Time to Licensure for Driving Among U.S. Teens: Survival Analysis of Interval-Censored Survey Data

Abstract

Objective:

Novice drivers who delay in driving licensure may miss safety benefits of Graduate Driver Licensing (GDL) programs, potentially putting themselves at higher crash-risk. Time to licensure relates their access to independent transportation to potential future economic- and educational-related opportunities. The objective of this study was to explore time to licensure associations with teens’ race/ethnicity and GDL restrictions.

Methods:

Secondary analysis using all seven annual assessments of the NEXT Generation Health Study, a nationally representative longitudinal study starting with 10^th grade (N=2785; 2009–2010 school year). Data were collected in U.S. public/private schools, colleges, workplaces, and other settings. The outcome variable was interval-censored time to licensure (event = obtained driving licensure). Independent variables included race/ethnicity and state-specific GDL restrictions. Covariates included family affluence, parent education, nativity, sex, and urbanicity. Proportional hazards (PH) models were conducted for interval-censored survival analysis based on stepwise backward elimination for fitting multivariate models with consideration of complex survey features. In the PH models, a hazard ratio (HR) estimates a greater (>1) or lesser (<1) likelihood of licensure at all timepoints.

Results:

Median time to licensure after reaching legal driving age for Latinos, African Americans, and Non-Latino Whites was 3.47, 2.90, and 0.41 years respectively. Multivariate PH models showed that Latinos were 46% less likely (HR=0.54, 95%CI: 0.35–0.72) and African Americans were 56% less likely (HR=0.44, 95%CI: 0.32–0.56) to have obtained licensure at any time compared to Non-Latino Whites. Only learner minimum age GDL restriction was associated with time to licensure. Living in a state with a required learner driving minimum age of ≥16 years (HR=0.57, 95%CI: 0.16–0.98) also corresponded with 43% lower likelihood of licensure at legal eligibility compared to living in other states with a required learner driving minimum age of <16 years.

Conclusion:

Latinos and African American teens obtained their license approximately three years after eligibility on average, and much later than Non-Latino Whites. Time to licensure likelihood was associated with race/ethnicity and required minimum age of learner permit, indicating important implications for teens of different racial/ethnic groups in relation to licensure, access to independent transportation, and exposure to GDL programs.

INTRODUCTION

Teens in the U.S. are at a high risk for serious nonfatal injury, disability, and death caused by motor vehicle crashes and motor-vehicle crashes are the leading cause of unintentional injury death for male and female youth aged 16 to 20 years (Webb, 2018, February).

To reduce the overall risk of motor vehicle crashes among teens, Graduated Driver Licensing (GDL) programs were put in place to allow novice teen drivers to safely gain driving experience before obtaining full independent driving privileges. Previous studies have indicated that comprehensive GDL programs are significantly associated with reducing fatal crash involvement by 20% to 40% among teen drivers (Shope, 2007). Beginning in 1996, the adoption of GDL policies in all states began with the incorporation of additional novice driver requirements and restrictions during the learning period (Williams et al., 2016). In GDL policy practice, teen drivers are generally supervised by experienced licensed drivers, restricted in driving certain numbers of young passengers, mandated to log a policy-defined number of practice hours of driving, and forbidden to drive during late nighttime hours (Masten et al., 2011). Those teen drivers participating in GDL programs hold learner’s permits when they begin driving, gradually progress to restricted driving licensure, and eventually receive a fully-independent driving licensure at the end of the GDL program (Hedlund, 2007).

Across the U.S., the minimum age to obtain a driver permit can range from 14 to 16 years old, a restricted license from 14 to 17 years old, and a full license typically from 16.5 to 18 years old (Williams et al., 2012). Despite the well-known safety benefits of GDL programs, many teens choose not to obtain their license when they reached the legal driving age; instead they delay in driving licensure (DDL). The rate of teens that delay in driving licensure (DDL) has increased, which could be due to GDL or to other national trends such as increases in ride sharing and social media. Between 1996 and 2010, the percentage of U.S. high school seniors with a driver’s license decreased from 83% to 73% (Shults & Williams, 2013). Since the 2008 great recession, the proportions of teens with a driver’s license decreased at different age groups from 2008 (ages 17, 18, and 19: 50.0%, 65.4% and 75.5%), to 2011 (ages 17, 18, and 19: 45.0%, 60.3% and 69.3%) and 2014 (ages 17, 18, and 19: 44.9%, 60.1% and 69.0%) (Sivak & Schoettle, 2016). This trend is of interest as GDL programs have been established to decrease overall crash risk in novice drivers, but most of these programs only extend to drivers aged 18 years and younger (Curry et al., 2018). Consequently, those licensed at later ages might miss out on safety benefits of GDL. As of 2017, only seven states required some GDL restrictions for novice drivers ≥ 18 years old (Curry et al., 2017). Currently, no studies have been found to conduct the comparison in states with GDL restrictions for older novices (e.g., 18 years or older) versus those states without GDL restrictions for older novices. While DDL may lead to less crash risk due to the reduced overall and age-related high-risk driving exposure, the initial reduction in crash risk may be reversed later when these youth start driving and have insufficient driving exposure, experience, and instruction (Tefft et al., 2013). For instance, if teens do not become licensed before the age of 18, they might miss important graduated driving safety practices that are intentionally designed to reduce crashes among teen novice drivers. Furthermore, there are also potential benefits from licensure beyond exposure to GDL that could be diminished by DDL. Mobility enabled by driving licensure can improve economic and educational opportunities (Fransen et al., 2018). Therefore, disparities in DDL are a concern.

Possible reasons for DDL have been examined, and some demographic and socioeconomic variables are associated with the timing of driving licensure. For example, African American and Latino youth DDL at higher rates than their Non-Latino White counterparts (Shults & Williams, 2013; Tefft et al., 2014). Furthermore, low income teens are more likely to DDL compared to high income teens (Thigpen & Handy, 2018).

Currently, there is conflicting evidence on how GDL programs may be associated with delays in obtaining driving licensure (Wang et al., 2020). A retrospective study indicated that GDL policies increase the likelihood of DDL (Thigpen & Handy, 2018). However, other studies did not find substantial association between GDL and DDL among teens (Tefft et al., 2014). Finally, while very recent research has shed additional light on factors that contribute to DDL, the knowledge base in this area remains incomplete in need of further investigation (Vaca et al., 2020).

Given limited and some mixed findings, this study aimed to evaluate the longitudinal associations of state-level GDL restrictions with time to licensure among U.S. high school students using survival analysis and to assess the interaction between state-level GDL restrictions and race/ethnicity on time to licensure.

METHODS

Sampling

Data used were from all seven waves of the NEXT Generation Health Study (NEXT). This longitudinal study began in the 2009–2010 school year and followed a nationally representative cohort of U.S. 10^th grade students ((16.3 y/o (se=0.03)) using a multistage sampling design stratified by nine census divisions. The data collection and sampling strategy have been previously published (Li et al., 2013). A total of 2785 students from 81 high schools participated in the NEXT study. African American participants were oversampled to provide a better population estimate as well as an adequate sample to examine racial/ethnic differences. The study protocol was reviewed and approved by the Institutional Review Board of the Eunice Kennedy Shriver National Institute of Child Health and Human Development.

Outcome Variable

Time to licensure:

Participant’s state of residence and their date of birth were used to determine when they reached the legal driving age, the minimum age when a participant was eligible to obtain his/her independent, unsupervised driver’s license based on their state law. For participants (0.4% of the sample) who did not provide any state information, it was assumed they reached the legal driving age on their 16^th birthday. Date of birth, driving licensure status at each wave, and survey administration date at each wave were used to determine when licensure was obtained.

Time to licensure was defined as the length of time that passed between when an individual reached legal driving age and when they obtained a license (licensure time). In our study, the licensure time T of an individual was observed to lie in an interval of survey administration dates. For example, if a participant responded that they had a permit at W2 and then a license at W3, we concluded that they obtained a license between their survey administration dates for W2 and W3. The interval was further refined if the participant reached the minimum legal driving age between these two administration dates because the exact date of licensure was unknown.

Independent Variables

State-specific GDL restrictions:

GDL laws effective during the NEXT study (from 2009 to 2016) for the learner’s and intermediate phases for participant’s state were coded to be consistent with methods by McCartt et al. (McCartt et al., 2010), such that zero points indicate no or low GDL restrictions and one or more points indicate higher restrictions. The variables derived were: 1) minimum permit age (0 points=less than 16 years, 1 point=16 years or older); 2) permit holding period (0 points=less than 6 months, 1 point=6 or more months); 3) permit required practice driving hours (0 points=less than 30 hours, 1 point=30 or more hours); 4) limits on nighttime driving points (0 points=no restriction, 1 point=after 10 p.m., 2 points=10 p.m. or earlier); and 5) limits on young passenger(s) allowed - age of 18 to 25 or younger, depending on state (0 points=3 or more passengers or no restriction, 1 point=2 passengers, 2 points=zero or 1 passenger).

Demographic Variables

Race/ethnicity was one primary independent variable in the analyses. Other demographic variables were covariates including sex (females vs. males), family affluence (high, moderate vs. low), nativity (born in the U.S. and born elsewhere), parent education level (some college, bachelor’s degree or higher vs. high school or less), participant post-high school education status (attending vs. not attending school after high school), and urbanicity (suburban, rural, vs. urban derived from participants’ W1 school location based on the seven National Center for Education Statistics categories).

Statistical Analyses

All statistical analyses were performed in R (version 3.6.1). Significance level was set a priori at p<.05 (2-sided) based on 95% confidence intervals. Missing values for demographic covariates were imputed using hot-deck imputation. The patterns of missingness were examined. The percent of imputed missing data for each covariate was 9.0% (partly imputed) for urbanicity, 0.1% for family affluence, 9.4% for family structure, 0.2% for sex, 1.1% for US nativity, 9.2% for parental education, 10.8% for post-high school education, 2.4% for social media use, and 0.4% GDL laws. After confirming the high school name and location for these participants, 204 of the 251 missing values were considered “urban”. These 204 missing values were thus changed to urban while the remaining 47 missing urbanicity values from other PSUs were imputed. The most common missing value was post-high school education, but this was missing at random (MAR). Of the individuals missing a value for family structure and parental education, the majority of them were from the same PSU and the patterns do not appear MAR. However, considering that family structure and parental education cannot be inferred from location, they were imputed with hot-deck imputation. The remaining variables were missing only for a very small number of subjects. As a result, we assumed that use of hot-deck imputation here would not be problematic.

To address differences in time to licensure by race/ethnicity, the non-parametric maximum likelihood estimator (NPMLE) of the survival distribution was obtained for each racial/ethnic sub-group (Gentleman & Vandal, 2012; R Core Team, 2014). To determine whether time to licensure was longer for Latinos than Non-Latino Whites and African Americans, the NPMLE survival curves for Latinos, African Americans, and Non-Latino Whites were compared with an interval-censored generalization of the classical log-rank test, i.e., weighted log-rank tests (WLRT) and test statistic, G², of the WLRT (Bogaerts et al., 2017). The data are interval-censored because the date when the driver’s license was obtained was not known. Rather, at each wave (i.e., annual participant assessment) participants reported whether or not they had a driver’s license. For participants that did not obtain a license before Wave 7 (the last wave of the study), the time to licensure is right censored. Estimated median survival times for time to licensure were determined by finding the region of support/Turnbull interval in which $\widehat{S}\left(t\right)=0.5$. We identified point estimates for the median survival times for time to licensure by assuming that the NPMLE is linear in the regions of support.

We used semi-parametric Proportional Hazard (PH) models (Bogaerts et al., 2017) to test bivariate associations of time to licensure with GDL law restrictions, race/ethnicity, and other demographic covariates. After that, a multivariate model was fitted including all the variables (i.e., demographic covariates and GDL law restrictions) that had significant associations in the bivariate models. The multivariate model can be used to assess whether these associations are significant when controlling for the effects of other variables. Stepwise backward elimination was used to remove variables until all the remaining variables showed significant associations with time to licensure.

To explore whether the effect of the race/ethnicity and GDL law restrictions on time to licensure was stronger for Latinos than non-Latinos, a multivariate model was fitted that included interaction terms between race/ethnicity (reference=Non-Latino Whites) and GDL law variables that were significant from the previous multivariate model with jackknife confidence intervals for random forests. In the PH models, a hazard ratio (HR) >1 indicates a greater likelihood and < 1 a lower likelihood of having a driving license at any timepoint during the study. The implication of a HR < 1 is a that those in the category have longer times to licensure than those in the referent group.

The features (i.e., census division, primary sampling unit, and survey weights) of the complex survey design were taken into consideration in the analyses. The jackknife repeated replication (JRR) method of variance estimation is applicable to complex survey designs in which there are two or more primary sampling units (PSUs) selected from each stratum. The method consists of four steps recommended by Heeringa et al. (2017).

RESULTS

We excluded participants that identified as “Other” race/ethnicity (N=142) and those (N=27) that had missing information for race/ethnicity, licensure status, or date of birth. Therefore, this study included 2,616 participants, of whom 829 (32%) were Latino, 683 (26%) were African American, and 1,104 (42%) were Non-Latino White. By W7, 81.5% Latinos (N=683), 80.4% African American (N=586), and 92.4% Non-Latino Whites (N=900) reported having an independent, unsupervised driver’s license. As shown in Figure 1, the median time to licensure after reaching the legal driving age was 3.47 years for Latinos (1263.5 days, Turnbull interval=[1236; 1264]), 2.90 years for African Americans (1057 days, Turnbull interval=[1056; 1058]), and 0.41 years for Non-Latino Whites (147.5 days, Turnbull interval=[146; 148]). The NPMLE survival curve for median time to licensure in Non-Latino Whites did not cross the curves for Latinos or African Americans. Instead, it was well underneath the curves for Latinos and African Americans (Figure 1). There were statistically significant differences in the survival curves for median time to licensure for Latinos versus African Americans (G²=6.13, p=.013), Latinos versus Non-Latino Whites (G²=442.0, p=2.2е−16), and African Americans versus Non-Latino Whites (G²=309.1, p=2.2е−16).

Figure 1.

NPMLE Survival Curves for Time to Licensure among Latinos, African Americans and Non-Latino Whites

Note. The straight horizontal dotted line indicates median time to licensure. NPMLE: the non-parametric maximum likelihood estimator.

To test the association between demographic covariates of interest, GDL law restrictions and time to licensure, individual bivariate semiparametric PH models were fit with each demographic covariate or GDL law restrictions as the predictor and licensure status as the response. For each model, the HR and corresponding 95% confidence interval (CI) are provided in Supplementary Table 1 for demographic covariates and in Table 1 for GDL law variables. Crude mean and median time to licensure by each demographic covariate and graduated driver licensing variables are provided in Supplementary Table 2.

Table 1.

Bivariate Models for Associations Between Time to Licensure and Graduated Driver Licensing Variables

Variable	Level	NState	N	Weighted %	Hazard Ratioa	95% CIb
Learner minimum age	Less than 16	17	2110	79.9	Ref
	16 or older	5	415	20.1	0.62	(0.39, 0.85)
Permit hold period	Less than 3 months	17	1885	82.1	Ref
	6 or more months	5	640	17.9	1.11	(0.57, 1.64)
Required learner hours	Less than 30 hours	2	155	11.4	Ref
	30 or more hours	20	2370	88.6	0.72	(0.24, 1.21)
Restriction on night driving	After 10 pm	15	1940	79.8	Ref
	10 pm or earlier	7	585	20.2	0.99	(0.71, 1.28)
Restriction on underage passengers	≥3 passengers or no restriction	2	555	13.5	Ref
	2 passengers	3	206	22.7	0.56	(0.40, 0.73)
	0 or 1 passenger	17	1764	63.8	0.77	(0.57, 0.96)

N_State indicates the # of states and N indicates the # of participants; Weighted % indicate weighted proportion of participants.

Significant associations (p<.05) are indicated in boldface.

Notes:

^a.)A hazard ratios (HR) >1 means increased likelihood of having a license at any given time (less time to licensure, decreased DDL). A HR<1 means decreased likelihood of having a license at any given time (greater time to licensure, increased DDL);

^b)The 95% confidence intervals (CI) were calculated using the jackknife repeated replication method.

As shown in Table 1, of the eligible NEXT sample, the percentage of participants residing in a state with learner phase restrictions was 20.1% (weighted, hereafter; in 5 states) for ≥16 yrs. minimum age, 17.9% for ≥6 month holding period (in 5 states), and 88.6% for ≥30 hours required practice (in 20 states). For intermediate phase restrictions, 20.2% of participants had a night driving limit of no later than 10 PM (in 7 states), and 22.7% (in 3 states) and 63.8% (in 17 states) had limits on young passengers of ≤2 and ≤1, respectively.

Table 2 shows the results of the multivariate PH model examining the associations of having a driving license with race/ethnicity and learner minimum age controlling for other covariates. Latinos and African Americans had a longer time to licensure compared to Non-Latino Whites. Specifically, Latinos were 46% less likely (HR=0.54, 95%CI: 0.35, 0.72) and African Americans were 56% less likely (HR=0.44, 95%CI: 0.32, 0.56) to have a driving license at any timepoint compared to Non-Latino Whites. Living in a state with a minimum age of ≥16 years for a learner’s permit (HR=0.57, 95%CI: 0.16, 0.98) corresponded with 43% lower likelihood of having a driving license compared to living in a state with a minimum age of < 16 years. Youth born outside of the U.S. were 25% less likely (HR=0.75, 95%CI: 0.54, 0.96) to have a driving license compared to those born within the US. Those living in suburban (HR=0.82, 95%CI: 0.66, 0.97) and rural (HR=0.78, 95%CI: 0.57, 0.99) areas were 18% and 22% less likely to have a driving license compared to those living in an urban area, respectively. Youth from moderate (HR=1.44, 95%CI: 1.22, 1.65) and high affluent families (HR=1.83, 95%CI: 1.50, 2.16) were 44% and 83% more likely to have a driving license compared to those from low affluence families, respectively. Those who had a parent with a Bachelor’s degree or higher were 42% more likely (HR=1.42, 95%CI: 1.03, 1.81) to have a driving license compared to those who had a parent with high school or less education. Participants with post-high school education also corresponded with a 62% higher (HR=1.62, 95%CI: 1.31, 1.93) likelihood of having licensure.

Table 2.

Multivariate Model for Associations Between Time to Licensure and Graduated Driver Licensing Variables

Independent Variablesa	Levels	Hazard Ratiob	95% CIc
Race/ethnicity (Ref: Non-Latino White)	Latino	0.54	(0.35, 0.72)
	African American	0.44	(0.32, 0.56)
Learner minimum age (Ref: Less than 16)	16 or older	0.57	(0.16, 0.98)
Family affluence (Ref: Low)	Moderate	1.44	(1.22, 1.65)
	High	1.83	(1.50, 2.16)
U.S. nativity (Ref: U.S. born)	Born outside U.S.	0.75	(0.54, 0.96)
Parent level of education (Ref: High school or less)	Some college	1.23	(0.99, 1.48)
	Bachelor’s degree or higher	1.42	(1.03, 1.81)
Post HS education (Ref: No)	Yes	1.62	(1.31, 1.93)
Urbanicity (Ref: Urban)	Suburban	0.82	(0.66, 0.97)
	Rural	0.78	(0.57, 0.99)

Significant associations (p<.05) are indicated in boldface.

Notes:

^a.)All independent variables were included in the same model to control for each other;

^b.)A hazard ratios (HR) >1 means increased likelihood of having a license at any given time (less time to licensure, decreased DDL). A HR<1 means decreased likelihood of having a license at any given time (greater time to licensure, increased DDL);

^c.)The 95% confidence intervals (CI) were calculated using the jackknife

The interaction between race/ethnicity and the minimum age for a learner’s permit which was significant in the bivariate models was included in the multivariate PH model but was not statistically significant. Therefore, we reported the results only from the multivariate model without including the interaction term.

DISCUSSION

To the best of our knowledge, this study is the first to not only use a survival analysis approach (i.e., interval-censored time to event) to evaluate longitudinal associations of time to licensure and state-level GDL restrictions among a recent cohort of U.S. high school students but to also assess moderation effects of state-level GDL restrictions on the association between race/ethnicity and time to licensure. Study results demonstrate that Latino and African American teens were more likely to have longer times to licensure than Non-Latino White teens. GDL laws with older learner minimum age (i.e., ≥16 years) were significantly associated greater time to licensure for teens, this was not true for some other facets (e. g., supervised practiced driving hours and passengers/nighttime driving restrictions) of the state GDL laws.

In this nationally representative sample study, we have advanced the vital understanding of the associations between DDL and demographic characteristics in U.S. teens. In our study, we found that the estimated time from eligibility to licensure for Latino and African American teens was about 3 years (i.e., more than 2 and half year compared to 0.4 years for non-Latino White teens). This is a considerable delay in independent driving that may result in lost economic and occupational opportunities.

These findings on race and other demographic characteristics in association with delayed licensure are consistent with other research (Shults & Williams, 2013; Tefft et al., 2014; Thigpen & Handy, 2018). We also found that low family affluence, being born outside the U.S., having a parent with a high school education or less, living in a suburban or rural area, and not having education after high school were all associated with decreased likelihood of licensure at any timepoint in the study. Teens who belong to any of those groups may have a greater financial challenge to owning a vehicle and the cost of licensure and associated vehicle costs (e.g., gas, registration, insurance) may further heighten the burden on family finances. Therefore, all the family economic strains together may result in the delay of licensure.

In addition to demographic characteristics and socioeconomic status, we hypothesized that DDL, at least in part, may also be attributed to desires to avoid GDL imposed restrictions by delaying licensure. Our study shows that the GDL laws specifying the learner minimum age by state may be a key factor in time to licensure among teen drivers, whereas other GDL provisions were not associated with time to licensure. However, the association between GDL laws specifying learner minimum age and time to licensure did not vary by race/ethnicity.

We recognize that there are limitations in our study. First, it would have been ideal to know the exact date on which participants obtained their driving license. However, that level of specificity is not available in the NEXT survey data. As a result, we were limited to using intervals with a median length of approximately one year during which an individual obtained their license. Despite this limitation, the study results present compelling evidence for a reasonable and logical estimate of the teen’s licensure date. Second, the school-based recruitment limits the generalization to youth in school at 10^th grade but not those who were not enrolled in school. Third, only a limited number of covariates were collected and analyzed. This may exclude other important factors that could influence DDL (e.g., youth time conflict, parents’ schedules).

The findings of this study have implications for improving teen driver safety. Using this interval-censored survival analysis, time to licensure was found to be significantly associated with teen’s demographics, social structural features, and GDL regulations, which might provide future directions to provide interventions to ensure teen’s driving safety.

These results indicate differences in the likelihood of exposure to GDL programs among Latino and African American youth, given that only seven states impose restrictions on novice drivers over 18 years of age (Curry et al., 2017). Further, indicators of socio-economic status, such as lower parent education and family influence, birth outside the U.S., and lower participant educational attainment also contribute to time to licensure. Given the strong evidence that GDL restrictions have been shown to effectively reduce crashes among teen novice drivers, it is logical to consider that they may be further protective for novice drivers beyond the age of 18 years. Expansion of GDL to older ages could extend driver safety and crash prevention to vulnerable youth regardless of when they obtain their license. However, more research on such expansion is merited.