0039)	(0.0039)	(0.0040)	(0.0070)	(0.0056)
Pre‐BTB Mean	0.873	0.834	0.882	0.703	0.759	0.727
$N$	4,252,550	537,406	1,133,514	3,325,614	569,482	768,335
Panel C
Ln(Wage)	−0.0175*	−0.0567**	−0.0208	−0.0139	−0.0375***	−0.0079
	(0.0089)	(0.0274)	(0.0135)	(0.0087)	(0.0116)	(0.0129)
Pre‐BTB Mean	49716.82	34549.81	35947.81	31561.15	28724.83	25879.47
$N$	4,019,617	587,891	1,101,690	3,246,548	591,797	766,667

Note: Based on data from the American Community Survey. Panel A includes low‐skilled individuals in the labor force. Panels B and C include employed low‐skilled individuals. The estimates are obtained with a two‐way fixed effects OLS model. The dependent variables are binary variables for employment and full‐time employment and the natural logarithm of wages. See Section 3.2 for a list of controls. The table reports the pre‐BTB mean of wages instead of the logarithm of wages.

^*statistically significant at the 10% level.

^**at the 5% level.

^***at the 1% level.

My findings confirm the results from previous studies. The estimates suggest that the relationship between BTB laws and labor market outcomes is stronger among Black and Hispanic individuals without a college degree than among their White counterparts. Both dynamic and static estimates provide evidence that BTB laws are negatively associated with an intensive margin of labor supply among racial minorities. The TWFE estimates in Table 2 suggest that following the policy adoption, the probabilities of working full‐time for Black and Hispanic individuals decrease by 1.2–1.7% points (1.2%–2% with respect to the mean) with the largest decrease among Black men. ¹⁸ The static OLS and BJS estimates also suggest that BTB laws are negatively associated with employment among racial/ethnic minorities, although the employment trend in event studies appears less pronounced. The relationship between BTB laws and wages appears to be the strongest among Black men. The estimates suggest that after the passage of BTB laws, wages of Black men declined by 5.5% $\left(1-e^{(-0.0567)}\right)$. Similarly, after BTB, the wages of Black women declined more than the wages of their Hispanic and White counterparts.

Supporting Information S1: Appendix Table A.4 shows the estimates obtained with Equation (2) for individuals with a college degree. Consistent with studies by Doleac and Hansen (2020) and Sabia et al. (2021), the estimated employment effects are much less evident among individuals with a college degree. Only Hispanic men with a college degree experienced a reduction in the probability of full‐time employment after the enactment of BTB laws. However, this estimated effect is offset by White women with a college degree, who experienced an increase in the probability of working full‐time after BTB laws were passed. These findings suggest that BTB laws appear to disproportionately impact those who are already more susceptible to substance abuse, as individuals without a college degree are at higher risk of dying from the diseases of despair (Case and Deaton 2017).

Next, I examine whether the hypothesis of statistical discrimination is substantiated independently of the dataset used for the analysis. The findings obtained with the CPS data are presented in Supporting Information S1: Appendix Tables A.5 and A.6 and Appendix Figures A.5 and A.6. ¹⁹ The pattern of the results is similar to that obtained with the ACS data. However, estimates obtained with the CPS data suggest that BTB laws are associated with a larger decrease in employment and wages among Black workers, while for Hispanic men, BTB laws are mostly associated with a decrease in full‐time employment.

Overall, these findings provide suggestive evidence that the worsening of labor market conditions among low‐skilled racial minority groups is one of the channels through which BTB laws may affect drug‐related mortality. Those who lack higher education already face a higher risk of substance abuse. Therefore, laws that diminish their labor market opportunities appear to exacerbate this problem.

Considering the impact of BTB laws on labor market outcomes, their effect sizes on the drug‐related mortality rate may seem large. Hence, it is important to compare the findings from this study with those from previous studies that investigated the relationship between labor market outcomes and drug‐related mortality.

The findings of this study suggest that the probability of employment among Black and Hispanic men decreased by approximately 1% point after the passage of BTB laws. The drug‐related mortality rate of this demographic group increased by approximately 25% in the first year after the passage of BTB laws and by approximately 40% in the second year after the passage of BTB laws (i.e., 2.5 to 4 deaths per 100,000 and 5.5 to 6 deaths per 100,000 relative to the pre‐treatment means shown in Supporting Information S1: Table A.7). Pierce and Schott (2020) document that exposure to a bill granting permanent normal trade relations to China is associated with a one‐percentage‐point increase in the unemployment rate and an increase in the death rate from drug overdoses of 2–3 deaths per 100,000. Thus, given that the current drug epidemic is driven by more lethal substances such as fentanyl, the estimates of the effects of BTB laws on drug‐related mortality are broadly consistent with the findings of Pierce and Schott (2020). ²⁰

Another way to compare the magnitude of the effects of BTB laws on drug‐related mortality with their labor market impacts is as follows. There are approximately 14 million Black workers without a college degree in the US. ²¹ Suppose 1% of these workers lose their jobs due to BTB laws, which would correspond to 140,000 individuals. BTB laws are associated with about a 20% increase in drug‐related deaths among Black men and women. Given that the average number of deaths among working‐age Black men and women between 2005 and 2020 was roughly 5000, this corresponds to an increase of approximately 1000 deaths. This implies that if a decrease in employment is the sole channel through which BTB affects drug‐related deaths, approximately 0.7% of those who lose jobs overdose. However, I find that BTB laws are also associated with a decrease in full‐time employment and wages. Hence, an increase in drug‐related mortality attributed to a decrease in employment is likely to be lower than discussed above.

Moreover, there are also other channels through which BTB laws may affect drug‐related mortality. Since BTB laws delay criminal background checks, they may decrease the future costs of drug possession. In 2020, approximately 800,000 arrests were made for drug possession, with the majority of penalties resulting in fines or probation. Compared to these penalties, the fear of being unemployed due to a criminal record might be a significant deterrent. Thus, BTB laws may lower the perceived costs of drug‐related crimes. Moreover, Sabia et al. (2021) show that BTB laws increase crime among Hispanic low‐skilled men by 16%. Even though their study focuses on property crime, in the same manner, BTB laws may increase incidences of sale and manufacture of drugs, thus increasing the supply of drugs.

4.3. Heterogeneity and Robustness Checks

4.3.1. Heterogeneity

So far, I have examined the effect of BTB laws that apply to both private and public employers. Supporting Information S1: Appendix Figures A.5 to A.8 and Appendix Tables A.8 to A.10 show differential effects by law types on drug‐related mortality and labor market outcomes. The estimates in Supporting Information S1: Table A.8 suggest that the effects of private BTB laws on drug‐related mortality are larger in magnitude compared to the effects of public BTB laws. However, the statistical test fails to reject the null hypothesis that the effect of BTB laws on drug‐related mortality does not differ by type of law. Similarly, I fail to reject the null hypothesis that private and public BTB laws do not have differential effects on labor market outcomes. ²²

Next, I investigate if the effects of BTB laws differ across age cohorts. Supporting Information S1: Figure A.9 presents heterogeneity by two age groups: 15–34 and 35–64. The estimates for Black individuals and Hispanic men suggest that both age groups experienced an increase in the drug‐related mortality rate after BTB laws were enacted. The estimates in Supporting Information S1: Appendix Tables A.12 and A.13 also indicate that the effects of BTB laws on labor market outcomes are evident among both age groups. After the passage of BTB laws, the younger age group experienced a higher incidence of job losses, while the older age group had a lower probability of working full‐time. ²³

Furthermore, I investigate the heterogeneity by type of drugs: opioids versus. stimulants. According to Supporting Information S1: Appendix Figure A.10, BTB laws are positively associated with both opioid‐related mortality and mortality from stimulants. This pattern of results suggests that the effect of BTB laws is not limited to a specific type of substance, which provides suggestive evidence that the effects are not driven by the fentanyl crisis.

4.3.2. Robustness Checks

Factors related to the fentanyl crisis. To provide more explicit evidence that the results are not driven by differential exposure to the fentanyl crisis, I perform a series of robustness checks. Powell and Pacula (2021) show that exposure to OxyContin reformulation is positively associated with synthetic opioid deaths. I control for the exposure to OxyContin reformulation by interacting state‐level OxyContin and pain‐reliever misuse rates in the pre‐reformulation period (2004–2009) with time‐fixed effects. The results are shown in Supporting Information S1: Appendix Figures A.11 and A.12. The estimates suggest a similar pattern of findings, indicating no specific pre‐trends, but an increase in drug‐related mortality among Black individuals and Hispanic men following the adoption of BTB laws.

Next, following Moore et al. (2023), I control for legal trade facilities through which fentanyl might be smuggled. To do so, I interact the logarithm of state‐level import per resident in 2008 with time‐fixed effects. ²⁴ The results presented in Supporting Information S1: Appendix Figures A.13 and A.14 are consistent with the main results. I also assess whether the legal diversion of fentanyl and the legal status of fentanyl testing strips affect my findings. To account for these factors, I add the logarithm of the legal distribution of fentanyl per resident and an indicator variable that takes the value one if possession of fentanyl testing strips is not criminalized to Equation (1). The results are presented in Supporting Information S1: Appendix Figures A.15 and A.16. Controlling for these confounding factors has little impact on my findings.

Some states that adopted BTB laws are located on the border with Mexico and Canada. To provide evidence that the effect of BTB laws is not contaminated by drug smuggling across the Canadian and Mexican borders, I follow Moore et al. (2023), and exclude 14 states that have a land border with Mexico and Canada from my sample. The findings presented in Supporting Information S1: Appendix Figures A.17 and A.18 show that the estimates are robust to the exclusion of border states.

Sensitivity to other control variables. To examine whether the results are sensitive to the choice of other control variables, I make several changes to the set of controls. First, I omit all the control variables. Second, to account for baseline differences between treated and control states, which may be problematic for the causal interpretation of the estimates (Kahn‐Lang and Lang 2020), I add the interactions of baseline levels of continuous control variables with the year fixed effects. ²⁵ Third, I add a separate control variable “local only BTB” to the Equation (1). Specifically, this variable is coded as zero if no local laws are in place or if a state‐level law has been adopted, and it is equal to the share of the population of the state covered by a city or county BTB law otherwise. ²⁶ The share is constructed as the population of a city or county in 2003 divided by the state population in 2003. ²⁷ Supporting Information S1: Appendix Figures A.19 to A.24 show that the estimates are not sensitive to the set of control variables. The estimated effects are also robust to the inclusion of state‐specific trends, which control for unobservables that have a constant effect in states over time. The estimates obtained from the regressions that account for state‐specific trends are shown in Supporting Information S1: Appendix Figures A.25 and A.26. Furthermore, Supporting Information S1: Appendix Figures A.27 and A.28 show that estimates are robust to estimating the effects with the estimator proposed by Borusyak et al. (2024), which in some cases can handle the control variables better than the estimator proposed by Sun and Abraham (2021).

Sensitivity analysis on the validity of the pre‐trends. The identification of the effect of BTB laws on drug‐related mortality relies on the assumption of parallel trends between treatment and control states. In an attempt to account for baseline differences between treated and control states, I have already interacted the baseline continuous variables with the time fixed effects. However, examining the sensitivity of adding these control variables to the Equation (1) provides only suggestive evidence that DiD methods can be used to identify the causal effect of BTB laws (Kahn‐Lang and Lang 2020). To assess the sensitivity of the estimates to potential violations of the parallel trends assumption, I employ the methodology developed by Rambachan and Roth (2023). Instead of imposing the parallel trends assumption, I estimate the maximum deviation from a linear trend in the post‐treatment period relative to the maximum observed pre‐treatment deviation that would make the estimated results inconsistent. I employ this sensitivity analysis for the coefficients $t+2$ and $t+3$ when the effect of BTB laws on drug‐related mortality reaches its maximum. The findings are provided in Supporting Information S1: Appendix Figures A.29 and A.30. The results suggest that when the effect of BTB laws on drug‐related mortality reaches its maximum, the coefficient for Black men would remain significant up to $\bar{M}=0.6$ while the coefficient for Hispanic men would remain significant up to $\bar{M}=1.4$. In other words, for the estimated results to be inconsistent, a violation in parallel trends in the post‐period would need to be at least 60% and 140% of the size of the maximum deviation in the pre‐period for Black and Hispanic men, respectively. For Hispanic and Black women, the breaking values are 0.6 and 1, respectively.

Logarithm of zeros and functional form of the dependent variable. I also investigate if Sun and Abraham (2021) estimates are robust to estimating a log‐linear model without missing values. To address the issue with missing values resulting from taking the logarithm of zero, I add a constant to the dependent variable. This constant is set to half of the minimum positive value of the drug‐related death rate within the corresponding group. By doing so, I ensure that all observations have non‐zero values, enabling the logarithmic transformation without missing data. Supporting Information S1: Figures A.31 and A.32 in the Appendix do not provide any evidence that estimates are of different magnitude or statistical significance.

To examine whether the pattern of the results is robust to the functional form of the dependent variable, I estimate the effect of BTB laws on the levels instead of the logarithmic transformation of the dependent variable. The estimates depicted in Supporting Information S1: Appendix Figures A.33 and A.34 show that the effects are larger in magnitude for Black and Hispanic men, but the pattern of the results across demographic groups remains the same. The larger magnitude of the results arises due to an exponential increase in the drug‐related mortality rate, and the size of the effects provides evidence that the logarithmic transformation is justified.

Dates of BTB laws. For the main analysis, I have used the effective dates of BTB laws to identify the effect of the policy on drug‐related mortality and labor market outcomes. Supporting Information S1: Appendix Figures A.35 and A.36 present the estimates from the specification that replaces the effective dates of the laws with the dates when the laws were signed. In this specification, it takes 1 year longer for drug‐related mortality to increase following the adoption of BTB laws. However, the pattern of the results is similar to the one presented before. Moreover, since Virginia's BTB law only prohibits employers from obtaining information about arrests for simple marijuana possession, in an alternative specification, I code Virginia as a non‐BTB state. The estimates in Supporting Information S1: Appendix Figures A.37 and A.38 show that the findings are robust to this modification.

Sample selection for labor market outcomes. Given that the ACS does not provide specific information on the reasons for being out of the labor force (i.e., retirement, disability, or other factors), the main analysis focuses solely on individuals in the labor force. However, if BTB laws encourage or discourage some workers to stay in the labor force, this may lead to an endogenous sample selection. To examine if this is the case, I analyze the effect of BTB for all workers. ²⁸ The findings are shown in Supporting Information S1: Appendix Table A.14. When the analysis sample includes individuals out of the workforce, ACS data estimates decrease in both magnitude and precision, while CPS data estimates appear to remain largely unaffected. This discrepancy may arise because the ACS tends to underestimate the number of employed individuals with irregular or unstructured jobs (U.S. Census Bureau, 2019).

Similarly, the sample of workers for the outcomes that are conditional on employment may be endogenous as BTB laws affect employment. To examine this concern, I compare the original estimates with the estimates obtained with the regressions that exclude individual‐level controls. The findings are reported in Supporting Information S1: Tables A.15 and A.16. The coefficients obtained in the model without individual‐level controls are similar to those obtained in the regressions that include individual‐level controls. This provides suggestive evidence that the sample of workers after BTB laws is similar to the sample of workers before BTB laws in terms of observables. The samples may still be different in terms of unobservables, which is why the findings for the outcomes conditional on employment have to be viewed with caution.

5. Conclusion

Ban the Box laws are gaining popularity across the United States. Several studies have examined the impact of BTB laws on labor market outcomes (Craigie 2020; Doleac and Hansen 2020; Kaestner and Wang 2024). This study provides evidence of the negative, indirect consequences of BTB laws on the leading cause of despair‐related deaths – that is, drug‐related mortality.

Using state‐level mortality data, I find that the adoption of BTB laws that apply to both private and public employers is associated with an increase in drug‐related fatalities among Black individuals and Hispanic men. One year after the adoption of BTB laws, the drug‐related mortality rate among Black and Hispanic men increased by 22–27%; two years after the adoption of BTB laws, the drug‐related mortality rate among Black and Hispanic men increased by more than 35%.

The main mechanism driving this relationship appears to be diminished labor opportunities resulting from statistical discrimination against low‐skilled Black and Hispanic individuals. Consistent with the results suggested by Doleac and Hansen (2020) and Sabia et al. (2021), I show that BTB laws are associated with a reduction in wages, the probability of employment, and the probability of full‐time employment among low‐skilled minority groups.

These results suggest a negative relationship between labor opportunities and drug‐related mortality, confirming the findings of Autor et al. (2019); Pierce and Schott (2020) and Charles et al. (2019). Consequently, addressing demand factors in substance abuse might be crucial to combat the drug epidemic effectively.

Conflicts of Interest

The author declares no conflicts of interest.

Supporting information

Supporting Information S1

Cheipesh, Oleksandra . 2025. “Hidden Costs of Ban the Box Laws: Unraveling the Effects on Drug‐Related Deaths.” Health Economics: 2059–2071. 10.1002/hec.70018.

Data Availability Statement

The data that support the findings of this study are openly available in CDC WONDER at https://wonder.cdc.gov/.