Naloxone Access Laws and Unintended E�ects on the

Opioid Crisis: Increase in Overdose Deaths and Drug

Abuse

Estrella Ndrianasy

February 10, 2019

Abstract

Naloxone Access Laws (NALs) are one of several policies intended to �ght the rampant

opioid epidemic by allowing the public to carry naloxone (Narcan), an overdose reversal

drug. All states have currently adopted NALs with New Mexico being the �rst one in

2001. This study extends the literature by examining the potential for moral hazard from

the ready availability of naloxone to opioid abusers and the general public. E�ects on

opioid overdose deaths, opioid rehabilitation admissions, and the legal supply of opioids are

analyzed via a �xed e�ects regression method and propensity score matching techniques.

The estimation framework accounts for potential biases in a state's decision to adopt NALs

and concurrent laws aimed at combating the crisis such as the Good Samaritan Laws and

Prescription Drug Monitoring Programs. Each year, opioid overdose deaths signi�cantly

increased by 130 to almost 190 and the supply of legal opioid drugs signi�cantly went

up from 3,200 to 4,000 more grams per 100,000 population due to NALs during the time

period analyzed. On the other hand, the e�ects of NALs on opioid rehabilitation admission

is inconclusive. The results, in particular the rise in opioid overdose deaths, seem to indicate

a moral hazard issue with more addicts misusing dangerous levels of opioids with naloxone

as a safety net.

1

1 Introduction

The opioid epidemic is prevalent in the U.S. claiming over 60,000 lives in 2016 [Wonder, 2017].

The opioid crisis is fueled by misuse and abuse of prescription pain relievers. Oftentimes, the

addiction starts with legitimate needs for pain relievers whose euphoric e�ects can quickly turn

into dependency. Once addicted, the abuser may seek more opioids through �doctor-shopping�

wherein they attempt to obtain more of the drugs through multiple prescribers. Addicts also

acquire opioids through friends and family's prescriptions. At the other extreme, opioids are

available illegally where they present an even greater risk to the individual when laced with

the cheaper but powerful fentanyl, a synthetic opioid. Opioid overdoses are now the number

one drug killer in the US surpassing others such as cocaine and methamphetamine. Various

measures such as Prescription Drug Monitoring Programs, insurance and pharmacy bene�t

manager strategies, state legislation, clinical guidelines, safe storage and disposal, and naloxone

distribution were put in place to prevent doctors from over-prescribing opioids or patients from

abusing opioids [Haegerich et al., 2014]. Most of the measures were supply-side interventions

with the intent of cutting o� an addict's access to the substance, resulting in the substitution to

other less restrictive but oftentimes more dangerous alternatives like heroin [Alpert et al., 2017].

Naloxone distribution measures are oriented towards existing drug consumers, their friends and

families, and service providers by educating them on the overdose risk factors, signs of overdose,

appropriate response, and the administration of naloxone.

The administration of Naloxone Hydrochloride (naloxone) reverses an opioid overdose and

can prevent death. Naloxone is an opioid receptor antagonist that can be administered via

intramuscular, intravenous, and intranasal routes. It works by displacing opioid agonists, such

as heroin or oxycodone, from opioid receptors [Doe-Simkins et al., 2009]. Naloxone is a relatively

inexpensive medication with a price tag of $20 to $40. Naloxone is can also be obtained at no

cost from overdose prevention programs in 2016. It works within mere minutes, with prior

allergic reaction as the only contraindication [Illinois, 2017]. Most importantly, naloxone has no

e�ect in the absence of opioid consumption. Without any agonist properties, it has no potential

for abuse and minimal likelihood for diversion or misuse [Heller and Stancli�, 2007]. In the wake

of the opioid crisis, the immediate availability of naloxone is crucial. Additionally, naloxone is a

2

prescription drug, but not a controlled substance. This implies that the prescription of naloxone

is legal whereas the dispensing of the drug by medical professional at the point of service is

subject to varying rules for each state. Naloxone Access Laws (NALs) are policies passed by

states to curb the opioid overdose epidemic. As of July 2017, all �fty states and the District

of Columbia have a form of NALs to prevent unintentional drug overdose deaths. Generally,

NALs removes some criminal, civil, and professional liabilities for prescribers, dispensers, and

laypersons administering the opioid antagonist.

This paper explores the potential e�ect for moral hazard. Even when accounting for timing

di�erences in passing NALs, the number of prescription opioids and heroin overdose deaths have

increased alarmingly in the past decade and a half. Drug user's perception of opioid consumption

posing less of a risk because of the availability of naloxone is the focus of this paper. Plausible

unintended e�ects of NALs are increased consumption of the opioid resulting in drug rehabili-

tation treatments and accidental overdose deaths. However, this could also be due to a success

of NALs leading to addicts seeking help in overcoming their addiction. It will be assumed that

no matter what drives the decision to seek rehabilitation, it must begin with some prior opioid

consumption turning to addiction on the part of addict. Moreover, there are some limitations

in the reliability of mortality data due to under-reporting and/or human error in accounting

for the true causes of deaths as documented in the literature, especially in the causes of drug

overdoses [Ruhm, 2018]. The analysis will rely on the International Classi�cation of Diseases-10

(ICD-10) to best account for the potential mortality cause discrepancies. Overall, this paper ex-

tends the literature by looking at the unintended e�ects of NALs, a public health e�ort intended

to reduce opioid overdose deaths. Most of the literature hails NALs as successes whereas the

evidence suggest an ever increasing trend in opioid deaths, particularly heroin. This study aims

to answer the mechanism behind this phenomenon while allowing for a national level as com-

pared to previous community and city level assessments for smaller pilot prevention programs

from the early and mid 2000s. The paper will be divided into the following sections: literature

review, identi�cation strategy, data description, results analysis, and discussion. Overall, the

results show the presence of moral hazard for opioid overdose deaths nationally. Further, the

total grams for legal opioid drugs per capita is signi�cantly increased due to the NALs, whereas

there was no e�ect on opioid rehabilitation treatment.

3

2 Literature Review

NALs are widely depicted as successful policies in the �ght against the opioid epidemic in the

literature. Various pilot programs were implemented in large metropolitan areas in the early

2000s resulting in a reduction of unintended opioid overdose deaths by as much as nine per

cent in New York City [Heller and Stancli�, 2007, Worthington et al., 2006] to a high of 20%

in Chicago in 2001 [Maxwell et al., 2006]. Other programs such as the Project Lazarus in

Wilkes County, North Carolina, led a signi�cant drop of opioid overdose mortality from a high

of 46.6 to 29 per 100,000 inhabitants in just one year in 2010 [Albert et al., 2011]. Naloxone

distribution programs in other cities such as San Francisco DOPE program in 2010 led to a

greater awareness of the bene�ts provided by NALs [Straus et al., 2013]. The initiative also

led to a take-back initiative by the Drug Enforcement Administration in 2012, where 520 tons

of unwanted or expired medication were disposed o� by the public. The city of Baltimore,

Maryland, enacted a similar prevention program called Staying Alive resulting in 22 successful

overdose reversals performed by 19 out of 43 individuals recruited through street-based outreach

and advertising [Tobin et al., 2009]. Chicago's own program also led to a 20% decrease in heroin

overdoses in 2001, followed by a 10% decrease for 2002 and 2003, respectively through the

distribution of 3,500 naloxone vials resulting in over 300 overdose reversals [Maxwell et al.,

2006]. Overall, naloxone overdose prevention programs have overwhelmingly reported positive

outcomes of successful opioid overdose reversals via peer administration of naloxone [Dettmer

et al., 2001, Worthington et al., 2006, Maxwell et al., 2006, Sporer and Kral, 2007, Mueller et al.,

2015].

NALs have been successful in reducing opioid overdose unintentional deaths. There were

more than 64,000 drug overdose deaths estimated in 2016, with the largest increase stemming

from synthetic opioids such as fentanyl and fentanyl analogs with over 20,000 deaths [Wonder,

2017]. Heroin and natural and semi-synthetic opioids were second and third in number of drug

deaths with 15,446 and 14,427 respectively. Another opiate, methadone, was the sixth highest

killer with over 3,000 deaths, preceded by cocaine and methamphetamine. In the early 2000s,

several studies have documented the e�ects of NALs and GSLs on reducing the number of

opioid deaths in small pilot programs in large cities in the U.S and the world [Maxwell et al.,

4

2006, Heller and Stancli�, 2007, Kerr et al., 2008, Piper et al., 2008, Tobin et al., 2009, Doe-

Simkins et al., 2009, Albert et al., 2011, Wheeler et al., 2012]. The success of NALs is closely

tied to a bystander's willingness to intervene and adequacy in the administration of naloxone

in the event of an overdose. Overall, participants in NALs training program programs have a

positive attitude towards peer administration as they feel empowered to help others by reducing

mortality and morbidity in minimizing delays to treatment, feel as they contribute in preserving

ambulance services for other medical emergencies, and most of all avoid authority involvement

[Kerr et al., 2008]. Nonetheless, some literature has suggested the potential for moral hazard

from the NALs due to the reduced expected cost of opioid consumption. The unintended e�ects

range from an increased consumption of non-prescription opioids to the transition to harder

drugs such as heroin, cocaine, and methamphetamine [Bachhuber et al., 2014, Seal et al., 2003,

2005]. Others look at competing drug laws, such as Medical Marijuana Laws, and �nd mixed

results in subsequent opioid distribution [Rees et al., 2017, Powell et al., 2015, DiNardo and

Lemieux, 2001]. Moreover, some suggest prescription opioids are potential gateway drugs where

misuse is considered a key feature of trajectories into injection drug use such as heroin [Lankenau

et al., 2012, Fiellin et al., 2013]. These are attributed to opioid overdose victims continuing or

even increasing drug abuse in order to alleviate withdrawal symptoms [Seal et al., 2003, Lagu

et al., 2006].

Naloxone then becomes a safety mechanism enabling at risk individuals to consume greater

amounts of opioids in the absence of consequences [Albert et al., 2011]. Moral hazard may result

in increased opioid addiction and overdose deaths, although the literature is fairly inconclusive.

In a survey of injection drug users in San Francisco, 35% felt comfortable using greater amounts

of heroin in the presence of naloxone while 46% of overdose victims were planning on using

heroin again to alleviate withdrawal symptoms [Seal et al., 2003]. The same authors also found

reduced heroin consumption later after adequate education and training in overdose reversal,

illustrating the mixed nature of the literature [Seal et al., 2005]. Another naloxone distribu-

tion program in Providence, Rhode Island, found rare adverse e�ects whereby the likelihood of

getting addicted on heroin actually increased in opioid addicts when symptoms of withdrawals

are precipitated, driving them to seek relief in additional substance abuse [Lagu et al., 2006].

Others highlight the potential for naloxone to increase polysubstance abuse, especially the as-

5

cension to harder opiates such as heroin using prescription opioids as gateway drugs [Bachhuber

et al., 2014, Sporer and Kral, 2007, Kerr et al., 2008]. Additionally, there has been a signi�-

cant increase in the amount of substances distributed in the U.S., including methadone, heroin,

benzodiazepines and barbiturates, cocaine, other opioids, and their related overdoses [Walley

et al., 2013, Rees et al., 2017]. Finally, there is some evidence that competing laws such as

Medical Marijuana Laws, could have a reducing e�ect on prescription opioids misuse through

a substitution mechanism [Powell et al., 2015, DiNardo and Lemieux, 2001]. Another example

of a measure to combat the prescription opioid overdose deaths but yielded much higher heroin

fatalities is the OxyContin Reformulation of 2010, which marked the introduction of an abuse

deterrent opioid. The reformulation caused an opioid supply disruption that turned drug addicts

to much harder substances such as heroin and fentanyl, illicit and dangerous forms of opiates.

The measure marginally decreased prescription painkiller fatalities but signi�cantly increased

heroin and synthetic opioid (fentanyl) overdose deaths [Alpert et al., 2017]. Some law enforce-

ment entities also have at best a diverging and at worst a skeptical opinion on the e�ectiveness

of NALs, primarily focusing on their perceived enabling potential, especially with regards to

immunity from prosecution [Gaston et al., 2009, Banta-Green et al., 2013].

This paper extends the scarce and mixed literature of the unintended e�ects of NALs. From

an economics perspective, the potential issue of moral hazard is important to evaluate the

practical costs and bene�ts associated in the implementation of NALs. The moral hazard itself

seems to originate from the sense of safety provided by the presence of naloxone, which may cause

an opioid addict to continue misusing at his own and society's cost. This analysis follows the

literature by looking at the commonly used substance abuse outcomes: treatment/rehabilitation,

unintended overdose deaths, and the amount of opioids distributed for all states. The Treatment

Episode Data Set (TEDS), the National Vital Statistics System (NVSS) Multiple Cause of Death

Mortality deaths, and Automation of Reports and Consolidated Orders System (ARCOS) will

be used to conduct the analysis at the state level. The TEDS and NVSS data are maintained

by the CDC by the Substance Abuse and Mental Health Services Administration and National

Center for Health Statistics respectively. TEDS reports treatment information on addicts seeking

rehabilitation. NVSS records mortality data and their causes among other things. Finally, the

ARCOS data is an automated, comprehensive drug reporting system monitoring the �ow of DEA

6

controlled substances from their point of manufacture through commercial distribution channels

to point of sale. This paper updates the literature by providing a national level analysis on the

e�ects of NALs since most states passed the measure into law in the early 2010s. Previous work

focused on small pilot programs limited with small sample size and weaker power for external

validity. Moreover, an extension of the di�erence-in-di�erence identi�cation strategy, a �xed

e�ects regression model, as well as a propensity score matching framework will be conducted to

analyze the impact of NALs on the opioid overdose epidemic.

3 Identi�cation Strategy

3.1 Fixed E�ects Model

A standard di�erence-in-di�erence (DID) setup is inadequate for instances where there are

multiple time periods and multiple groups. An expansion of the standard DID is possible by

including time and individual or group �xed e�ects. Moreover, the variation in treatment timing

is exploited to get a glimpse of the e�ects of the policy implementation. The expanded regression

model is as given below:

Yst = yeart + states + βNaloxonest + θXs + εst (1)

where Drugst is the outcome of interest referring to categories of drugs such as opioids and

heroin in state s and year t, yeart is the year �xed e�ect, states is the individual state �xed e�ect,

Naloxonest is a dummy equal to one if a state adopted NALs and 0 otherwise, Xs is a vector of

individual state covariates including other relevant drug related policies, εst is an error term. β

is the causal e�ect of interest and is interpreted as the average treatment e�ect of the adoption of

Naloxone laws. The �xed-e�ect construct aims to reduce the omitted variable bias unaccounted

for within the states such as di�ering demographics, prevalence of drug consumption, or the

existence of pill mills a�ecting the supply of opioids. The identifying assumption relies on

variations in opioid abuse and its corresponding predictors to be time-invariant. Moreover, it

is important that changes in within state factors a�ecting drug abuse are uncorrelated with

the state's decision to adopt the Naloxone Access Laws. In other words, the decision to adopt

7

a Naloxone law may be endogenous. The potential for endogeneity is tentatively alleviated

through the preliminary �xed-e�ect model. This preliminary model assumes endogeneity due

to omitted variable is removed or at the very least reduced for interpretable results.

3.2 Propensity Score Matching

The interpretation of NALs e�ect is complicated by the wide variation of states with regards

to their characteristics and the timing of the law's adoption. The inherent di�erences between

states, included but not limited to the severity of their opioid epidemic, acts as a critical factor

in their decision to adopt NALs and other measure combating the crisis. This introduces biases

in the interpretation of law's e�ect at best and impossible at worst. The core of the issue lies

in creating appropriate treatment and control states. Simply put, a state's decision to adopt

NALs is endogenous to its speci�c opioid problem. Assuming that the opioid issue arises from

a combination of observable and unobservable factors within the state, a matching method can

be utilized to create treatment and control groups by relying not only on the timing of the

NALs adoption but on its characteristics. A common method widely used in the literature is

propensity score matching where states are matched into control and treatment groups based

on their observable characteristics. This method seeks in part to alleviate the issue of common

trend when comparing states before and after the adoption of a policy.

Each observation's propensity score derives from the conditional or predicted probability

of receiving treatment given pre-treatment characteristics. In this case, each state is given

a propensity score on their likelihood to adopt the NALs given their characteristics. States

are assigned to control and treatment groups based their population characteristics, adoption

of other opioids laws such as PDMPs and GSLs, and crime rates. Following Callaway and

Sant'Anna [2018], the standard DID construct with two time periods and no treatment in

period 1 is extended to multiple time periods. Let t = 1, ..., τ represent individual time periods

for each year analyzed, NALt denotes a binary variable equal to one if a state is treated at year

t and zero otherwise. Furthermore, let NALg be a binary variable denoting a state that is �rst

treated at year g, while C is a binary variable accounting for states that are never treated at any

point in time. Each state will have exactly one of either NALg or C equal to one, depending

8

on whether it is receives treatment or not. The generalized propensity score pg(X) indicates

the probability of a state being treated conditional on having covariates X and conditional on

being treated or not. Following, the generalized propensity score for multiple time periods and

multiple states is de�ned as:

pg(X) = prob(NALg = 1|X, NALg + C = 1) = E(NALg|X, NALg + C = 1) (2)

The treatment outcomes are de�ned as Yt =

Yt(1) if NALt=1

Yt(0) otherwise

The observed outcome in each year t is Yt = NALtYt(1) + (1 − NALt)Yt(0). The actual

treatment e�ect is some part of the previous function as a state cannot be simultaneously a

treatment and a control. For a standard DID, the average treatment e�ect is the di�erence

between treated and control, which becomes a simple t-test between both groups. In the case of

multiple group and time periods, the �group average treatment e�ect� at time t for states �rst

treated at year g is expressed as in Callaway and Sant'Anna [2018] as:

ATE(g, t) = E[Yt(1)− Yt(0)|NALg = 1] (3)

Furthermore, it is assumed that {Ys1, Ys2, ..., Ysτ , Xs, NALs1, NALs2, ..., NALsτ}ns=1 is

independently and identically distributed. The outcomes are assumed to be independent of

selection into treatment conditional on a state characteristics, i.e. treatment is exogenous. Also,

the conditional parallel trend states that control and treatment observations would have followed

similar trends in the absence of treatment given state characteristics X. The average outcomes

for groups �rst treated at year g and control observations are assumed to have parallel trends at

year g and any other subsequent time periods, conditional on covariates X which will be time

speci�c.

E[Yt(0)− Yt−1(0)|X, NALg = 1] = E[Yt(0)− Yt−1(0)|X, NALg = 0]

The irreversibility of treatment condition assumes that once treated, a state will be treated

in the next time period. That is NALt = 1implies that NALt+1 = 1 for time periods t = 2, ..., τ .

The overlap or matching condition assumes that for each value of X, there are both treated and

9

control observations. Speci�cally, there will be a positive number of observations that will be

treated in period g and that there is concurrently a positive probability that an individual is

not treated given state covariates X. For all g = 2, ..., τ , it is assumed that p(NALg = 1) > 0

and 0 < pg(X) < 1 states that there is a matched control observation with similar X. Finally,

the balancing condition states that assignment to treatment is independent of the X covariates,

given the same propensity score between di�erent observations.

4 Data

This analysis combines several data sources to look at various outcomes of the opioid epidemic.

The Treatment Episode Data Set (TEDS) provides individual state information on the count

of individuals seeking opioid addiction rehabilitation from 2000 to 2014. TEDS removes all

identi�able information preventing the analysis from capturing individuals seeking treatment

across multiple time periods. Each observation is therefore considered unique. The second

outcome of interest is opioid overdose deaths coming from the National Vital Statistics Sys-

tem (NVSS) Multiple Cause of Death Mortality dataset. The International Classi�cation of

Diseases (ICD-10) is used to identify deaths with external causes of injury related to opioid

misuse. The ICD-10 codes used for opioids are X40-X44, X60-X64, X85, and Y10-Y14 from

years 2000 to 2014. Although the NVSS put restriction on identifying detailed mortality data

since 2005, aggregate information on state death count is still available. Finally, the supply

of legal controlled substances is analyzed through the Drug Enforcement Agency's statistical

reporting arm, the Automation of Reports and Consolidated Orders System (ARCOS) from

2006 to 2014. The ARCOS data is used to look at trends in the supply of legally sold opioids

especially in light of the NALs. Information on commonly distributed opioids such as Codeine,

Fentanyl, Hydrocodone, Hydromorphone, Meperidine, Methadone, Oxycodone, Oxymorphone,

and Tapendatol are analyzed at the state and national levels. Data on violent and property

crimes are obtained through the Uniform Crime Reporting (UCR) system for all years studied.

The UCR data is used as a proxy for the required law enforcement resources per state which

may be related to its corresponding opioid abuse levels. State demographics data come from

the Current Population Survey to provide a background on overall population characteristics.

10

Table 1: Summary Statistics of Selected VariablesVariable (per 100,000

population)

N Mean Standard

Deviation

Minimum Maximum

Opioids Outcomes

Total Grams of Opioids 572 51,031.10 17,581.27 6,777.00 118,645.26

Opioid Overdose Deaths 918 756.48 864.76 13.00 5,403.00

Opioids Rehabilitation Admissions 640 80.47 88.19 0.43 651.33

Demographic Characteristics

White 867 107.54 106.02 16.48 590.51

Black 867 13.19 33.35 0.16 276.83

Hispanic 867 13.31 14.41 0.55 102.31

Male 867 65.67 61.32 11.89 302.63

Female 867 69.03 63.71 12.59 308.70

Married 867 53.91 49.91 10.31 272.92

Divorced 867 10.05 9.66 1.58 47.67

Single 867 62.34 59.75 10.66 276.96

Employed 867 64.14 62.06 12.34 295.36

Unemployed 867 4.03 4.13 0.40 25.89

Medicaid 867 20.13 21.56 1.34 116.95

Medicare 867 16.10 14.81 2.90 66.07

Below Federal Poverty Level 867 16.33 15.13 1.79 93.22

Education Level

Less than High School 867 19.01 16.70 3.58 83.09

High School 867 29.92 27.91 5.90 144.75

Some College 867 18.56 17.88 3.17 103.10

Associate 867 8.87 9.34 1.00 45.54

Bachelor's 867 16.86 16.93 2.09 87.87

Master's 867 6.39 8.14 0.72 67.98

Age

15 to 19 867 10.65 10.18 1.67 58.02

20 to 24 867 7.79 7.41 1.31 36.71

25 to 29 867 6.51 6.53 1.10 40.13

30 to 34 867 9.12 8.73 1.82 46.83

35 to 39 867 7.68 7.26 1.70 45.49

40 to 44 867 10.39 10.06 1.95 56.20

45 to 49 867 10.02 9.70 1.49 49.46

50 to 54 867 9.00 8.63 1.68 39.92

55 to 59 867 7.35 7.09 1.09 32.33

60 to 64 867 5.74 5.50 0.89 27.13

Over 65 867 14.35 13.50 2.60 64.15

Crime Rate

Property Crime 765 3,194.15 837.68 1,524.40 6,409.00

Violent Crime 765 411.22 216.80 78.20 1,637.90

11

4.1 Summary Statistics

The data in table 1 show an average of over 50,000 grams for the most commonly legally dis-

tributed opioids over the between 2006 and 2014 across all states for every 100,000 population.

In 2018, the CDC provided guidelines on the lowest e�ective dosages when prescribing opioids,

with thresholds equal or greater to 50 morphine milligrams equivalents (MME) per day needing

to be carefully assessed, and titration of dosages greater or equal to 90 MME to be avoided or

referred to a pain specialist. For reference a 50 MME/day is equivalent to 50 mg of Hydrocodone

(10 tables of Hydrocodone/acetaminophen 5/300), 33 mg of Oxycodone, or 12 mg of Methadone.

At the rate of 50 MME/day, the opioid supply average of 50,000 grams translates to prescrip-

tions of over 1 million MME/day of hydrocodone, more than 1.5 million MME/day of oxycodone,

and greater than 4.5 million MME/day of methadone on average per 100,000 population in the

US for the time period analyzed. Second, there were on average a little more than 750 opioid

overdose deaths in per 100,000 population nationally from 2000 and 2014, with the number still

alarmingly rising. These deaths are categorized in ICD-10 as accidental, intentional self-harm,

and events of undetermined intents involving poisoning and exposure to noxious substances

such as nonopioid analgesics, antipyretics and antirheumatics; antiepileptic, sedative-hypnotic,

antiparkinsonism and psychotic drugs (not elsewhere classi�ed); narcotics and psychodysleptics

(hallucinogens), not elsewhere classi�ed; other drugs acting on the autonomic nervous system;

other and unspeci�ed drugs, medicaments and biological substances. Furthermore, the TEDS

data show an average of above 80 opioid rehabilitation admission per 100,000 population in the

US between 2000 to 2014. An opioid rehabilitation admission is de�ned as one where �opiates

and synthetics� were reported at admission regardless of whether such substance was primary,

secondary, or tertiary. This de�nition excludes non-prescription Methadone. Admissions in-

volving �opiates and synthetics� were reported in close to 12% of all drug admissions at a count

of about 2.2 million, and accounts for 7% of all primary substance abuse in the time period

analyzed.

Overall, the Current Population Survey dataset shows a generally higher proportion of whites

at about 108 per 100,000 population, a slightly higher proportion of females compared to males

at 13.19 per capita, and the majority is employed. Close to every 20 people out of 100,000

12

has less than a twelfth grade education, about 30 graduated high school, 19 has some college

education whereas about 17 obtained a bachelor's degree. The age distribution is rather even

among all age groups with a higher concentration of the US population concentrated at 40 years

old and up. Finally, the Uniform Crime Report dataset shows property crime rate to be an

average of 3,200 and violent crime at about 410 per 100,000 population, respectively. Property

crimes include burglary, larceny, and motor vehicle theft whereas violent crimes are comprised

of murder and non-negligent manslaughter, rape, robbery, and aggravated assault.

4.2 Parallel Trends Assumption

The parallel trend assumption asserts the enactment of a policy should have no e�ect on the

trends of the outcome of interests. Speci�cally, the adoption of NALs should be completely

random as to avoid the issue of selection bias. Per �gure 4, it is shown the bulk of states

passed the laws between 2013 and 2015. The gap between the �rst state (New Mexico) passing

the laws and the few last ones is about 16 years from 2001 to 2017. There seems to be no

pattern as to the timing of the adoption of the law, other than a collective movement from 2013

onward where states from all over the country decided to adopt the law at once. There were

no documented federal or state incentives to adopt the laws either to the author's knowledge.

Due to data limitation, only states that passed NALs in 2013 and 2014 are aggregated to verify

the parallel trends assumption for all three outcomes of interests. In �gure 1 ,the states passing

NALs in 2013 seem to have steadily followed an upward trend in opioid rehabilitation admission,

regardless of policy adoption; and thus parallel trend is assumed. Those adopting NALs in 2014

experienced an immediate sharp decline in rehab admission during the year of its policy adoption

and then returns to their previous levels. This patterns violates the common trend assumption

but provides insight into the potential spillover e�ects from other states passing NALs in the

previous years.

The picture is much of the same in �gure 2 for the opioid overdose deaths with steady decline

in both years considered, then a particularly large drop occurs followed by an immediate return

to its previous levels. It is important to note the opioid deaths might include heroin deaths. The

OxyContin reformulation of 2010 reduced the number of all opioid deaths while simultaneously

13

Figure 1: Opioid Rehabilitation Admissions

Figure 2: Opioids Overdose Deaths

Figure 3: Total Grams of Legally Distributed Opioids

14

Figure 4: Naloxone Access Laws E�ective Dates

increasing heroin overdoses due to a substitution e�ect [Case and Deaton, 2015]. The most

recent renewed uptick in opioid overdose deaths might be due to the rapidly increasing impact

of the deadly fentanyl. Overall, there seems to be little evidence to suggest a common trend.

Finally, the total grams of legally distributed opioids follows a steady trend of declining amount

prior and after the adoption of NALs in both years considered, as seen in �gure 3. There is

generally a decrease in the legal supply of opioid that can be due to physicians prescribing less

painkillers due to the various laws restricting their ability to do so, especially the PDMPs. The

CDC also advises alternative approaches in treating pain and heavily monitoring the allowable

dosage of opioids for treatment to 50 MME or 90 MME/day. In this case, the common trend

assumption seems plausible.

5 Results

5.1 Fixed-E�ects Regression

NALs have a negative although insigni�cant impact on opioid rehabilitation admissions as seen

in table 2. This result is consistent across various model speci�cations. These �ndings are on par

15

with the literature where various opioid driven policies have no impact on admission into reha-

bilitation. One plausible explanation is the lack of information on what drives an addict to seek

help. The vast majority of patients were admitted through court referrals and thus were involun-

tary. Second, the limits on Medicaid reimbursements for behavioral health an substance abuse

treatment is a deterrent to admitting patients needing help, thus e�ectively under-reporting

the total numbers of addicts impacted by NALs with non-fatal overdoses reversed by Naloxone

for instance. Essentially, the data on opioid rehabilitation admissions fails to re�ect the actual

extent of the epidemic, and perhaps clouds the potential e�ects of NALs.

Also, table 2 shows NALs positively and signi�cantly a�ect opioid overdose deaths. There

are about 140 more opioid deaths per 100,000 population due to NALs, all else being equal. Fur-

thermore, GSLs signi�cantly reduce opioid mortality by a factor of more than 210 less deaths

per capita. The con�icting e�ects of GSLs and NALs are signi�cant and consistent across all

speci�cations, and the interaction of all three policies of interests signi�cantly increase opioid

mortality rate. The signi�cant increase in opioid deaths by NALs seems to con�rm the hypothe-

sis of moral hazard where the presence of an overdose reversal drug as a safety net leads to riskier

behaviors, and ultimately resulting in fatal overdoses. Nevertheless, it is important to note that

opioid overdose deaths are generally trending upward during the time period analyzed. More to

the point, the widespread substitution to a harder narcotic, heroin, and the lacing of heroin with

fentanyl, are very likely a driving force behind the increase in opioid deaths. Nonetheless, it can

be said that policies such as NALs are also responsible for the ampli�cation of the magnitude of

an already growing opioid mortality rate. The deterrent e�ect of GSLs is a positive in curbing

the opioid epidemic by encouraging third parties to intervene with naloxone administration in

the event of an emergency.

Finally, table 2 illustrates how the implementation of NALs results in a signi�cant increase in

the legal supply of opioid by an amount of about 3,000 grams to close to 4,000 grams per 100,000

population. In a scenario of 50 MME/days dosages, NALs resulted in per capital prescriptions

of between 60,000 and 80,000 of hydrocodone, between 91,000 and 120,000 of oxycodone, and

between 250,000 to over 330,000 of methadone during the time period analyzed. Year and state

�xed e�ects were used to mitigate issues of linear trends. However, the interpretation of such

16

Table 2: Fixed E�ect-Regression Results

Outcomes (per100,000 population)

Policies (1)

NAX

(2)

GSL

(3)

PDMP

(4)

Full Model

Opioid Rehabilitation

Admission

NAX 2.402(5.747)

1.238(6.172)

2.171(5.753)

-2.499(7.096)

GSL 3.219(6.202)

-2.627(8.123)

PDMP 4.285(4.543)

4.173(4.563)

All Policies 10.01(10.08)

Number of Observations 600 600 600 600

Overall R-Squared 0.905 0.905 0.905 0.906

Opioid Overdose Deaths

NAX 133.1***(39.66)

187.1***(41.40)

133.4***(39.70)

139.8**(47.41)

GSL -166.3***(41.83)

-241.4***(56.06)

PDMP -15.50(30.88)

-7.685(30.43)

All Policies 137.7*(67.77)

Number of Observations 600 600 600 600

Overall R-Squared 0.955 0.953 0.955 0.953

Total Grams of Opioid

Drugs

NAX 3239.3*(1295.6)

2893.1**(1342.2)

3209.9*(1300.1)

3957.4**(1519.1)

GSL -2292.4(1294.3)

-2091.3(1889.0)

PDMP -463.5(1243.9)

-279.3(1248.4)

All Policies -300.9(2327.6)

Number of Observations 333 333 333 333

Overall R-Squared 0.938 0.939 0.939 0.939

Standard Errors in parentheses* p<0.05, ** p<0.01, *** p<0.001

17

Table 3: Propensity Score Matching Average Treatment E�ectsOutcomes

(per 100,000 population)

Nearest

Neighbor

Matching

Kernel

Matching

Strati�cation

Matching

Opioid Rehabilitation

Admission-46.75(44.51)

-38.83(36.29)

-63.14*(28.64)

Opioid Overdose Deaths 369.0*(158.2)

333.6*(151.0)

366.2*(170.7)

Total Grams of Opioid

Drugs1573.0(5385.0)

4980.9(5264.7)

5735.7(4808.5)

Standard Errors in parentheses* p<0.05, ** p<0.01, *** p<0.001

an increase could be potentially due to a legitimate need for opioids used as painkillers for an

aging population for instance. There has also been a shift in medicine leading to a preference for

opioids to treat pain since the mid nineties. While taking those potential factors into account,

it is also crucial to note that opioid misuse is one of the leading causes for the epidemic. The

availability of more naloxone would encourage a higher consumption of opioids, either prescribed

legitimately or not. This could be in the form of doctor shopping and various forms of drug

seeking behaviors. In that sense, the hypothesis of NALs culminating in riskier behavior is

plausible.

5.2 Propensity Score Matching

The propensity score results matching closely match those of the �xed e�ects regression models

although they are of much larger magnitudes. The average treatment e�ects via various matching

techniques are found in table 3 after randomly assigning treatment and control groups for the

states analyzed. The results suggest NALs have a signi�cant and negative e�ects on opioid

rehabilitation admissions with 63 less per 100,000 population. This could stem from addicts

being able to maintain harmful habits by resorting to naloxone injection in the event of an

emergency. Namely, naloxone is used in lieu of treatment in order to survive an overdose.

Opioids overdose deaths are still signi�cantly impacted by NALs with almost double the size of

the e�ects when using a matching technique. The results in table 3 show there are between 330

to 360 more opioid deaths from NALs per capita, all else being equal. Finally, matching lead to

18

NALs having a positive but insigni�cant impact on the legal supply of opioid.

6 Discussion

Higher rates of opioids mortality and consumption of narcotics, as evidenced by a higher supply

of legally distributed opioid, lend credibility to the likelihood of moral hazard from the adop-

tion of Naloxone Access Laws. The study shows signi�cant and positive e�ects for overdose

deaths and legal distribution of opioids while it �nds no impact on rehabilitation admission

from the opioid law. The lack of information on the cause for seeking addiction treatment is

a data limitation. There are also no information on the degree of access to naloxone for each

state, preventing the study from knowing whether an attempt was made to reverse an overdose.

Overall, these weaknesses were mitigated by assuming complete access to naloxone, if not by

third parties then by �rst responders.

The administration of naloxone is a reactionary measure to an already present addiction

at its worst possible outcome, an overdose. The link between NALs and opioid rehabilitation

admission seems therefore unlikely as emergency treatment is the direct response to a non fatal

overdose rather than rehabilitation. Instead, the relationship between NALs and rehabilitation

rests in determining the actual reason for seeking cure. Also, this issue is further exacerbated

by government reimbursement restrictions for substance abuse treatment, not to mention the

social stigma associated with addiction. A bipartisan bill known as the Medicaid Coverage for

Addiction Recovery Expansion Act was introduced in 2017 to ease the restrictions on Medicaid

reimbursements for substance use disorder treatment centers with up to 40 beds for stays of up

to 60 consecutive days.

Most importantly, this paper aims to study the likelihood of moral hazard from enacting

NALs. According to the results, this hypothesis is plausible from the increasing opioid mortality

rates and the growing legal supply of opioids, excluding illicit narcotics such as heroin. Of

course, the establishment of causality between possession of naloxone and an addict's decision

to undertake riskier behavior is impossible without case by case evidence. However, the growing

trend of overdose deaths coinciding with the implementation of NALs suggest a strong correlation

that should be cause for worry. This study suggests that amid the broader opioid crisis and its

19

plethora of factors, NALs carry a non-negligible responsibility in its worsening. The mechanism

for the e�ectiveness of naloxone is two fold: overdose inducing levels of opioid consumption,

followed by timely and appropriate use of naloxone for reversal. Opioids overdose deaths occur

in either the absence of naloxone, dangerously high level of opioid consumption, or lack of

knowledge on the use of naloxone. To prevent more deaths, there needs to be more education on

the use and limitations of naloxone. Initiatives such as Syringe Services Programs widely known

as needle exchange programs (NEPs) providing medically supervised and safer consumption sites

are a step in the right direction in empowering people su�ering from addiction. The Centers

for Disease Control and Prevention also supports such community e�orts by providing federal

funding to state and local communities via the Consolidated Appropriations Act of 2016.

Finally, the growing supply of legally distributed opioids in relation to the enactment of

NALs is cause for concern. The more opioid is available, the more opportunities there are

for misuse. The safety net of naloxone might encourage more doctor shopping to increase

consumption. Take-back initiatives such as the Drug Enforcement Administration's National

Prescription Drug Take Back Day programs to safely dispose of drug are opportunities to reduce

the supply of legally available drugs, and hopefully reduce the need for naloxone in the event of a

drug overdose. Furthermore, regulations aimed at prescribers to curb over prescription of opioids

could greatly impact the epidemic. In the future, research on NALs can be extended by looking

separately at the components of the law, speci�cally the level of liability protection a�orded

by the law. More insight can be gleaned from understanding the di�erences in the criminal,

civil, or professional immunity the law provide, especially with states varying widely in their

restrictions. NALs might have a di�erential impact on laypersons, prescribers, and dispensers

of the opioid antagonist that could paint a very di�erent picture of the opioid epidemic.

References

Su Albert, II Brason, W Fred, Catherine K Sanford, Nabarun Dasgupta, Jim Graham, and Beth

Lovette. Project lazarus: Community-based overdose prevention in rural north carolina. Pain

medicine, 12(s2), 2011.

20

Abby Alpert, David Powell, and Rosalie Liccardo Pacula. Supply-side drug policy in the presence

of substitutes: Evidence from the introduction of abuse-deterrent opioids. Technical report,

National Bureau of Economic Research, 2017.

Marcus A Bachhuber, Brendan Saloner, Chinazo O Cunningham, and Colleen L Barry. Medical

cannabis laws and opioid analgesic overdose mortality in the united states, 1999-2010. JAMA

internal medicine, 174(10):1668�1673, 2014.

Caleb J Banta-Green, Leo Beletsky, Jennifer A Schoeppe, Phillip O Co�n, and Patricia C

Kuszler. Police o�cers and paramedics experiences with overdose and their knowledge and

opinions of washington state drug overdose-naloxone-good samaritan law. Journal of Urban

Health, 90(6):1102�1111, 2013.

Brantly Callaway and Pedro H. C. Sant'Anna. Di�erence-in-Di�erences with Multiple Time

Periods and an Application on the Minimum Wage and Employment. DETU Working Papers

1804, Department of Economics, Temple University, March 2018. URL https://ideas.

repec.org/p/tem/wpaper/1804.html.

Anne Case and Angus Deaton. Rising morbidity and mortality in midlife among white non-

hispanic americans in the 21st century. Proceedings of the National Academy of Sciences, 112

(49):15078�15083, 2015.

Kerstin Dettmer, Bill Saunders, and John Strang. Take home naloxone and the prevention of

deaths from opiate overdose: two pilot schemes. Bmj, 322(7291):895�896, 2001.

John DiNardo and Thomas Lemieux. Alcohol, marijuana, and american youth: the unintended

consequences of government regulation. Journal of health economics, 20(6):991�1010, 2001.

Maya Doe-Simkins, Alexander Y Walley, Andy Epstein, and Peter Moyer. Saved by the nose:

bystander-administered intranasal naloxone hydrochloride for opioid overdose. American

Journal of Public Health, 99(5):788�791, 2009.

Lynn E Fiellin, Jeanette M Tetrault, William C Becker, David A Fiellin, and Rani A Ho�.

Previous use of alcohol, cigarettes, and marijuana and subsequent abuse of prescription opioids

in young adults. Journal of Adolescent Health, 52(2):158�163, 2013.

21

Romina Lopez Gaston, David Best, Victoria Manning, and Ed Day. Can we prevent drug related

deaths by training opioid users to recognise and manage overdoses? Harm Reduction Journal,

6(1):26, 2009.

Tamara M Haegerich, Leonard J Paulozzi, Brian J Manns, and Christopher M Jones. What we

know, and donât know, about the impact of state policy and systems-level interventions on

prescription drug overdose. Drug and alcohol dependence, 145:34�47, 2014.

Daliah I Heller and Sharon Stancli�. Providing naloxone to substance users for secondary

administration to reduce overdose mortality in new york city. Public health reports, 122(3):

393�397, 2007.

Stop Overdose Illinois. Naloxone fact sheet. http://stopoverdoseil.org/assets/

naloxone-fact-sheet.pdf, 2017. Accessed: 2018-02-11.

Debra Kerr, Paul Dietze, Anne-Maree Kelly, and Damien Jolley. Attitudes of australian heroin

users to peer distribution of naloxone for heroin overdose: perspectives on intranasal admin-

istration. Journal of Urban Health, 85(3):352�360, 2008.

Tara Lagu, Bradley J Anderson, and Michael Stein. Overdoses among friends: drug users are

willing to administer naloxone to others. Journal of substance abuse treatment, 30(2):129�133,

2006.

Stephen E Lankenau, Michelle Teti, Karol Silva, Jennifer Jackson Bloom, Alex Harocopos, and

Meghan Treese. Initiation into prescription opioid misuse amongst young injection drug users.

International Journal of Drug Policy, 23(1):37�44, 2012.

Sarz Maxwell, Dan Bigg, Karen Stanczykiewicz, and Suzanne Carlberg-Racich. Prescribing

naloxone to actively injecting heroin users: a program to reduce heroin overdose deaths.

Journal of addictive diseases, 25(3):89�96, 2006.

Shane R Mueller, Alexander Y Walley, Susan L Calcaterra, Jason M Glanz, and Ingrid A

Binswanger. A review of opioid overdose prevention and naloxone prescribing: Implications

for translating community programming into clinical practice. Substance abuse, 36(2):240�253,

2015.

22

Tinka Markham Piper, Sharon Stancli�, Sasha Rudenstine, Susan Sherman, Vijay Nandi, Allan

Clear, and Sandro Galea. Evaluation of a naloxone distribution and administration program

in new york city. Substance use & misuse, 43(7):858�870, 2008.

David Powell, Rosalie Liccardo Pacula, and Mireille Jacobson. Do medical marijuana laws

reduce addictions and deaths related to pain killers? Technical report, National Bureau of

Economic Research, 2015.

Daniel I Rees, Joseph J Sabia, Laura M Argys, Joshua Latshaw, and Dhaval Dave. With a little

help from my friends: The e�ects of naloxone access and good samaritan laws on opioid-related

deaths. Technical report, National Bureau of Economic Research, 2017.

Christopher J Ruhm. Deaths of despair or drug problems? Technical report, National Bureau

of Economic Research, 2018.

Karen H Seal, Moher Downing, Alex H Kral, Shannon Singleton-Banks, Jon-Paul Hammond,

Jennifer Lorvick, Dan Ciccarone, and Brian R Edlin. Attitudes about prescribing take-home

naloxone to injection drug users for the management of heroin overdose: a survey of street-

recruited injectors in the san francisco bay area. Journal of Urban Health, 80(2):291�301,

2003.

Karen H Seal, Robert Thawley, Lauren Gee, Joshua Bamberger, Alex H Kral, Dan Ciccarone,

Moher Downing, and Brian R Edlin. Naloxone distribution and cardiopulmonary resuscitation

training for injection drug users to prevent heroin overdose death: a pilot intervention study.

Journal of Urban Health, 82(2):303�311, 2005.

Karl A Sporer and Alex H Kral. Prescription naloxone: a novel approach to heroin overdose

prevention. Annals of emergency medicine, 49(2):172�177, 2007.

Michele M Straus, Udi E Ghitza, and Betty Tai. Preventing deaths from rising opioid overdose in

the us�the promise of naloxone antidote in community-based naloxone take-home programs.

Substance abuse and rehabilitation, 4:65, 2013.

Karin E Tobin, Susan G Sherman, Peter Beilenson, Christopher Welsh, and Carl A Latkin. Eval-

23

uation of the staying alive programme: training injection drug users to properly administer

naloxone and save lives. International Journal of Drug Policy, 20(2):131�136, 2009.

Alexander Y Walley, Maya Doe-Simkins, Emily Quinn, Courtney Pierce, Ziming Xuan, and

Al Ozono�. Opioid overdose prevention with intranasal naloxone among people who take

methadone. Journal of Substance Abuse Treatment, 44(2):241�247, 2013.

Eliza Wheeler, Peter J Davidson, T Stephen Jones, and Kevin S Irwin. Community-based opioid

overdose prevention programs providing naloxoneâunited states, 2010. MMWR. Morbidity and

mortality weekly report, 61(6):101, 2012.

CDC Wonder. Overdose death rates. https://www.drugabuse.gov/related-topics/

trends-statistics/overdose-death-rates, 2017. Accessed: 2018-02-05.

Nancy Worthington, Tinka Markham Piper, Sandro Galea, and David Rosenthal. Opiate users'

knowledge about overdose prevention and naloxone in new york city: a focus group study.

Harm Reduction Journal, 3(1):19, 2006.

Appendix

24

Table 4: Opioid Rehabilitation Admission Fixed E�ects Results

Variables(1)

NAX

(2)

GSL

(3)

PDMP

(4)

Full Model

Opioid Related Policies

NAX2.402

(5.747)

1.238

(6.172)

2.171

(5.753)

-2.499

(7.096)

GSL3.219

(6.202)

-2.627

(8.123)

PDMP4.285

(4.543)

4.173

(4.563)

NAX*GSL*PDMP10.01

(10.08)

Demographic Characteristics

White0.250

(0.262)

0.245

(0.262)

0.216

(0.264)

0.233

(0.265)

Black1.819**

(0.658)

1.783**

(0.662)

1.759**

(0.661)

1.682*

(0.662)

Amerindian1.143*

(0.483)

1.167*

(0.485)

1.113*

(0.484)

1.084*

(0.483)

Asian0.757*

(0.299)

0.748*

(0.299)

0.717*

(0.302)

0.717*

(0.302)

Hispanic-0.936*

(0.461)

-0.882

(0.473)

-0.963*

(0.462)

-0.898

(0.476)

Male3.376***

(0.958)

3.426***

(0.964)

3.384***

(0.959)

3.196***

(0.943)

Married0.433

(0.833)

0.381

(0.840)

0.387

(0.834)

0.534

(0.803)

Divorced4.549**

(1.465)

4.526**

(1.4657)

4.424**

(1.471)

4.518**

(1.465)

Single-2.146**

(0.767)

-2.171**

(0.769)

-2.171**

(0.767)

-2.080**

(0.758)

Unemployed-1.426

(1.307)

-1.399

(1.309)

-1.362

(1.309)

-1.487

(1.302)

Education Level

Less than High School1.028

(0.828)

0.995

(0.831)

1.026

(0.828)

1.043

(0.829)

High School Degree1.114

(0.711)

1.090

(0.713)

1.237

(0.723)

1.245

(0.726)

Bachelor's Degree-1.214

(1.025)

-1.181

(1.028)

-1.196

(1.025)

-1.128

(1.029)

Master's Degree4.603**

(1.514)

4.602**

(1.515)

4.737**

(1.521)

4.813**

(1.520)

Professional Degree7.850*

(3.483)

7.761*

(3.490)

8.427*

(3.537)

8.634*

(3.523)

N 600 600 600 600

Overall R-Squared 0.955 0.953 0.955 0.953Standard errors in parentheses

* p<0.05, ** p<0.01, *** p<0.001

25

Table 5: Continued Opioid Rehabilitation Admission Fixed E�ects Results

Variables (1)

NAX

(2)

GSL

(3)

PDMP

(4)

Full Model

Age Levels

20 to 24 -0.724

(1.473)

-0.717

(1.474)

-0.700

(1.474)

-0.727

(1.472)

25 to 29 2.484

(1.652)

2.476

(1.653)

2.558

(1.654)

2.435

(1.654)

30 to 34 -4.572**

(1.484)

-4.525**

(1.488)

-4.466**

(1.488)

-4.701**

(1.461)

35 to 39 0.0178

(1.645)

0.0714

(1.649)

0.135

(1.650)

0.128

(1.653)

40 to 44 -2.858

(1.484)

-2.799

(1.490)

-2.806

(1.485)

-2.990*

(1.462)

45 to 49 -2.014

(1.300)

-1.947

(1.308)

-1.873

(1.309)

-1.960

(1.305)

50 to 54 -5.726***

(1.393)

-5.706***

(1.395)

-5.617***

(1.398)

-5.714***

(1.389)

55 to 59 -0.656

(1.469)

-0.687

(1.471)

-0.799

(1.477)

-0.813

(1.479)

60 to 64 6.159***

(1.659)

6.210***

(1.663)

6.189***

(1.659)

6.099***

(1.648)

Other Covariates

Medicaid Recipient 1.172**

(0.433)

1.163**

(0.434)

1.120*

(0.437)

1.039*

(0.440)

Employer Provided

Insurance

-2.991***

(0.698)

-2.994***

(0.698)

-3.005***

(0.698)

-3.041***

(0.697)

Below Federal Poverty

Level

-1.172*

(0.569)

-1.170*

(0.569)

-1.107

(0.573)

-1.045

(0.577)

Larceny Theft 0.0453***

(0.00895)

0.0456***

(0.00898)

0.0464***

(0.00902)

0.0461***

(0.00907)

N 600 600 600 600

Overall R-Squared 0.955 0.953 0.955 0.953

Standard errors in parentheses* p<0.05, ** p<0.01, *** p<0.001

26

Table 6: Opioid Overdose Deaths Fixed E�ects Results

Variables(1)

NAX

(2)

GSL

(3)

PDMP

(4)

Full Model

Opioid Related Policies

NAX133.1***

(39.66)

187.1***

(41.40)

133.4****

(39.70)

139.8**

(47.41)

GSL-166.3***

(41.83)

-241.4***

(56.06)

PDMP-15.50

(30.88)

-7.685

(30.43)

NAX*GSL*PDMP137.7*

(67.77)

Demographic Characteristics

White1.658

(1.720)

1.922

(1.697)

1.740

(1.729)

2.120

(1.703)

Black-9.751*

(4.471)

-8.899*

(4.414)

-9.635*

(4.480)

-8.811*

(4.409)

Asian0.495

(1.981)

0.782

(1.954)

0.598

(1.993)

0.835

(1.960)

Amerindian3.036

(3.307)

1.798

(3.276)

3.089

(3.311)

1.643

(3.273)

Male-5.014

(6.090)

-5.971

(6.010)

-4.965

(6.095)

-7.027

(6.022)

Divorced1.981

(9.004)

0.843

(8.882)

2.327

(9.036)

1.173

(8.891)

Single-6.322

(5.302)

-7.168

(5.232)

-6.275

(5.307)

-6.737

(5.226)

Unemployed3.727

(8.908)

2.045

(8.793)

3.311

(8.953)

1.388

(8.813)

Education Level

Less than High School21.69**

(7.731)

20.66**

(7.627)

22.03**

(7.766)

20.69**

(7.642)

High School Degree18.74**

(5.711)

18.69***

(5.631)

18.67**

(5.717)

18.84***

(5.622)

Some College21.80**

(7.061)

21.83**

(6.962)

22.19**

(7.109)

21.47**

(6.995)

Bachelor's Degree5.614

(8.929)

3.838

(8.815)

5.897

(8.953)

3.854

(8.816)

N 640 640 640 640

Overall R-Squared 0.953 0.955 0.953 0.955

27

Table 7: Continued Opioid Overdose Deaths Fixed E�ects Results

Variables (1)

NAX

(2)

GSL

(3)

PDMP

(4)

Full Model

Age Levels

15 to 19 -19.63

(10.52)

-16.04

(10.41)

-19.88

(10.54)

-15.51

(10.41)

20 to 24 -6.317

(11.14)

-5.841

(10.99)

-6.604

(11.17)

-4.710

(11.00)

25 to 29 -17.61

(12.70)

-15.69

(12.53)

-18.13

(12.75)

-15.60

(12.55)

30 to 34 -4.456

(10.62)

-3.446

(10.48)

-4.853

(10.66)

-3.234

(10.49)

35 to 39 -12.77

(12.14)

-12.31

(11.97)

-13.37

(12.21)

-12.84

(12.01)

40 to 44 -4.215

(10.52)

-2.408

(10.39)

-4.548

(10.55)

-2.312

(10.39)

45 to 49 -14.08

(9.522)

-14.62

(9.389)

-14.85

(9.652)

-15.19

(9.490)

50 to 54 -4.744

(10.30)

-3.409

(10.16)

-5.347

(10.37)

-2.634

(10.22)

55 to 59 -28.47**

(10.33)

-25.64*

(10.21)

-28.06**

(10.37)

-24.78*

(10.22)

60 to 64 -25.38*

(11.12)

-24.29*

(10.97)

-25.62*

(11.14)

-22.84*

(10.98)

Over 65 -27.74

(14.80)

-22.05

(14.66)

-28.06

(14.83)

-22.36

(14.65)

Other Covariates

Medicaid Recipient -1.156

(2.993)

-0.730

(2.953)

-0.975

(3.017)

-1.435

(2.993)

Medicare Recipient 10.69

(14.22)

5.742

(14.08)

10.64

(14.24)

6.323

(14.05)

Employer Provided

Insurance

5.266

(4.842)

5.765

(4.776)

5.257

(4.845)

5.656

(4.766)

Below Federal Poverty

Level

1.655

(3.888)

1.766

(3.834)

1.504

(3.903)

2.549

(3.860)

Aggravated Assault -0.884**

(0.298)

-1.118***

(0.300)

-0.896**

(0.299)

-1.195***

(0.302)

N 640 640 640 640

Overall R-Squared 0.953 0.955 0.953 0.955

28

Table 8: Total Grams of Legally Distributed Opioid Fixed E�ects Results

Variables(1)

NAX

(2)

GSL

(3)

PDMP

(4)

Full Model

Opioid Related Policies

NAX3239.3*

(1295.6)

2893.1**

(1342.2)

3209.9*

(1300.1)

3957.4**

(1519.1)

GSL-2292.4

(1294.3)

-2091.3

(1889.0)

PDMP-463.5

(1243.9)

-279.3

(1248.4)

NAX*GSL*PDMP-300.9

(2327.6)

Demographic Characteristics

White229.5

(125.2)

229.1

(124.6)

230.9

(125.4)

228.5

(125.6)

Black-648.7*

(316.7)

-627.1*

(315.6)

-652.4*

(317.3)

-630.3*

(317.0)

Male-248.8

(295.4)

-311.6

(296.3)

-244.0

(296.1)

-303.2

(300.1)

Married238.5

(249.2)

273.8

(249.0)

233.0

(250.1)

273.5

(252.0)

Single526.1*

(250.5)

521.3*

(249.5)

531.9*

(251.4)

524.2*

(251.1)

Unemployed386.1

(298.8)

404.7

(297.8)

370.1

(302.4)

94.6

(302.1)

Education Level

High School Degree161.1

(223.1)

168.8

(222.3)

143.5

(228.4)

155.9

(228.5)

Associate343.4

(361.8)

335.1

(360.4)

339.0

(362.6)

331.0

(362.1)

Bachelor's Degree-559.1

(299.6)

-587.4*

(298.8)

-562.4

(300.2)

-590.7*

(300.3)

N 333 333 333 333

Overall R-Squared 0.938 0.939 0.938 0.939

29

Table 9: Continued Total Grams of Legally Distributed Opioid Fixed E�ects Results

Variables (1)

NAX

(2)

GSL

(3)

PDMP

(4)

Full Model

Age Levels

15 to 19 -252.2

(360.0)

-229.5

(358.8)

-252.0

(360.6)

-230.8

(360.2)

20 to 24 -1304.3**

(450.5)

-1279.6**

(448.9)

-1308.2**

(451.3)

-1284.4**

(451.0)

25 to 29 268.2

(465.5)

357.5

(466.3)

263.1

(466.4)

351.8

(468.6)

30 to 34 -1096.4*

(491.5)

-1050.6*

(490.2)

-1077.5*

(494.9)

-1044.6*

(496.2)

35 to 39 -1091.9*

(484.4)

-1067.0*

(482.7)

-1097.9*

(485.5)

-1069.0*

(485.0)

40 to 44 -477.8

(408.3)

-443.1

(407.1)

-483.7

(409.3)

-450.8

(409.9)

45 to 49 -116.1

(435.7)

-98.46

(434.0)

-136.6

(439.9)

-114.3

(439.8)

50 to 54 -575.5

(405.2)

-544.9

(404.0)

-563.9

(407.1)

-541.4

(407.4)

55 to 59 -562.7

(407.8)

-537.0

(406.4)

-530.1

(417.7)

-516.9

(417.0)

60 to 64 849.1

(473.0)

875.2

(471.4)

860.6

(474.8)

877.9

(475.1)

Over 65 -595.4

(528.5)

-520.2

(528.1)

-583.0

(530.4)

-514.8

(531.0)

Other Covariates

Medicaid Recipient -88.13

(105.6)

-67.89

(105.8)

-86.55

(105.9)

-65.21

(107.3)

Medicare Recipient 233.8

(494.3)

178.9

(493.3)

230.2

(495.2)

176.2

(495.2)

Below Federal Poverty

Level

-356.9*

(173.6)

-377.6*

(173.3)

-358.8*

(174.0)

-378.8*

(174.0)

Aggravated Assault 39.88**

(13.73)

35.34*

(13.92)

38.55**

(14.21)

34.84*

(14.54)

N 333 333 333 333

Overall R-Squared 0.938 0.939 0.938 0.939

30