AER: Insights 2021, 3(4): 523–539

https://doi.org/10.1257/aeri.20200653

Teenage Driving, Mortality, and Risky Behaviors†

By Jason Huh and Julian Reif*

We investigate the effect of teenage driving on mortality and risky

behaviors in the United States using a regression discontinuity

design. We estimate that total mortality rises by 5.84 deaths per

100,000 (15 percent) at the minimum legal driving age cutoff, driven

by an increase in motor vehicle fatalities of 4.92 deaths per 100,000

(44 percent). We also find that poisoning deaths, which are caused

primarily by drug overdoses, rise by 0.31 deaths per 100,000 (29 per-

cent) at the cutoff and that this effect is concentrated among females.

Our findings show that teenage driving contributes to sex differences

in risky drug use behaviors. (JEL I12, J13, J16, R41)

Suicides and motor vehicle accidents are the two leading causes of death for teen-

agers in the United States. Both often involve substance abuse, which itself is also a

leading cause of teenage death. Over 25 percent of all teenage hospitalizations are

related to mental health or substance abuse disorders (Heslin and Elixhauser 2016).

Yet, the determinants of risky behaviors among youth remain poorly understood,

and what little we know about the drivers of risky behaviors among adults may not

apply to teenagers. For example, drug overdose deaths among teenagers declined

between 2007 and 2014, in stark contrast to the significant nationwide increase in

adult “deaths of despair” that has received much attention from researchers (Case

and Deaton 2015, 2017).

We investigate the effects of teenage driving on mortality and risky behaviors.

While driving undoubtedly increases motor vehicle mortality risk, the magnitude

of this risk is hard to quantify. Moreover, driving enables teenagers to participate in

unsupervised risky behaviors away from home, which may in turn lead to changes in

mental health or drug use that have additional effects on mortality risk. Investigating

these links can shed light on the determinants of risky behaviors among youth.

However, identifying the effects of driving on teenage outcomes is challenging.

­Individual-level data on driving behaviors are scarce, and comparing the behaviors

* Huh: Department of Economics, Rensselaer Polytechnic Institute (email: [email protected]); Reif: Gies

College of Business, University of Illinois, and NBER (email: [email protected]). Amy Finkelstein was coeditor

for this article. We are thankful for comments from Kitt Carpenter, Tatyana Deryugina, Carlos Dobkin, Yingying

Dong, Michael Frakes, Don Fullerton, Jonathan Hall, Ellen Meara, Nolan Miller, Nicolas Ziebarth, and participants

at the Chicago Booth Junior Health Economics Summit and the UIUC Applied Economics Workshop.

†

Go to https://doi.org/10.1257/aeri.20200653 to visit the article page for additional materials and author

disclosure statement(s).

523

524 AER: INSIGHTS DECEMBER 2021

of drivers to ­nondrivers is unlikely to yield causal estimates because the decision

to obtain a license is voluntary. In addition, detecting changes in important but rare

outcomes such as ­drug-related mortality requires large sample sizes.

We overcome these challenges by using a regression discontinuity (RD) approach

to identify the causal effect of teenage driving on a number of outcomes.1 Our research

design exploits variation in driving eligibility caused by the minimum legal driving age

(henceforth, “minimum driving age” or “MDA”), which creates large differences in

the teenage driver population on either side of the MDA cutoff. We employ a confiden-

tial dataset that includes information about month and year of birth for over 500,000

teenage deaths during 1983–2014, which enables us to compare mortality rates for

teenagers just above the MDA to mortality rates for teenagers just below the MDA.

We estimate that driving eligibility increases teenage mortality by 5.84 deaths per

100,000 (15 percent) at the MDA cutoff. This effect is driven primarily by an increase

in motor vehicle fatalities of 4.92 per 100,000 (44 percent).

We also estimate that teenage poisoning deaths rise by 0.314 deaths per 100,000

(29 percent) at the cutoff. This poisoning effect is driven by a stark rise in female

drug overdose deaths of 0.646 per 100,000 (78 percent) and an accompanying rise

in female gas poisoning deaths of 0.127 per 100,000 (82 percent).2 These deaths

reflect changes in both suicides and accidental deaths, although our analysis sug-

gests that the increase in poisoning suicides reflects substitution away from other

methods of suicide. By contrast, we find no effect on male poisoning deaths. This

null effect is meaningfully precise: the 95 percent confidence interval rules out the

corresponding female estimate.

The most common MDA during our sample period is 16, which raises the con-

cern that our results might be driven by a “birthday effect” or some other regulation

that takes effect at age 16, such as the federal minimum legal working age or a

state’s minimum school-leaving age. However, the discontinuities that we observe

are l­ong lasting, which is inconsistent with a temporary birthday effect, and we

do not detect changes in the probability of working or leaving school at the MDA

cutoff. Furthermore, our main results are similar when we limit our analysis to the

subsample of states with MDAs other than 16. We therefore conclude that the MDA

is the causal mechanism underlying our results.

Youth risky behaviors differ significantly by sex. During our sample period,

29 percent of young female deaths from suicides and accidents were caused by

poisonings, compared to just 15 percent for young males. Prior studies have also

noted that poisonings make up a significantly larger share of young female suicides

than young male suicides (Ruch et al. 2019). Our results reveal that teenage driving

contributes to these behavioral differences among male and female youth.

Much of the research on the causal determinants of drug abuse comes from stud-

ies of adults. For example, several studies find that drug overdose deaths increase

after the first of the month and following the receipt of money, suggesting a causal

relationship between a “full wallet” and substance abuse (Phillips, Christenfeld, and

1

Our analysis controls for the ­family-wise error rate in order to address the multiple inference concern that

arises when testing many hypotheses. See Section III for details.

2

During our sample period, about 80 percent of teenage poisoning deaths are caused by drug overdoses, and

about 20 percent are caused by gas poisonings. Most gas poisoning deaths are caused by carbon monoxide.

VOL. 3 NO. 4 HUH AND REIF: TEENAGE DRIVING, MORTALITY, AND RISKY BEHAVIORS 525

Ryan 1999; Riddell and Riddell 2006; Dobkin and Puller 2007; Evans and Moore

2012). Less is known about the causal determinants of drug abuse among adoles-

cents.3 Indeed, Gruber (2001) calls for economists to pay more attention to the

­risk-taking behavior of youth, noting that while economic incentives appear to mat-

ter, most of the variation remains unexplained. Anderson (2010), who finds no effect

of a Montana a­ nti-drug advertising campaign on methamphetamine use among high

schoolers, emphasizes the need for further research on the determinants of illegal

drug use among the young. Our study advances this literature by using a novel

source of exogenous variation to uncover a strong, causal relationship between driv-

ing and drug overdose deaths among teenagers.

The remainder of our paper is organized as follows. Section I provides back-

ground information. Section II describes our data. Section III outlines our empirical

strategy. Section IV describes our results, and Section V concludes.

I. Background

A. Teenage Driving

Every US state requires drivers to be licensed. Teenagers begin the licensing

process by obtaining a learner’s permit, which allows them to drive under adult

supervision. Depending on the state, the adult must be at least 18–25 years of age

and have up to five years of driving experience. The minimum legal age for obtain-

ing a learner’s permit ranges from 14 to 16 over our 1983–2014 sample period.

With rare exception, teenagers must then complete a driver’s education course and

­behind-the-wheel training to become eligible to take their state’s driving test, which

typically consists of two components: a written test and a b­ ehind-the-wheel test. The

teenager receives her driver’s license after passing both components.

Beginning in 1996, states began adopting Graduated Driver Licensing (GDL)

programs. Teenagers who live in a state with a GDL program earn an “intermedi-

ate” driver’s license after passing their state’s driving test. An intermediate license

imposes driving restrictions on newly licensed teenagers in an effort to encourage

safer driving. For example, it typically prohibits unsupervised driving during certain

nighttime hours and limits the number and age of passengers in the teenage driver’s

vehicle. These restrictions are lifted after a set period of time that ranges from three

months up to two years, at which point the teenager earns a “full” license.

Our study focuses on the MDA, the minimum legal age at which teenagers become

eligible to pass their state’s driving test. In states with GDL programs in place, this

eligibility threshold corresponds to the minimum legal intermediate licensing age.

The MDA, which varies by state and over time, ranges from 14 to 18 years of age

during our 1983–2014 sample period. On average, about 40 percent of ­16-year-olds

3

Most of the prior literature on youth drug use focuses on tobacco, alcohol, and marijuana (e.g., Glied 2002;

Cawley, Markowitz, and Tauras 2004; Carpenter et al. 2019). Drug overdose deaths among teenagers, however, are

mostly caused by opioids (both illegal and prescription) and sedatives. The teenage driving literature has investi-

gated the effects of graduated driver licensing laws on motor vehicle fatalities (Dee, Grabowski, and Morrisey 2005;

Morrisey et al. 2006; ­Karaca-Mandic and Ridgeway 2010; Gilpin 2019) and crime (Deza and Litwok 2016) and

has investigated the effects of the MDA on crash risk (Foss et al. 2011; Chapman, Masten, and Browning 2014;

Curry et al. 2015).

526 AER: INSIGHTS DECEMBER 2021

have a driver’s license during this period, although this declines to less than 30

percent in more recent years (online Appendix B.1). A similar decline occurs for

vehicle miles traveled.

B. Teenage Mortality

Accidents and suicides are the two leading causes of death among teenagers,

respectively. Motor vehicle fatalities, which comprise over 60 percent of all acciden-

tal teenage deaths, are nearly as common as teenage suicides (CDC 2018). About

25 percent of teenage motor vehicle fatalities are ­alcohol related, and over 50 per-

cent occur during the nighttime (Dee and Evans 2001). Motor vehicle fatality rates

have declined significantly in recent decades, falling by over 50 percent during our

1983–2014 sample period. Prior studies have attributed this decline to a reduction

in drunk driving, an increase in seat belt use, and the introduction of GDL pro-

grams (Dee and Evans 2001, Gilpin 2019). As of 1998, all states have implemented

­zero-tolerance drunk driving laws that set a limit of 0.02 percent blood alcohol con-

centration (​∼​1 drink) or lower for drivers under age 21 (Carpenter 2004). Air bags

were also introduced during this time period, but Dee and Evans (2001) argue that

they played only a small role in reducing teenage motor vehicle fatalities.

Poisonings are the ­second leading cause of accidental death among teenagers (CDC

2018). ­Two-thirds of drug overdose deaths for youth ages 15–24 are related to her-

oin and illegal opioids, and ­one-third are related to sedatives and prescription opioids

(National Institute on Drug Abuse 2019). Prior studies find that young females exceed

males in their nonmedical use of sedatives and prescription opioids and are also more

likely to overdose than males (Cotto et al. 2010, Lyons et al. 2019).

Poisonings also make up a significantly larger share of young female suicides

than young male suicides (Ruch et al. 2019). Teenage females show higher rates

of suicidal thinking and are more likely to attempt suicide than teenage males, but

males die by suicide at higher rates (Swahn and Bossarte 2007). Section IIB docu-

ments sex disparities in poisoning deaths—both accidental and suicidal—over our

sample period.

II. Data

A. MDA Laws

We obtained data on MDA laws from the Insurance Institute for Highway Safety

for the years 1995–2014 (IIHS 1995–2014). Data for the years 1983–1994 were

­hand collected from databases of state session laws (HeinOnline 1983–1994). These

data are reported in the online Appendix (Table B.1).

B. Mortality

We measure mortality using the National Vital Statistics (NCHS 1983–2014).

This dataset is based on death certificate records and includes information on dece-

dents’ month and year of death, cause of death, race, and sex. The medical classi-

fication codes used to define the cause of death outcomes employed in our analysis

VOL. 3 NO. 4 HUH AND REIF: TEENAGE DRIVING, MORTALITY, AND RISKY BEHAVIORS 527

are reported in the online Appendix (Table B.2). We obtained a r­ estricted-use ver-

sion of the National Vital Statistics for the years 1983–2014 that includes infor-

mation on decedents’ state of residence and month and year of birth. We use these

data to calculate age in months at death for all decedents. We then aggregate to the

­age-in-months level and calculate a­ ge-specific deaths per 100,000 p­ erson-years by

combining these count data with population estimates provided by the Surveillance,

Epidemiology, and End Results (SEER) Program (NCI 1983–2014).4 Our final

dataset includes information on 501,193 teenage deaths.

The average death rate for teenagers ages 15–19 is 68.63 deaths per 100,000

during our sample period (online Appendix Table B.3). The majority of these deaths

are attributable to external causes, and the leading external cause of teenage death

is motor vehicle accidents.5 The motor vehicle mortality rate for ages 15–19 is over

five times larger than that for ages 10–14. This mortality rate increases further for

ages 20–24 but then decreases for ages 25–29.

Suicides and accidents comprise 25 percent of all deaths for ages 15–19. Males in

this age group are four times more likely than females to die by suicide or accident.

The cause of these deaths varies by sex. For ages 15–19, 38 percent of male suicides

and accidents are caused by firearms, compared to just 22 percent for females. By

contrast, only 15 percent of male suicides and accidents are caused by poisonings,

compared to 29 percent for females.

Deaths from all causes and deaths from motor vehicle accidents have both

declined significantly for males and females during our sample period (online

Appendix Figure B.3). However, although male deaths from suicides and acci-

dents have declined steadily since 1994, female deaths from suicides and accidents

increased beginning in 2003.

Male and female poisoning deaths evolved similarly from 1983 until 2000

(online Appendix Figure B.4). The trends then diverged: the male poisoning death

rate climbs and falls significantly over the next 15 years, while the female poison-

ing death rate remains relatively constant after 2003. Neither group experiences

sustained increases in poisoning deaths after 2007. By contrast, Case and Deaton

(2015) document a steeply increasing trend in poisoning deaths among midlife

whites for 2007–2013.

Our main estimates combine suicides and accidents into one category to mini-

mize measurement error concerns (Cutler, Glaeser, and Norberg 2001; Alexander

and Schnell 2019). When someone dies from a drug overdose, for example, it may

not be clear whether the death should be classified as a suicide or an accident. We

report estimates separately for suicides and accidents in the online Appendix.

C. Driving Behaviors

We measure driving behaviors using the National Longitudinal Study of

Adolescent to Adult Health (hereafter, “Add Health”; Add Health 1995–1996). This

4

These population data are available only for integer ages. When calculating ­age-specific death rates, we divide

the count of deaths for a specific age in months by ­one-twelfth of the corresponding integer-age population.

5

Deaths not caused by external factors are attributable to “internal causes.” Cancer, circulatory system diseases,

and nervous system diseases are the three largest subcategories within internal causes, and they make up more than

half of all deaths due to internal causes at these ages.

528 AER: INSIGHTS DECEMBER 2021

nationally representative study began in 1994 with a classroom survey of about

20,000 students in grades 7–12. The study then followed up with a series of ­in-home

­interviews in 1995 and 1996. We obtained a r­estricted-use version of the i­n-home

survey data that includes month and year of birth. After excluding observations

with missing data, our sample includes 32,307 ­person-year observations (online

Appendix B.4). This sample includes respondents ranging in age from 11 to 21;

97.9 percent of respondents are between the ages of 13 and 19. We aggregate these

data to the a­ ge-in-months level using Add Health’s ­cross-sectional weights.

The ­in-home survey asks respondents whether they have a driver’s license and

whether they drive 0, 1–50, 51–100, or “over 100” miles per week, which we use

to measure vehicle miles driven. We assign values of 25 and 75 to respondents who

selected the ranges 1–50 and 51–100, respectively. For “over 100,” our baseline

specification assigns a value of 150. By way of comparison, the typical adult driver

drove 265 miles per week in 1996 (Federal Highway Administration 1997, 1992–

2002). To account for uncertainty, we also report results from an alternate specifica-

tion that instead assigns a value of 265 to the “over 100” response.

Add Health also asks questions about drug consumption and mental health. We

do not present results for these outcomes because the low prevalence of the most

important behaviors (suicide attempts and illegal drug consumption) and the sur-

vey’s small sample size cause the analysis to be underpowered.

III. Empirical Strategy

We employ an RD design to identify the effect of driving eligibility on teenage driv-

ing behaviors and mortality. Eligibility depends on year, age, and state of residence.

For analytical convenience, we recenter the age variable for decedents in our data by

measuring it in months from the MDA law in force during the month of death.

Our main identifying assumption is that assignment to either side of the MDA

threshold is as good as random. This assumption is very reasonable: age cannot be

manipulated, and we do not suffer from sample selection bias because we observe

the universe of deaths.

We estimate the following model:

​Y​a​​  = αAG​Ea​​​  + βPOS​T​a​​  + γ​(POS​Ta​​​  × AG​E​a​​)​  + δ​D​a​​  + ​ε​a​​​.

(1) ​

The dependent variable, ​​Y​a​​​, is an outcome for the ­one-month age cell a​ ​. The run-

ning variable, ​AG​E​a​​​, is measured in months from the MDA, and ​POS​Ta​​​​is an indi-

cator equal to 1 if ​AG​Ea​​​  ≥ 0​. This indicator suffers from measurement error at

the age cutoff because we do not know whether a teenager who died in the month

she reaches the MDA was over or under the MDA on the day of her death. We

remove the bias associated with this measurement error by including the indicator

​ ​a​​​, which is equal to 1 when A

variable ​D ​ G​Ea​​​  = 0​and is 0 otherwise (Dong 2015).

All of our regressions employ a triangular kernel. Our preferred specification

employs a m ­ ean-squared error (MSE) optimal bandwidth that is constant on each side

of the cutoff but allowed to vary across outcomes. We report robust ­bias-corrected

confidence intervals that account for the possibility that our estimating equation is

misspecified (Calonico, Cattaneo, and Titiunik 2014).

VOL. 3 NO. 4 HUH AND REIF: TEENAGE DRIVING, MORTALITY, AND RISKY BEHAVIORS 529

We address multiple inference concerns by controlling for the f­ amily-wise error

rate using the ­Sidak-Holm ­step-down correction.6 In our setting, the family includes

all 13 mortality outcomes reported in our main table. When estimating models sep-

arately for males and females, we include outcomes from both subgroups in the

family, that is, we report p​ ​-values that adjust for testing 26 different hypotheses.

IV. Results

A. Driving

We begin by estimating the effect of driving eligibility on license status and

vehicle miles driven. Figure 1, panel A shows that about 25 percent of teenagers

obtain a license within their first two months of eligibility and that this proportion

rises to over 50 percent after 12 months. The increase at the age cutoff (value 0 on

the ­x-axis) is attenuated because of the measurement error discussed in Section III.

Figure 1, panel B shows a corresponding rise in vehicle miles driven. Miles driven is

positive prior to the MDA because teenagers with learner’s permits can drive under

parental supervision.

The first three rows of panel A in Table 1 report RD estimates of ​β​from equation

(1) for our driving outcomes. Column 2 reports that driving eligibility increases a

teenager’s probability of obtaining a driver’s license by 18.6 percentage points. It

also increases her annual driving amount by 375 miles using the baseline defini-

tion of average vehicle miles driven, or by 575 miles using the alternate definition

(online Appendix Figure A.1). The increase in licensing rates at the cutoff is similar

for males and females, but the increase in vehicle miles driven is larger for males.

B. Mortality

Figure 1, panels C and D show the effect of driving eligibility on total mortality

and motor vehicle fatalities. Both of these mortality outcomes increase by about

5 deaths per 100,000 during the first two months of driving eligibility, for both males

and females. There is little change in their overall trends: total mortality and motor

vehicle fatalities both increase with age at about the same rate in the periods before

and after the age cutoff.

Figure 2, panel A illustrates a stark increase in female poisoning deaths in the

months immediately following the age cutoff. Figure 2, panels B and C show that

female deaths from drug overdoses and gas poisonings—which together comprise

all poisoning deaths—both rise significantly at the age cutoff. By contrast, Figure 3

shows that trends in male poisoning deaths appear continuous at the age cutoff.

Panel B in Table 1 reports estimates of β

​ ​from equation (1) for 13 different causes

of death.7 Column 2 reports that total mortality increases by 5.84 deaths per 100,000

6

This correction is conservative because it does not account for the significant collinearity among our outcomes

(see Appendix C of Jones, Molitor, and Reif 2019). We are unaware of any r­ esampling-based multiple testing cor-

rections for RD designs with discrete running variables.

7

Because bandwidths vary by outcome, estimates for specific causes of death do not add up to the estimate for

total deaths. Estimates from a specification that uses a constant bandwidth of 24 months are available in the online

Appendix (Table A.12). When using that specification, subcategory estimates do add up to the total estimate.

530 AER: INSIGHTS DECEMBER 2021

Panel A. Proportion with a driver’s license Panel B. Average annual vehicle miles driven

0.6

2,500

2,000

0.4

1,500

0.2 1,000

500

0 0

−12−10 −8 −6 −4 −2 0 2 4 6 8 10 12 −12−10 −8 −6 −4 −2 0 2 4 6 8 10 12

Age (in months) since MDA Age (in months) since MDA

Panel C. Deaths from all causes Panel D. Motor vehicle fatalities

30

80

Deaths per 100,000

Deaths per 100,000

25

65

20

50

15

35 10

20 5

−12−10 −8 −6 −4 −2 0 2 4 6 8 10 12 −12−10 −8 −6 −4 −2 0 2 4 6 8 10 12

Age (in months) since MDA Age (in months) since MDA

Males Females

Figure 1. Teenage Driving and Mortality

Notes: The figure shows the proportion of teenagers with a driver’s license, average annual vehicle miles driven

(baseline specification), a­ ll-cause death rates, and motor vehicle accident death rates by age relative to the MDA.

Estimates in panels A and B are based on weighted responses to the 1995–1996 Add Health surveys. Estimates in

panels C and D are based on death counts from the 1983–2014 National Vital Statistics and population data from

the 1983–2014 SEER Program. The fitted lines are estimated using equation (1) with a bandwidth of 24 months.

Table 1 provides RD estimates.

at the age cutoff, an increase of 15 percent relative to a mean of 38.9 deaths per

100,000. The estimated effect on deaths from internal causes is small and statisti-

cally insignificant. By contrast, the estimated effect on deaths from external causes

is 5.20 deaths per 100,000 (19 percent) and remains marginally significant after con-

servatively accounting for multiple inference (­family-wise ​p = 0.0523​). Scaling

the a­ ll-cause mortality estimate by the driver’s license estimate reported in panel A

implies that driving increases total mortality at the cutoff by 31.4 deaths per 100,000

(81 percent) among teenagers who receive a license.

Columns 4 and 6 of Table 1 report RD estimates separately for males and females.

Comparing column 3 to column 5 shows that—for every cause of death—death

rates are higher for males than females in the year prior to reaching the MDA. Thus,

while the absolute increase in a­ ll-cause mortality at the age cutoff is about the same

for both males and females, the relative increase for females (22 percent) is double

the relative increase for males (11 percent).

Table 1 also decomposes deaths due to external causes into four main subcatego-

ries: motor vehicle accidents, suicides and accidents, homicides, and other. Column 2

VOL. 3 NO. 4 HUH AND REIF: TEENAGE DRIVING, MORTALITY, AND RISKY BEHAVIORS 531

Table 1—Effect of Driving Eligibility on Teenage Driving and Mortality

Full sample Male Female

Outcome variable Mean RD Mean RD Mean RD

(1) (2) (3) (4) (5) (6)

Panel A. Driving

Has driver’s license 0.0130 0.186 0.0163 0.193 0.0101 0.179

[0.124, 0.231] [0.139, 0.231] [0.103, 0.232]

Miles driven (miles/yr) (baseline) 514 375 569 486 458 234

[159, 530] [195, 734] [−105, 479]

Miles driven (miles/yr) (alternate) 549 575 613 753 484 327

[231, 856] [328, 1,194] [−144, 676]

Panel B. Mortality

All causes 38.9 5.84 50.6 5.72 26.7 5.76

[1.99, 9.36] [−0.809, 11.3] [4.35, 7.53]

{0.0252} {0.643} {​<​0.0001}

Internal causes 12.2 0.406 13.8 −0.0589 10.5 0.820

[−0.120, 1.17] [−0.979, 1.03] [−0.0420, 2.00]

{0.560} {1.00} {0.554}

External causes 26.7 5.20 36.8 5.56 16.1 4.82

[1.42, 8.47] [0.0377, 10.3] [2.81, 6.66]

{0.0523} {0.524} {​<​0.0001}

Motor vehicle accident 11.2 4.92 13.6 5.67 8.75 4.46

[2.36, 7.07] [2.76, 8.10] [2.41, 6.14]

{​<​0.001} {​<​0.01} {​<​0.001}

Suicide and accident 10.5 0.167 15.6 −0.0506 5.07 0.337

[−0.680, 0.924] [−1.63, 1.22] [−0.0259, 0.849]

{0.998} {1.00} {0.555}

Firearm 3.64 0.0914 5.87 0.529 1.29 −0.333

[−0.326, 0.474] [0.0108, 1.04] [−0.715, -0.0560]

{0.998} {0.524} {0.313}

Poisoning 1.08 0.314 1.17 0.133 0.984 0.747

[0.183, 0.522] [−0.218, 0.458] [0.591, 1.07]

{​<​0.001} {0.998} {​<​0.0001}

Drug overdose 0.864 0.315 0.897 0.0447 0.830 0.646

[0.233, 0.496] [−0.242, 0.305] [0.476, 0.999]

{​<​0.0001} {1.00} {​<​0.0001}

Carbon monoxide and other gases 0.214 0.103 0.270 0.0798 0.154 0.127

[−0.0301, 0.215] [−0.149, 0.258] [0.0333, 0.243]

{0.593} {0.998} {0.163}

Drowning 1.53 −0.294 2.64 −0.690 0.367 0.126

[−0.576, −0.0967] [−1.20, -0.352] [−0.00258, 0.270]

{0.0523} {​<​0.01} {0.544}

Other 4.23 0.105 5.93 0.0406 2.43 0.0749

[−0.316, 0.463] [−0.511, 0.512] [−0.519, 0.639]

{0.998} {1.00} {1.00}

Homicide 4.80 −0.0423 7.33 −0.0320 2.14 −0.0779

[−0.623, 0.534] [−1.18, 1.10] [−0.335, 0.154]

{0.998} {1.00} {0.998}

Other external 0.243 0.00608 0.328 −0.0571 0.154 0.143

[−0.148, 0.154] [−0.316, 0.154] [0.0872, 0.247]

{0.998} {0.998} {​<​0.001}

Notes: The driving estimates in panel A are based on weighted responses to the 1995–1996 Add Health surveys. The

mortality estimates in panel B, which are measured in deaths per 100,000 ­person-years, are based on death counts

from the 1983–2014 National Vital Statistics and population data from the 1983−2014 SEER Program. Columns

1, 3, and 5 report means of the dependent variable one year before reaching the MDA. Columns 2, 4, and 6 report

­MSE-optimal estimates of β ​ ​from equation (1). Robust, b­ ias-corrected 95 percent confidence intervals are reported

in brackets. Family-wise ​p​-values, reported in braces, adjust for the number of outcome variables in each family

and for the number of subgroups.

532 AER: INSIGHTS DECEMBER 2021

Panel A. All poisonings

1.8

1.6

Deaths per 100,000

1.4

1.2

1

0.8

−12 −10 −8 −6 −4 −2 0 2 4 6 8 10 12

Age (in months) since MDA

Panel B. Drug overdoses Panel C. Carbon monoxide and other

gas poisonings

1.6 0.4

Deaths per 100,000

1.4

Deaths per 100,000

0.3

1.2

1

0.2

0.8

0.6 0.1

−12 −10 −8 −6 −4 −2 0 2 4 6 8 10 12 −12−10 −8 −6 −4 −2 0 2 4 6 8 10 12

Age (in months) since MDA Age (in months) since MDA

Figure 2. Female Poisoning Deaths, 1983–2014

Notes: The figure shows female death rates for different causes of death by age relative to the MDA. Deaths from all

poisonings, shown in panel A, equal the sum of deaths from drug overdoses and from carbon monoxide and other

gas poisonings. The fitted lines are estimated using equation (1) with a bandwidth of 24 months. Table 1 provides

RD estimates.

reports a significant increase in motor vehicle fatalities of 4.92 deaths per 100,000

(44 percent) at the age cutoff. Columns 4 and 6 report that this increase is significant

for both males and females (­family-wise ​p < 0.01​) and can explain the majority

of the increase in total mortality for both subgroups. As with a­ ll-cause mortality,

the absolute increase in motor vehicle fatalities at the cutoff is similar for males and

females, but the relative increase is larger for females.

VOL. 3 NO. 4 HUH AND REIF: TEENAGE DRIVING, MORTALITY, AND RISKY BEHAVIORS 533

Panel A. All poisonings

3

2.5

Deaths per 100,000

2

1.5

1

0.5

−12 −10 −8 −6 −4 −2 0 2 4 6 8 10 12

Age (in months) since MDA

Panel B. Drug overdoses Panel C. Carbon monoxide and

other gas poisonings

1

2.5

Deaths per 100,000

0.8

Deaths per 100,000

2

0.6

1.5

0.4

1

0.2

0.5

−12−10 −8 −6 −4 −2 0 2 4 6 8 10 12 −12−10 −8 −6 −4 −2 0 2 4 6 8 10 12

Age (in months) since MDA Age (in months) since MDA

Figure 3. Male Poisoning Deaths, 1983–2014

Notes: The figure shows male death rates for different causes of death by age relative to the MDA. Deaths from all

poisonings, shown in panel A, equal the sum of deaths from drug overdoses and from carbon monoxide and other

gas poisonings. The fitted lines are estimated using equation (1) with a bandwidth of 24 months. Table 1 provides

RD estimates.

Column 2 of Table 1 also reports a significant increase in poisoning deaths

of 0.314 per 100,000 (29 percent). This poisoning effect can be further decom-

posed into a 0.315 per 100,000 (36 percent) increase in drug overdose deaths and

a 0.103 per 100,000 (48 percent) increase in gas poisoning deaths.8 Comparing the

8

Nearly 90 percent of all gas poisoning deaths are caused by carbon monoxide, an odorless gas produced by

motor vehicle exhaust (based on authors’ calculations using the 1983–1998 National Vital Statistics).

534 AER: INSIGHTS DECEMBER 2021

e­ stimates in column 6 to those in column 4 reveals that this effect is driven by female

poisoning deaths, which increase by 0.747 per 100,000 (76 percent) at the cutoff

(­family-wise ​p < 0.0001​). This increase can be attributed primarily to a 0.646 per

100,000 (78 percent) increase in drug overdose deaths (­family-wise ​p < 0.0001​).

Gas poisoning deaths rise by 0.127 per 100,000 (82 percent) at the cutoff, although

this effect is not statistically significant after accounting for multiple hypothesis

testing (­family-wise ​p = 0.163​).

Column 4 of Table 1 reports that the estimate for male poisoning deaths is small

and statistically insignificant and that its 95 percent confidence interval rules out the

corresponding female estimate. We do estimate a statistically significant decrease

in male drownings of 0.690 per 100,000 (26 percent) (online Appendix Figure

A.3a). Unlike the change in female poisoning deaths shown in Figure 2, panel A,

this change in male drownings is s­ hort lived. Similarly, while column 6 of Table 1

reports a statistically significant increase in female deaths due to “other external”

causes, the RD plot does not provide compelling evidence of an effect (online

Appendix Figure A.4b).

Finally, we assess how our estimates change over time by estimating our model

for different ­four-year bins. Figure 4, panels A and C reveal a steady decline in

our estimated effects for male and female motor vehicle fatalities beginning in the

­mid-1990s. Likewise, Figure 4, panel D exhibits a declining trend in the estimates

for female poisoning deaths. By contrast, Figure 4, panel B shows that the effects

of driving eligibility on male poisoning deaths remain centered around zero for the

majority of our sample period.

We cannot directly investigate the cause of the declines shown in Figure 4, pan-

els A, C, and D because our driving behavior data are available only for 1995 and

1996. That said, these declines coincide with a decline in teenage driving documented

in national statistics. For example, license counts published by the Federal Highway

Administration show a steady decline in the proportion of US teenagers with a driv-

er’s license over our sample period (online Appendix Figure B.1), and McGuckin

and Fucci (2018) report a decline in average vehicle miles traveled among licensed

teenagers beginning in 1990 (online Appendix Figure B.2). In addition, prior studies

of GDL programs—which were introduced beginning in 1996—suggest that these

programs successfully reduced teenage motor vehicle fatalities by limiting, rather than

improving, teenage driving (­Karaca-Mandic and Ridgeway 2010, Gilpin 2019).

C. Extensions

Our RD estimates account neither for accidents with multiple fatalities nor for peo-

ple killed by a teenage driver who survives the accident. According to data p­ ublished

by the Fatality Analysis Reporting System (FARS), on average an additional 0.24

people died for every car accident where a newly eligible teenage driver died at the

wheel. In addition, among all fatal car accidents involving a newly eligible teenage

driver at the wheel, the accidents where that teenage driver died accounted for only

45 percent of the total fatalities.9 If we assume these statistics apply proportionally

9

These statistics are calculated using Fatality Analysis Reporting System data published for the years 1983–

2014. Those data report only integer ages.

VOL. 3 NO. 4 HUH AND REIF: TEENAGE DRIVING, MORTALITY, AND RISKY BEHAVIORS 535

Panel A. Motor vehicle fatalities (males) Panel B. Poisoning deaths (males)

15 2

Deaths per 100,000

Deaths per 100,000

1

10

0

5

−1

0 −2

1983 1987 1991 1995 1999 2003 2007 2011 1983 1987 1991 1995 1999 2003 2007 2011

Year Year

Male estimate 95% confidence interval

Panel C. Motor vehicle fatalities (females) Panel D. Poisoning deaths (females)

10 2

Deaths per 100,000

Deaths per 100,000

1.5

5

1

0.5

0

0

−0.5

−5

1983 1987 1991 1995 1999 2003 2007 2011 1983 1987 1991 1995 1999 2003 2007 2011

Year Year

Female estimate 95% confidence interval

Figure 4. Trends in the Effect of Driving Eligibility on Motor Vehicle Fatalities and Poisoning Deaths

Notes: The figure plots ­MSE-optimal estimates of β​ ​from equation (1) separately for f­ our-year bins. The dependent

variable is deaths per 100,000 p­ erson-years. The dashed lines report robust b­ ias-corrected 95 percent confidence

intervals. Table 1 provides estimates for outcomes measured over the whole 1983–2014 sample period.

to our RD estimate, then they imply that total motor vehicle fatalities increase by

13.6​ ​(​​ = 4.92 × 1.24/0.45)​​deaths per 100,000 at the age cutoff.

In the online Appendix, we decompose the female poisoning death estimates by

estimating our model separately for deaths classified as suicides versus accidents

(Table A.1). A few results stand out in this exploratory analysis. First, the increase

in female drug overdose deaths is caused by both accidents and suicides. Second,

although most gas poisoning deaths among female teenagers are accidental in the

aggregate, the increase in these deaths at the age cutoff is driven by suicides. Finally,

the net effect of driving eligibility on total suicides is small and statistically insignif-

icant because of an offsetting reduction in firearm suicides, suggesting that female

teenagers who die by suicide substitute away from using firearms and toward using

drugs and gas poisoning upon gaining access to a car.

We caution that the reliability of the suicide/accident classification is unclear.

For example, some of the increase in accidental drug overdose deaths may in fact

be suicides. We also lack statistical power to discern with confidence whether the

increase in poisoning suicides reflects a net increase in suicide or substitution away

from other methods of suicide, making it difficult to test different theories of youth

suicide (Cutler, Glaeser, and Norberg 2001). Overall, we conclude that the rise in

536 AER: INSIGHTS DECEMBER 2021

female poisoning deaths represents changes in both accidental deaths and suicides

and that the effects on suicide might reflect a compositional shift in the method of

suicide.

D. Robustness

The most common MDA in our sample is 16 years, which also happens to be the

federal minimum legal working age as well as the minimum legal school-leaving

age in many states. We do not believe these other laws confound our estimates,

however. Analysis of the Add Health data shows small and statistically insignificant

changes in working for pay or leaving school at the MDA cutoff (online Appendix

Figure A.5). Moreover, our results are similar when we limit our analysis to states

with MDAs that are not 16 years (online Appendix Tables A.2 and A.3).10 Because

the MDA differs from the minimum legal ages for both working and leaving school

in this subsample (online Appendix B.5), we conclude that the MDA is the causal

mechanism underlying our results.

Our analysis examines a large number of outcomes across two different sub-

groups. Although we adjust for multiple inference, our outcomes and subgroups

were not specified prior to analysis. However, we emphasize that our most surpris-

ing result—the increase in female poisoning deaths illustrated in Figure 2—is far

too large to be spurious. A multiple testing correction would need to adjust for many

thousands of hypotheses to increase the unadjusted ​p​-value (​p < 0.00001​) above

the conventional significance level of 0.05.

Our online Appendix reports our main results separately by race, sex, and their

pairwise combinations (Tables A.4–A.7). Whites are more likely than non-Whites

to obtain a driver’s license upon becoming eligible, consistent with prior studies

(Shults and Williams 2013). This differential driving effect is reflected in the mortal-

ity estimates: motor vehicle fatalities increase the most at the cutoff for White males

and White females, and poisoning deaths increase the most for White females. We

also detect some modest seasonality in our estimate for motor vehicle fatalities, with

effect sizes peaking during the summer months in both absolute and relative terms

(Figure A.6).

Our estimates are not sensitive to using different bandwidth selection procedures

or polynomial approximations or to imposing a uniform bandwidth of 24 months

(online Appendix Tables A.10–A.12). Estimating our model using placebo cutoffs

generates estimates that are centered around zero, and our motor vehicle fatality and

female poisoning death estimates lie in the far tails of those placebo distributions

(online Appendix Figure A.8).

V. Conclusion

Employing a regression discontinuity design, we estimate that driving eligibility

increases teenage mortality at the MDA cutoff by 5.84 deaths per 100,000 over the

10

For example, female poisoning deaths rise by 0.509 deaths per 100,000 at the cutoff in states where the MDA

is not 16. This estimate is statistically significant, and its 95 percent confidence interval includes the full sample

estimate of 0.747 deaths per 100,000 (Table 1).

VOL. 3 NO. 4 HUH AND REIF: TEENAGE DRIVING, MORTALITY, AND RISKY BEHAVIORS 537

1983–2014 time period. The relationship between this threshold estimate and changes

in the MDA is ambiguous. For example, raising the MDA could reduce the effect of

eligibility on mortality if older teenagers are more careful drivers or increase the effect

if older teenagers are more likely to drink and drive. If we assume the effect remains

constant, then our estimate implies that a ­one-year increase in the MDA would have

saved 228 teenage lives annually during our sample period.11 This estimate, which

does not account for the additional people killed in motor vehicle accidents involving

newly eligible teenage drivers, is similar in magnitude to the estimated benefits of

raising the minimum legal drinking age (Carpenter and Dobkin 2009).

We also estimate that female poisoning deaths increase by 76 percent following

driving eligibility, but we find no effect on male poisoning deaths. These findings

reveal that teenage driving is an important causal factor behind sex disparities in

teenage poisoning deaths. While the increase in female poisoning deaths reflects

changes in both accidents and suicides, the specific behavioral mechanisms underly-

ing our results remain unclear. Driving may enable female teenagers to purchase or

consume drugs more easily and may affect mental health by altering social environ-

ments. We encourage future researchers to investigate these different possibilities.

REFERENCES

Add Health. 1995–1996. National Longitudinal Study of Adolescent to Adult Health [Restricted-Use

Version]. https://addhealth.cpc.unc.edu.

Alexander, Diane, and Molly Schnell. 2019. “Just What the Nurse Practitioner Ordered: Independent

Prescriptive Authority and Population Mental Health.” Journal of Health Economics 66: 145–62.

Anderson, D. Mark. 2010. “Does Information Matter? The Effect of the Meth Project on Meth Use

among Youths.” Journal of Health Economics 29 (5): 732–42.

Calonico, Sebastian, Matias D. Cattaneo, and Rocio Titiunik. 2014. “Robust Nonparametric Confi-

dence Intervals for Regression-Discontinuity Designs.” Econometrica 82 (6): 2295–2326.

Carpenter, Christopher. 2004. “How Do Zero Tolerance Drunk Driving Laws Work?” Journal of

Health Economics 23 (1): 61–83.

Carpenter, Christopher S., Tim A. Bruckner, Thurston Domina, Julie Gerlinger, and Sara Wakefield.

2019. “Effects of State Education Requirements for Substance Use Prevention.” Health Econom-

ics 28 (1): 78–86.

Carpenter, Christopher, and Carlos Dobkin. 2009. “The Effect of Alcohol Consumption on Mortality:

Regression Discontinuity Evidence from the Minimum Drinking Age.” American Economic Jour-

nal: Applied Economics 1 (1): 164–82.

Case, Anne, and Angus Deaton. 2015. “Rising Morbidity and Mortality in Midlife among White

Non-Hispanic Americans in the 21st Century.” Proceedings of the National Academy of Sciences of

the United States of America 112 (49): 15078–83.

Case, Anne, and Angus Deaton. 2017. “Mortality and Morbidity in the 21st Century.” Brookings Papers

on Economic Activity 46 (1): 397–476.

Cawley, John, Sara Markowitz, and John Tauras. 2004. “Lighting Up and Slimming Down: The

Effects of Body Weight and Cigarette Prices on Adolescent Smoking Initiation.” Journal of Health

Economics 23 (2): 293–311.

Centers for Disease Control and Prevention. 2018. “WISQARS (Web-Based Injury Statistics Query

and Reporting System).” https://www.cdc.gov/injury/wisqars/index.html.

Chapman, Eric A., Scott V. Masten, and Kelly K. Browning. 2014. “Crash and Traffic Violation Rates

before and after Licensure for Novice California Drivers Subject to Different Driver Licensing

Requirements.” Journal of Safety Research 50: 125–38.

Cotto, Jessica H., Elisabeth Davis, Gayathri J. Dowling, Jennifer C. Elcano, Anna B. Staton, and

Susan R.B. Weiss. 2010. “Gender Effects on Drug Use, Abuse, and Dependence: A Special Analy-

sis of Results from the National Survey on Drug Use and Health.” Gender Medicine 7 (5): ­402–13.

11

There was an average of 3.9 million ­16-year-olds alive in the United States during 1983–2014.

538 AER: INSIGHTS DECEMBER 2021

Curry, Allison E., Melissa R. Pfeiffer, Dennis R. Durbin, and Michael R. Elliott. 2015. “Young Driver

Crash Rates by Licensing Age, Driving Experience, and License Phase.” Accident Analysis and

Prevention 80: 243–50.

Cutler, David, Edward Glaeser, and Karen Norberg. 2001. “Explaining the Rise in Youth Suicide.” In

Risky Behavior among Youths: An Economic Analysis. NBER Chapters, edited by Jonathan Gruber,

­219–70. Chicago: University of Chicago Press.

Dee, Thomas S., and William N. Evans. 2001. “Teens and Traffic Safety.” In Risky Behavior among

Youths: An Economic Analysis, edited by Jonathan Gruber, 121–66. Chicago: University of Chi-

cago Press.

Dee, Thomas S., David C. Grabowski, and Michael A. Morrisey. 2005. “Graduated Driver Licensing

and Teen Traffic Fatalities.” Journal of Health Economics 24 (3): 571–89.

Deza, Monica, and Daniel Litwok. 2016. “Do Nighttime Driving Restrictions Reduce Criminal Partic-

ipation among Teenagers? Evidence from Graduated Driver Licensing.” Journal of Policy Analysis

and Management 35 (2): 306–32.

Dobkin, Carlos, and Steven L. Puller. 2007. “The Effects of Government Transfers on Monthly Cycles

in Drug Abuse, Hospitalization and Mortality.” Journal of Public Economics 91 (11–12): ­2137–57.

Dong, Yingying. 2015. “Regression Discontinuity Applications with Rounding Errors in the Running

Variable.” Journal of Applied Econometrics 30 (3): 422–46.

Evans, William N., and Timothy J. Moore. 2012. “Liquidity, Economic Activity, and Mortality.”

Review of Economics and Statistics 94 (2): 400–418.

Federal Highway Administration. 1992–2002. “Traffic Volume Trends 1992–2002.” US Department of

Transportation. https://www.fhwa.dot.gov/ohim/tvtw/archive/index.htm.

Federal Highway Administration. 1997. “Highway Statistics Series: Distribution of Licensed

­Drivers—1996 by Sex and Percentile in Each Age Group and Relation to Population.” US Depart-

ment of Transportation. https://www.fhwa.dot.gov/policyinformation/statistics/1996/dl20.cfm.

Foss, Robert D., Carol A. Martell, Arthur H. Goodwin, and Natalie P. O’Brien. 2011. Measuring

Changes in Teenage Driver Crash Characteristics during the Early Months of Driving. Washing-

ton, DC: AAA Foundation for Traffic Safety.

Gilpin, Gregory. 2019. “Teen Driver Licensure Provisions, Licensing, and Vehicular Fatalities.” Jour-

nal of Health Economics 66: 54–70.

Glied, Sherry. 2002. “Youth Tobacco Control: Reconciling Theory and Empirical Evidence.” Journal

of Health Economics 21 (1): 117–35.

Gruber, Jonathan. 2001. “Introduction to ‘Risky Behavior among Youths: An Economic Analysis.’” In

Risky Behavior among Youths: An Economic Analysis, edited by Jonathan Gruber, 1–28. Chicago:

University of Chicago Press.

HeinOnline. 1983–1994. Session Laws Library. https://home.heinonline.org/content/session-laws-

library (accessed July 30, 2020).

Heslin, Kevin C., and Anne Elixhauser. 2016. “Mental and Substance Use Disorders among Hospital-

ized Teenagers, 2012.” Healthcare Cost and Utilization Project Statistical Brief 202.

Huh, Jason, and Julian Reif. 2021. “Replication Data for: Teenage Driving, Mortality, and Risky

Behaviors.” American Economic Association [publisher], Inter-university Consortium for Political

and Social Research [distributor]. https://doi.org/10.3886/E133501V1.

Insurance Institute for Highway Safety. 1995–2014. “Effective Dates of Graduated Licensing Laws.”

http://www.iihs.org/iihs/topics/laws/graduatedlicenseintro?topicName=teenagers (accessed Octo-

ber 8, 2015).

Jones, Damon, David Molitor, and Julian Reif. 2019. “What Do Workplace Wellness Programs

Do? Evidence from the Illinois Workplace Wellness Study.” Quarterly Journal of Economics

134 (4): 1747–91.

Karaca-Mandic, Pinar, and Greg Ridgeway. 2010. “Behavioral Impact of Graduated Driver Licensing

on Teenage Driving Risk and Exposure.” Journal of Health Economics 29 (1): 48–61.

Lyons, Rachael M., Amy M. Yule, Davida Schiff, Sarah M. Bagley, and Timothy E. Wilens. 2019. “Risk

Factors for Drug Overdose in Young People: A Systematic Review of the Literature.” Journal of

Child and Adolescent Psychopharmacology 29 (7): 487–97.

McGuckin, N., and A. Fucci. 2018. Summary of Travel Trends: 2017 National Household Travel Sur-

vey. Washington, DC: Federal Highway Administration.

Morrisey, Michael A., David C. Grabowski, Thomas S. Dee, and Christine Campbell. 2006. “The

Strength of Graduated Drivers License Programs and Fatalities among Teen Drivers and Passen-

gers.” Accident Analysis & Prevention 38 (1): 135–41.

National Cancer Institute. 1983–2014. Surveillance, Epidemiology, and End Results (SEER) Program

Populations. Surveillance Research Program, Division of Cancer Control and Population Sciences.

https://seer.cancer.gov/popdata (accessed June 28, 2018).

VOL. 3 NO. 4 HUH AND REIF: TEENAGE DRIVING, MORTALITY, AND RISKY BEHAVIORS 539

National Center for Health Statistics. 1983–2014. “Mortality Data [Restricted-Use Version].” National

Vital Statistics System. https://www.cdc.gov/nchs/nvss/deaths.htm.

National Institute on Drug Abuse. 2019. “Drug Overdoses in Youth.” https://teens.drugabuse.gov/

drug-facts/drug-overdoses-youth.

Phillips, David P., Nicholas Christenfeld, and Natalie M. Ryan. 1999. “An Increase in the Number of

Deaths in the United States in the First Week of the Month—An Association with Substance Abuse

and Other Causes of Death.” New England Journal of Medicine 341 (2): 93–98.

Riddell, Chris, and Rosemarie Riddell. 2006. Welfare Checks, Drug Consumption, and Health: Evi-

dence from Vancouver Injection Drug Users.” Journal of Human Resources 41 (1): 138–61.

Ruch, Donna A., Arielle H. Sheftall, Paige Schlagbaum, Joseph Rausch, John V. Campo, and Jef-

frey A. Bridge. 2019. “Trends in Suicide among Youth Aged 10 to 19 Years in the United States,

1975 to 2016.” JAMA Network Open 2 (5): e193886.

Shults, Ruth A., and Allan F. Williams. 2013. “Trends in Driver Licensing Status and Driving among

High School Seniors in the United States, 1996–2010.” Journal of Safety Research 46: 167–70.

Swahn, Monica H., and Robert M. Bossarte. 2007. “Gender, Early Alcohol Use, and Suicide Ideation

and Attempts: Findings from the 2005 Youth Risk Behavior Survey.” Journal of Adolescent Health

41 (2): 175–81.

This article has been cited by:

1. Masoud Mohammad khanlou, Saeedeh Bazazian, Hamid Reza Vatankhah. 2022. The Effectiveness

of Cognitive-Behavioral Training on Emotion Regulation and High-Risk Behaviors in High School

Students. Iranian Evolutionary and Educational Psychology 4:2, 388-39⒎ [Crossref]