Data source and study population
Data was obtained from state Youth Risk Behavior Surveys (YRBSs), which are repeated cross-sectional surveys using a two-stage cluster sample design. State YRBSs are anonymous, voluntary surveys conducted every two years to obtain a representative state sample of 9th through 12th grade students attending public school.7-9
The YRBSs sampling design and methodology for combining and analyzing state-level data has been described previously.7-9,32 Only state with a response rate ≥ 60% would be weighted and access to public. States that asked a question about talking on a cellphone while driving in at least one of the years 2013, 2015, 2017, or 2019 surveys were included in this analysis. Participating states are listed in Additional File Table 1. The study inclusion criteria were students who had reached their state’s minimum age to obtain an intermediate license and had driven at least once in the 30 days prior to the survey administration date.
Data on state cellphone laws were obtained from the Insurance Institute for Highway Safety.21 Amendments to the laws and their effective dates were identified using the LexisNexis Academic database and state legislative documents.33 The distribution of enrollment in public elementary and secondary schools by areas were obtained from the National Center for Education Statistics.34 Detailed values of these variables for each state are listed in Additional File Table 2.
The study outcome was self-reported talking on a phone while driving (calling while driving, CWD was used as an abbreviation as opposed to TWD because TWD usually refers to texting while driving). CWD, which was measured using the question: “During the past 30 days, on how many days did you talk on a cell phone while driving a car or other vehicle?” Response options included seven ordinal categories ranging from 0 to 30 days. Students who responded “I did not drive” were excluded from the analysis. Analysis using the original seven ordinal categories is in Additional File Table 3. For the descriptive analysis, we categorized responses into never (0 days), sometimes (1-9 days) and frequent (10-30 days) engagement in CWD. For multivariable analysis, we created a binary outcome (never versus at least once) as any exposure to talking on a phone while driving may increase crash risk for teen drivers. A similar binary categorization was utilized by a previously published study using YRBSs data on texting/emailing while driving.35
The state status of handheld calling bans and young driver bans were classified as 1) no ban (the absence of both handheld calling ban and young driver ban); 2) young driver ban (the absence of a handheld calling ban and an enacted young driver ban ); and 3) concurrent ban (the enactment of both a handheld calling ban and a young driver ban), in which all drivers are not allowed to engage in handheld CWD and young drivers cannot engage in any type of cellphone while driving. No participating states enacted a handheld calling ban and in absence of a young driver ban during the study period. Cellphone law information for each state is listed in Additional File Table 2.
Previous studies have reported that teen driver cellphone use, varies by age, sex, race/ethnicity, and urban/rural status.24,28,35-37 We restricted our main analysis to students who had reached the state-dependent age to begin unsupervised driving under certain driving conditions as driving under the supervision of an adult driver may prohibit teen’s CWD behavior.30,38 For our study, urban/rural status was presented by the precent of students in rural areas, calculated by dividing the number of students enrolled in public elementary and secondary schools from rural areas by the total number of students enrolled in public elementary and secondary schools for each state. We used the enrollment from both elementary and secondary schools as data from only secondary schools is not available.
The association of cellphone laws and CWD was examined by adjusting for student demographics, the percent of students in rural areas, and survey year. None of the YBRS’s participating states changed cellphone law status during the study period, therefore, we estimated the difference of CWD between students of states with varing laws, but not the difference of pre-post law periods within states. Crude and adjusted prevalence ratios (PRs) with 95% confidence intervals (CIs) for CWD were estimated using Poisson regression models with robust variances estimation.39 Further we included interactions between cellphone laws and student demographics (age, sex, and race) to examine the associations between law types and CWD across the following subgroups, age (15/16 vs. ≥17 years), sex (female or male), race/ethnicity (White, Black or African American, Hispanic/Latino and others).
Complete case analysis was used as the percentage of missing data was low (approximately 2% of students reached the minimum age of intermediate license but did not answer the question on CWD). Data were weighted to adjust for school and student nonresponse, the distribution of students by grade, sex and race/ethnicity, and the complex design (strata and psu).9,32 Data analyses were performed in 2020 using SAS Enterprise Guide 7.1 (SAS Institute Inc., Cary, NC) and STATA 14.0 (StataCorp LLC, College Station, TX).
Several sensitivity analyses were conducted to assess potential biases: 1) restricting the analysis to the six states that participated in at least three survey years (Connecticut, Massachusetts, Missouri, Montana, Nebraska, North Dakota); 2) excluding Utah, which enacted their young driver ban during the same year the survey was conducted (2013), thus limiting the sample to states that enacted young driver bans before survey administration; 3) excluding Texas, which was weighted as 39% of the total study population (the total population in Texas is much larger than other participating states; 4) including all students who drove in the past 30 days regardless of their age or licensing status.
Lastly, to estimate the association between cellphone laws and CWD as a nominal outcome, we fitted Poisson regression models to calculate the prevalence ratios for 1) sometimes CWD vs. never CWD, and 2) frequent CWD vs. never CWD.