Examining Seat Belt Use Habits in Wyoming: A Bayesian Hierarchical Logit Approach

Abstract

Wyoming has one of the lowest seat belt law compliance rates in the United States. Understanding the factors associated with seat belt non-use is crucial for the development of appropriate mitigation measures that aid in increasing seat belt usage rates. In this study, Bayesian binary logistic regression models were developed to: i) investigate the factors associated with daytime seat belt non-use using observational survey data; and ii) assess the factors related to seat belt non-use in fatal crashes while accounting for possible intra-class correlation (effects of unobserved factors common to occupants in the same vehicle). The results indicated that males were less likely to use seat belts than females. Also, trips made in pick-up trucks, those made within urban areas, those made during weekdays, and those made by means of vehicles with Wyoming license plates were likely to involve seat belt non-use. Furthermore, counties with thriving oil and gas industries had lower seat belt use rates. It was also inferred that young occupants, rear seat occupants, impairment, weekdays, traveling around midnight, traveling on collector roads, and traveling on local roads were associated with seat belt non-use. A strong correlation in seat belt non-use among occupants within the same vehicle warrants the implementation of the hierarchical logistic model to investigate seat belt use among occupants involved in fatal crashes. The findings of this study highlight remarkable trends in seat belt use habits in Wyoming, which will aid legislatures in proposing road safety measures aimed at raising seat belt usage proportions.

Keywords

occupant protection safety seat belts/safety belts

Motor vehicle crashes are among the leading causes of death in the United States. The seat belt is known to be an effective safety device in preventing fatalities and severe injuries resulting from motor vehicle crashes. By restraining people in their seats during crashes, seat belts prevent the occupants from being ejected from the vehicle. According to Deutermann ( 1 ), ejected occupants involved in rollover crashes are more likely to be killed than restrained occupants, and seat belt use lowers ejection risk. Also, the National Highway Traffic Safety Administration (NHTSA) reported that seat belt usage lowers fatality and serious injury likelihoods for front-seat passengers by almost half ( 2 ). Buckling up prevented roughly 14,955 fatalities in 2017 ( 2 ).

Despite the effectiveness of seat belts in lowering mortality rates and lessening severe injury counts, seat belt non-use still remains a concern. Wyoming has one of the lowest seat belt law compliance rates in the United States according to NHTSA’s National Occupant Protection Use Survey (NOPUS) data ( 3 ). A comparison for Wyoming’s seat belt use rates with nationwide average is depicted in Figure 1.

Figure 1.

Seat belt use rates in Wyoming and nationwide. (3)

In 2019, the nationwide seat belt use rate was 90.7%. Yet, in Wyoming, it was 78.3%. In addition, although the overall seat belt use rate has gradually increased over the years in the U.S., it decreased from 86.3% in 2018 to 78.3% in 2019 in Wyoming as shown in the Figure 1 ( 3 ).

Among several factors, seat belt law non-compliance poses an obstacle to efforts to reduce the number of fatal and severe injury crashes in Wyoming. According to the Insurance Institute for Highway Safety (IIHS), the fatality rate per 100,000 residents in Wyoming was 19.2 compared with the nationwide rate of 11.2 in 2018 ( 4 ). Also, Wyoming exhibited a larger proportion of unrestrained occupants involved in fatal crashes (50.6%) than the national average, which was 43.1% ( 4 ).

Understanding the factors associated with seat belt non-use is crucial for the development of appropriate measures that aid in increasing seat belt law compliance rates. Boakye et al. ( 5 ) showed that proper seat belt use is influenced by several factors including the occupant’s demographic characteristics, interpersonal factors (interacting with passengers, seat location), seat belt laws, and environmental factors. Road safety researchers ( 6 – 9 ) examined the association between seat belt laws and seat belt use rates among different states. McCartt and Northrup ( 6 ) and Lee et al. ( 7 ) found that seat belt use rates were greatly dependent on the seat belt laws. Likewise, Boakye and Nambisan ( 8 ) found a positive association between seat belt laws and seat belt use rates. Harper ( 9 ) revealed that switching from secondary to primary laws increased the number of buckled fatally injured occupants. Several studies also stated that primary seat belt law enforcement is more effective at increasing seat belt compliance rates and lowering traffic fatalities compared with secondary law enforcement ( 10 ). In addition to types of seat belt law enforcement (primary versus secondary), increasing fines for seat belt non-compliance and types of coverage with regard to vehicle occupants (whether seat belts are required for all seating positions versus only front-seat occupants) have significant impacts on seat belt use rates ( 11 – 13 ). A detailed description of the studies is provided in the literature review below.

Other studies have demonstrated significant associations between drivers’ seat belt use choices and those of their passengers ( 5 , 14 –19), indicating correlations among occupants in the same vehicle. Williams and Shabanova ( 16 ) and Cooper et al. ( 15 ) indicated that a driver’s seat belt use is positively associated with the presence of other passengers and their seat belt use. In particular, Williams and Shabanova ( 16 ) reported that seat belt use among teenage drivers increased when traveling with parents or older passengers and decreased when traveling with peers. Moreover, the seat belt use of both front-seat occupants has been found to be highly correlated. Recently, Chakrabarty et al. ( 19 ) concluded that driver seat belt use choice has a substantial influence on the front-seat passenger’s seat belt use choice and vice versa signifying a strong positive correlation among front-seat occupants’ seat belt use. The finding is consistent with the previous studies ( 17 , 18 ). Boakye et al. ( 5 ) also found a positive association between seat belt use choices among front-seat occupants at night. That is, front-seat passengers tend to not wear seat belts if their accompanying drivers do not buckle up.

Most of the seat belt studies were limited to either analysis of fatal crash data or direct observational study. Fatal crash data may not represent the general motorist population since they are restricted to the persons involved in the fatal crashes while observational surveys compile information from front-seat occupants during daytime conditions only. In seat belt studies, no study was found simultaneously investigating the fatal crash data and using direct observational data. To this end, this study is focused on developing a Bayesian model, which includes both types of data. Moreover, this paper contributes to the body of literature by drawing a comparison between the two types of data for their sample population and modeling results.

The objectives of this study were: i) to investigate the factors associated with daytime seat belt non-use by implementing observational survey data; and ii) to develop a statistical model to assess the factors related to seat belt use in fatal crashes using the observational survey data as prior information while accounting for possible intra-class correlation effects (those of the unobserved factors influencing occupants’ seat belt use choices within the same vehicle). The binary logistic regression structure with Bayesian inference was employed to develop two separate models using the observational survey data and the NHTSA’s Fatality Analysis Reporting System (FARS) data comprising crashes in the years 2010 to 2019

In this study, prior information of seat belt use choice interpreted from an observational survey conducted in Wyoming was used. Also, factors associated with seat belt non-use in fatal crashes considering unobserved vehicle-specific heterogeneity effects were investigated. The findings of this study might assist local legislatures in suggesting appropriate programs to increase seat belt usage rates which would reduce the injury severity of motor vehicle crashes in Wyoming.

Literature Review

The road safety literature is replete with studies investigating the impact of seat belt usage on crash severity and evaluating the association between seat belt laws and seat belt usage rates. In this section, a brief account of the recent traffic safety literature pertaining to the objectives of this work is provided.

Studies assessing the efficacies of the seat belt laws indicated that such laws are the most effective approach for increasing seat belt use rates ( 6 , 7 ). In particular, it was found that the primary seat belt laws (whereby drivers can be halted for unbuckled occupants) are more effective than the secondary seat belt use laws (whereby drivers can only be halted if another traffic infraction was committed). Boakye et al. ( 5 ) examined the association between the adult seat belt use rates and the seat belt laws in the U.S. by processing data from FARS. They implemented the Tukey multiple comparison test and concluded that the states with primary seat belt laws, applicable to each vehicle’s occupant, had larger seat belt use rates than those with secondary or no seat belt laws. Houston et al. also found that seat belt use rates were higher in the states with primary seat belt laws than those with secondary seat belt laws ( 20 ). In addition, imposing higher fines along with standard enforcement has demonstrated significant impacts on seat belt use rates in the states with primary seat belt laws. Harper ( 9 ) explored the influence of enhancing seat belt law enforcement policies on road fatalities by employing a Bayesian data augmentation approach. The author inferred that although switching from secondary to primary laws resulted in a marginal influence on fatality toll reductions, the number of buckled fatally injured occupants rose by 16%. McCartt and Northrup ( 6 ) explored the seat belt use rates of deceased teenage drivers, aged 16 to 19, by implementing a multivariate logistic regression structure. The results showed that seat belt use rates among deceased teenage drivers varied widely among multiple states and were strongly dependent on the seat belt laws. The latter finding was in line with that of Lee et al. ( 7 ).

Previous studies have also shown that demographic, behavioral, and physiological attributes of vehicle occupants are associated with seat belt use habits. Lerner et al. ( 21 ) investigated several demographic characteristics associated with seat belt use choice pertaining to adults injured in road crashes. They concluded that elderly drivers, females, and wealthy individuals were more likely to buckle up. In addition, drivers were found to be more willing to use the seat belt as opposed to their passengers. Beck et al. ( 22 ) compared the vehicle occupant fatality tolls of rural areas to those of urban areas in relation to the seat belt use habits of the victims, aged 18 or above, across the United States. The results indicated that rural areas had larger rates of unbelted occupant fatalities, regardless of the seat belt laws. In another study, Boakye et al. ( 5 ) investigated the factors that contributed to seat belt use choice of front-seat passengers during evenings by employing the generalized estimating equation technique. The results revealed that male passengers were less disposed to use the seat belt than female passengers and that trips made along local roads discouraged seat belt use. Also, front-seat passengers were less inclined to buckle up when riding in cars and pick-up trucks as opposed to when riding in sports utility vehicles (SUVs). Recently, Chakraborty et al. ( 19 ) explored the seat belt use habits of front-seat occupants by employing the bivariate probit modeling framework. The findings indicated that the seat belt use habits of front-seat occupants differed depending on their socio-demographic characteristics. Age and education levels of the occupants also influenced seat belt use choice. Briggs et al. ( 23 ) examined high school students’ seat belt use choices. The authors concluded that students, aged 16 or above, were unwilling to buckle up when riding as passengers as opposed to when driving. Shinar et al. ( 24 ) and Houston et al. ( 20 ) found that occupants with higher education and income levels were more likely to buckle up. Seat belt use rates were also found to increase with age ( 25 , 26 ). Nambisan et al. ( 27 ) interpreted that seat belt use also depended on vehicle characteristics. They investigated the impacts of vehicle age and type on occupants’ seat belt use habits using five years of crash data, 2014 through 2018, collected from FARS. The researchers concluded that occupants of more recent vehicles were likely to buckle up relative to those of dated vehicles.

Several studies demonstrated that occupants in the same vehicle influenced each other in deciding whether to buckle up. Afghari et al. ( 26 ) defined this as the “vehicle atmosphere”. Han ( 14 ) examined passenger seat belt use choice and found that it was substantially dependent on the drivers’ discretions on using the seat belt. The likelihood of passengers opting not to buckle up rose considerably when the driver was unrestrained relative to when the driver was restrained. Chakraborty et al. ( 19 ) also concluded that there was a strong correlation among front-seat occupants making decisions on seat belt use. Nukenine and Daniel ( 28 ) examined whether the seat belt use habits of rear seat passengers induced an effect on the buckled driver’s injury severity. The results revealed that seat beat use status and the count of rear seat passengers affected the driver’s injury severity. Also, the risk of front-seat occupants being severely injured rose when the rear seat occupants were unbuckled. Jermakian and Weast ( 29 ) found that misconceptions about the grave ramifications of not buckling up, lack of knowledge, or lack of comfort incurred as a result of proper restraint use were some of the reasons for not fastening the seat belt among rear seat passengers. These findings were consistent with those of Boyle and Lampkin ( 30 ) and Kidd and McCartt ( 31 ). Eluru and Bhat ( 32 ) demonstrated the need for capturing the mixed effects of the parameters on the outcome. They modeled the concurrent effects of seat belt use and the resulting crash severities as a function of multiple relevant factors. The authors incorporated random parameters and considered correlations among the parameters influencing the crash severity.

Recently, random-effects models and hierarchical models with Bayesian inference have become popular in road safety studies. This is a result of their ability to account for the correlations among the parameters and the hierarchical nature of the data ( 33 ). Huang and Abdel-Aty ( 34 ) proposed a multi-tiered data framework to highlight the need to select hierarchical Bayesian modeling approaches in crash data analysis. Haung et al. ( 35 ) examined the influential factors that contributed to intersection-related crashes by implementing the hierarchical Bayesian binomial logistic regression model. Haque et al. ( 36 ) applied several hierarchical models to model motorcycle crash frequencies at signalized intersections. Ahmed et al. ( 37 ) employed the hierarchical ordered logit model with the no U-turns Hamiltonian Markov chain Monte Carlo (MCMC) technique in their analysis of resulting severities of hazardous material truck crashes. Haq et al. ( 38 ) used the Bayesian binary logistic regression framework to model injury severities of crashes that occurred as a result of tire failures. Afghari et al. ( 39 ) implemented the Bayesian multivariate binary model with latent variables to assess whether vehicle occupants influenced each other’s seat belt use choices.

Seat belt use studies were limited to either analysis of fatal crash data or direct observational study. Fatal crash data may not represent the general motorist population since they are restricted to the persons involved in the fatal crashes while observational surveys compile information from front-seat occupants during daytime conditions only. The current study is focused on developing a Bayesian model, which combines both types of data. Seat belt use choice data were obtained from observational surveys and FARS. The approach accommodates potentially interrelated seat belt use habits of both drivers and passengers. It also permits parameters to vary randomly across vehicle occupants, accounting for unobserved heterogeneity effects, such as roadway characteristics, vehicle attributes, and driver behaviors. To the best of the authors’ knowledge, no previous studies on seat belt use involved two data sets concurrently to capture the variability in the data.

Data Description

Two data sets were used in this study. The first was collected from a seat belt observational survey conducted by trained observers during the first week of June 2019 while the second was collected from FARS. It is worth mentioning that observational seat belt surveys are conducted every year to explore seat belt use rates among front-seat occupants in vehicles across many states. Wyoming observational surveys have also been conducted every year since 2012 in compliance with the national standards ( 40 ). Therefore, the data set represents a wider time frame since it is collected once every year. To ensure that the status of seat belt use with regard to the front-seat occupants within the same vehicle was recorded, the Uniform Criteria for State Observational Surveys of Seat Belt Use 23 CFR (Code of Federal Regulations) § 1340 standards were followed when conducting the survey ( 40 ).

For the first data set, a total of 24,821 observations of drivers and front-seat passengers in 18,286 vehicles were collected from 289 sites in 17 counties of Wyoming. According to the NHTSA, sample counties are selected as the primary sampling unit of the survey based on the vehicle miles traveled or population ( 41 ). It is worth noting that site locations are selected in a way that results in residents being likely to be observed ( 40 ). Moreover, multiple observation sites were picked to capture seat belt usage from a variety of motorists and avoid biases in data ( 41 ). More than two-thirds of the data were collected from secondary and local roads to guarantee that a majority of the observed vehicle occupants were local residents of the respective counties ( 40 ). This survey was conducted to explore seat belt usage by county and the association between seat belt use and several variables. These included population density, weather, vehicle registration, occupant gender, vehicle type, and occupant type (driver or passenger). Standard protocols were followed from observer training to data analysis to ensure the reliability and accuracy of the observed data ( 40 ). Figure 2 shows trends in seat belt use among occupants riding in vehicles in Wyoming as per the survey results and Table 1 lists summary statistics of the survey’s variables.

Table 1.

Summary Statistics of the Observational Survey’s Variables

Variables	Levels	Unbelted
Variables	Levels	Count	Proportion (%)	Total
Weather	Clear	3,492	20.07	17,399
Weather	Not clear	1,292	17.51	7,378
Area	Urban	1,857	31.33	5,927
Area	Rural	2,927	15.53	18,850
Road type	Primary	1,383	17.85	7,748
	Secondary	3,142	19.74	15,917
	Local	259	23.29	1,112
Number of lanes per direction	One lane	2,500	17.86	13,998
Number of lanes per direction	Two lanes	2,284	21.19	10,779
Vehicle type	Passenger car	1,043	17.06	6,112
	Van	1,081	14.16	7,634
	SUV	197	13.20	1,492
	Pick-up truck	2,463	25.82	9,539
Vehicle registration	Wyoming	3,373	23.88	14,123
Vehicle registration	Other state	1,364	13.00	10,496
Time of observation	7:30–9:30 a.m.	870	19.60	4,439
	9:30–11:30 a.m.	752	17.28	4,351
	11:30 a.m.–1:30 p.m.	1,061	18.71	5,672
	1:30–3:30 p.m.	743	18.87	3,938
	3:30–5:30 p.m.	1,352	21.22	6,371
Weekend	Weekend	580	14.92	3,888
Weekend	Weekdays	4,204	20.13	20,889
Gender	Male	3,326	22.89	14,528
Gender	Female	1,458	14.23	10,249

Note: SUV = Sports utility vehicle.

Figure 2.

Occupants’ seat belt use proportions in Wyoming for 2019.

The seat belt usage rate varied from a low of 63.5% belted occupants in Sweetwater County to a high of 97.8% in Niobrara County (Figure 2). In this survey, areas with more than 5,000 residents were identified as “urban,” while areas with less than 5,000 residents were considered “rural.” Also, 76.1% of the observations were witnessed in rural areas. According to the survey’s results, seat belt use rates were higher in rural than in urban areas. Similarly, seat belt usage rates were found to be higher on primary roads than on secondary and local roads. Furthermore, occupants were more likely to wear their seat belts during weekends. Although males comprised more than 80% of all vehicle occupants, they were 13.7% less likely to use their seat belts than female occupants.

Seat belt use was observed for a variety of vehicles, but they were broadly categorized as passenger cars, SUVs, pick-up trucks, and passenger minivans. With regard to vehicle types, pick-up truck occupants were likely to have the lowest seat belt use rate. Other than the vehicle type, weather conditions were condensed into two categories: clear and not clear. Observers also recorded the occupants’ vehicle license plates, whether they were Wyoming plates or not. With that, the survey results indicated that occupants of out-of-state vehicles exhibited higher rates of seat belt use than their Wyoming counterparts.

The second data set was collected from NHTSA’s FARS database. The database was queried to select records of Wyoming’s fatal crashes occurring from 2010 to 2019. This data set included records of 2,705 vehicle occupants who were involved in 1,135 fatal crashes, regardless of whether they suffered a fatal injury or not. The crash records included variables related to occupant demographics, vehicle characteristics, roadway characteristics, crash type, and environmental conditions at the times of the crashes. Categorical and quantitative variables related to seat belt use were extracted from the crash records to ascertain the factors that contributed to seat belt use habits. Roadway types were classified as interstates, primary arterials, and collector/local roads. The times of crashes, included in this analysis, were categorized into three classes: daytime (5 a.m.–5 p.m.), nighttime (5 p.m.–midnight) and midnight (midnight–5 a.m.). Although crashes occurred on all seven days of the week, the days were merged into weekdays and weekends for reasonable interpretation of the results. Likewise, the months, during which the crashes occurred, were conjoined into quarters: January to March as the first quarter, April to June as the second quarter, July to September as the third quarter, and October to December as the fourth quarter. Occupant characteristics, such as gender, age, seat location, and type of occupant (driver or passenger) were also included in this analysis. According to NHTSA, occupants 65 years and older are considered as elderly residents ( 42 ) while the Wyoming child restraint law states that children aged 9 years and younger are required to use child restraint systems ( 43 ). For maintaining consistency with NHTSA and Wyoming laws, the occupant ages were classified into seven categories: children (9 years and younger), teenage/young occupants (10–18 years), 19–30 years, 31–40 years, 41–50 years, 51–65 years, and elderly occupants (66 years and older). Note that the intermediate groups were classified such that the sample size of each age group was adequate and roughly equal to those of the others. The occupants’ seating locations were dichotomized as front and rear seats. In total, 1,104 occupants (41% of the total occupants) were unbuckled at the times of the crashes. Detailed analysis on seat belt non-use is provided in the following paragraph. Table 2 presents the summary statistics of the data set extracted from FARS while Figure 3 depicts the seat belt use rates among occupants riding in vehicles in Wyoming.

Table 2.

2010 to 2019 Fatal Crash Data Variables’ Summary Statistics

Variable	Category	Unbelted
Variable	Category	Count	Proportion (%)	Total
Months	First quarter (January–March)	229	45.17	507
	Second quarter (March–June)	274	36.73	746
	Third quarter (July–August)	294	42.18	697
	Fourth quarter (September–December)	307	40.66	755
Time	Daytime (5 a.m.–5 p.m.)	587	36.78	1,596
	Night (5 p.m.–midnight)	383	45.06	850
	Midnight (midnight–5 a.m.)	134	51.74	259
Roadway functional classification	Interstates	270	36.64	737
	Principal arterials	201	36.15	556
	Collectors and local roads	633	44.83	1,412
Gender	Female	351	38.03	923
Gender	Male	753	42.26	1,782
Alcohol	No	668	31.90	2,094
Alcohol	Yes	436	71.36	611
Drug	No	980	39.06	2,509
Drug	Yes	124	63.27	196
Age	Children (9 years and younger)	40	23.39	171
	10–18 years	132	47.14	280
	19–30 years	347	51.41	675
	31–40 years	176	42.11	418
	41–50 years	149	39.63	376
	51–65 years	200	37.59	532
	Elderly occupants (66 years and Older)	64	21.55	297
Occupant type	Driver	659	41.03	1,606
Occupant type	Passenger	445	40.49	1,099
Seating position	Front seat	850	40.36	2,106
Seating position	Rear seat	254	42.40	599
Day of week	Weekday	737	39.92	1,846
Day of week	Weekend	367	42.72	859

Figure 3.

Occupants’ seat belt use rates in fatal crashes from 2010 to 2019 in Wyoming.

Seat belt non-usage rates varied across counties from a high of 61% unbuckled occupants in Crook County to a low of 21% unbuckled occupants in Teton County (Figure 3). Table 2 shows that seat belt non-use rates ranged from 37% in the second quarter to 45% in the first quarter of the crash years. Furthermore, seat belt non-use rates were higher during weekends than weekdays. They were also observed to be the highest during midnight conditions.

The lowest passenger seat belt non-use rate was observed on principal arterials and the highest was observed on collector/local roads. In regard to gender, males had larger proportions of seat belt non-use rates than females. Seat belt non-use rates were lower in children and elderly occupants. By contrast, they were found to be higher among teenagers and young adults (aged 10–18 and 19–30, respectively). In regard to occupant types, seat belt non-use rates were higher among passengers than among drivers. Likewise, rear seat occupants had a larger proportion of seat belt non-use than front-seat occupants. Finally, alcohol or drug involvement substantially reduced seat belt use rates.

Research Methodology

In this study, Bayesian binary logit models with both fixed and random effects were applied to investigate the factors associated with seat belt use. Bayesian models treat the parameters as random variables, unlike frequentist models, which consider the parameters as fixed. In Bayesian analysis, the observed data are used to update the prior beliefs about the behavior of the parameters when estimating their distributional properties ( 44 ). One of the major advantages of Bayesian frameworks is their ability to select parametric families for prior probability distributions. Three types of prior distributions are mainly used in Bayesian approaches for estimating parameters including:

Informative priors based on the literature or previous relevant studies;

Weakly informative priors, which provide more realistic inferences being selected from trends (means and standard deviations); and

Non-informative priors, which allow the information from the likelihood to be interpreted.

Past studies ( 37 , 45 ) have shown that the incorporation of weakly informative priors is recommended since they enhance the understanding of the parameters’ distributional properties by confining the resulting posterior probabilities obtained within a reasonable range. In this study, prior distributions were drawn from the preliminary data analysis of the observational survey data and assumed to be normally distributed. Weekly informative priors were selected in a way that parameter estimates can be calculated with minimum bias and at the same time model convergence can be attained in substantially less time.

Various MCMC sampling techniques have been widely used, such as the Gibbs sampler, Metropolis Hastings algorithm, or the combination of both using the freeware, WinBugs, to obtain samples from the posterior distribution. Recently, the no U-turns Hamiltonian MCMC method, an extension of the Hamiltonian MCMC sampling technique, has been gaining momentum given its effectiveness and robustness over the other sampling techniques ( 37 , 46 ). In this study, the Bayesian Regression Models (brms) package of the R^® statistical software is implemented to estimate the models ( 47 ).

Bayesian logistic regression models were used to model the relationship between the dichotomous response variable (seat belt non-use versus use) and the explanatory variables including occupant characteristics, environmental attributes, and other factors while considering within vehicle correlations. The model structure is described as follows. Let Y_ij denote the dependent variable, which represents the seat belt use choice of the ith occupant within the jth vehicle. The outcome of this distribution is binary, either 0 (buckled) or 1 (unbuckled). The probability of not using the seat belt, Y_ij = 1, is denoted as P(Y_ij) and conforms to the logistic regression equation, expressed as:

\begin{matrix} Logit (P (Y_{ij} = 1)) = \ln (\frac{P (Y_{ij} = 1)}{1 - P (Y_{ij} = 1)}) \\ = α + \sum_{p = 1}^{P} β_{p} X_{p} \end{matrix}

(1)

\begin{matrix} Logit (P (Y_{ij} = 1)) = \ln (\frac{P (Y_{ij} = 1)}{1 - P (Y_{ij} = 1)}) \\ = (α + u_{j}) + \sum_{p = 1}^{P} β_{p} X_{p} \end{matrix}

(2)

Equation 1 represents the fixed effect binary logistic regression while Equation 2 presents the random effect binary logistic regression model. The term, α, is the model’s constant and u_j represents the random effects of the vehicle. It should be noted that u_j is assumed to be normally distributed with a mean of 0 and a standard deviation of σ, u_j∼N(0, σ²). Also, X_p denotes the explanatory variables and β_p is their respective regression coefficients. It is worth mentioning that the fixed effect Bayesian binary logit model was used to analyze the observational survey data while the random effect Bayesian binary logit model was employed to investigate the FARS data.

In Bayesian analysis, the resulting posterior distribution, P(θ|Y), is equivalent to the product of the likelihood of the observed data, P(Y|θ), and the prior information of the parameters, P(θ), divided by the average likelihood, P(Y), as shown in Equation 3.

P (θ | Y) = \frac{P (Y | θ) \times P (θ)}{P (Y)}

(3)

The average likelihood, P(Y), is also known as the marginal distribution of Y. It is computed as follows:

P (Y) = \int P (Y | θ) d θ

(4)

Since P(Y) is independent of θ, the posterior distribution is only proportional to the product of the likelihood and the prior as shown in Equation 5.

P (θ | Y) \propto P (Y | θ) \times P (θ)

(5)

An intra-class correlation (ICC) coefficient was incorporated in this study to quantify the unobserved variability of seat belt non-use among occupants within the same vehicle. The following equation was employed to compute the ICC.

ICC = \frac{σ^{2}}{σ^{2} + σ_{o}^{2}}

(6)

In Equation 6, σ² represents the variability of seat belt non-use rates across vehicles, and σ_O² denotes the variability within the vehicle, which is assumed to be fixed across vehicles and equal to π2/3 ≈ 3.293 for a hierarchical logistic distribution ( 37 ). An ICC value near 0 indicates a small variation that can be explained by the group-level differences and, thus, a non-random effect ordinary logit model may be adequate to fit the data. On the other hand, an ICC value near 1 indicates substantial variations among groups and the use of the hierarchical model is, therefore, justified ( 48 ).

In the analyses, Bayesian credible intervals (CIs) describe the uncertainty related to the covariates by assigning a range of values obtained from their posterior probability distributions. In this study, the 95th percentile Bayesian CI, which consists of the central portion (2.5%–97.5%) of the posterior distribution of the covariates, was used. Covariates, of which 95th percentile CIs did not contain 0, were identified as significant predictors ( 49 ).

Odds ratios (ORs) were estimated to interpret the effect of the significant variables on seat belt non-use. The OR is the odds of observing an unbuckled occupant in the presence of a particular parameter relative to those in the absence of the parameter, provided that all else was unchanged.

Before model fitting, multicollinearity was assessed among the explanatory variables using the correlation test, tolerance and variance inflation factor (VIF). Absolute correlation coefficients near 1 represent strong correlations between variable pairs and VIF values, while greater than 10 indicate the presence of strong multicollinearity ( 50 ). Also, tolerance levels less than 0.1 denote the existence of multicollinearity ( 51 ). In this study, the three measures’ results ensured the absence of multicollinearity among the explanatory variables.

Among several model goodness of fit measures, the Watanabe–Akaike information criterion (WAIC) ( 52 ), has been gaining popularity as a performance measure of Bayesian models. As opposed to the Akaike information criterion (AIC) and the deviance information criterion (DIC), WAIC is capable of averaging the likelihoods, P(Y|θ)s, over the posterior distribution ( 49 ). The WAIC includes the log-pointwise predictive density (LPPD) in its formulation, which is calculated as follows ( 49 ):

LPPD = \sum_{i = 1}^{N} \ln (\frac{1}{S} \sum_{s = 1}^{S} P (Y_{i} | θ_{s}))

(7)

The term, S, is the number of MCMC samples drawn from the posterior distribution. A correction factor that accounts for the effective number of parameters, p_WAIC, is included in the WAIC estimation. It is computed as ( 49 ):

p_{WAIC} = \sum_{i = 1}^{N} VAR (\ln [P (Y_{i} | θ_{1})], \ln [P (Y_{i} | θ_{2})], \dots, \ln [P (Y_{i} | θ_{s})])

(8)

Finally, the WAIC measure is computed by the following formula ( 49 ):

WAIC = - 2 LPPD + 2 p_{WAIC}

(9)

Posterior predictive assessments may also be undertaken to verify whether the model adequately explains the observed variability in the data. The rationale of those assessments is that they indicate whether the model can be applied to generate data that are similar to the observed data, provided that it exhibits a good fit ( 53 ). Both the WAIC metric and the posterior predictive assessments were employed in this study to evaluate model performances.

Results and Discussions

Two Bayesian models were developed to aid in understanding seat belt use habits in Wyoming. The first was a binary logit model with seat belt use as the response. The 2019 observational seat belt survey data for Wyoming was used for its estimation. The second was a random-effects model with the seat belt use as the outcome as well. It was estimated using the FARS data pertaining to Wyoming’s conditions from 2010 to 2019. Note that the means of the seat belt use choices, obtained from the observational survey, were incorporated as priors in the constant term of the latter model. As a method of statistical regularization, weakly informative priors shrank the parameter estimates toward 0 and reduced the Type I error rates ( 45 ). Similar to previous studies ( 37 , 38 , 54 , 55 ), incorporating such priors in this study helped to constrain posterior distributions obtained to a reasonable range, which in turn enabled to estimate the parameters with minimum bias. Weakly informative priors also facilitated attaining model convergence in substantially less time. Three Monte Carlo Markov chains were prepared for each model. With the aid of R’s “brms” package, the Hamiltonian MCMC sampling method was conducted to obtain the posterior distributions of the samples. For each chain, 3,000 iterations were simulated with 1,000 warm-up samples after rigorous fine-tuning of the model specifications to achieve convergence. In total, 6,000 posterior samples of the three chains were obtained. In addition, a target acceptance rate of 90% and a maximum tree depth of 10 were considered in the models. The maximum tree depth indicates the maximum number of allowable steps of no U-turn sampling. It is used to prevent premature termination ( 48 ). The target acceptance rate is used to control step size. The higher the target acceptance rate, the lower the step size resulting in less computation time.

Model convergence in this study was ensured by generating trace plots and results of the Gelman–Rubin convergence diagnostic, Rhat ( 56 ). The generated trace plots showed that the chains mixed well, indicating identical distributions of the simulations between and within chains. The Rhat measure, also known as the “potential scale reduction factor,” is the square root of the ratio of the variance of all the chains to the average within-chain variance ( 46 ). In the final model, variables were identified as significant based on the 95th percentile Bayesian CI. The effective sample size (ESS) metric gauges the level of autocorrelation of the samples within the chains. The closer the ESS to the total number of samples, the more independent samples are within the chains. In this study, the ESS results were found to be close to the total number of samples, which indicated that the samples were independent. The results of both models are presented in Tables 3 and 4. In the tables, the standard errors (SEs) are the approximate standard deviations of the sample populations. Figure 4 presents sample trace plots of several significant variables from the random-effects model.

Table 3.

Observational Survey Data Model Results

	Population-level effects			Credible interval
		Estimate	SE	2.5%	97.5%	OR
	Constant	−3.140	0.120	−3.370	−2.910	0.043
County	Campbell	1.090	0.100	0.900	1.270	2.974
	Carbon	1.360	0.100	1.170	1.550	3.896
	Converse	0.850	0.100	0.660	1.030	2.340
	Crook	−0.420	0.130	−0.680	−0.170	0.657
	Johnson	0.260	0.130	0.010	0.510	1.297
	Park	0.720	0.100	0.520	0.910	2.054
	Sheridan	0.180	0.100	−0.020	0.390	1.197
	Sweetwater	1.370	0.090	1.190	1.550	3.935
	Albany (reference)
Gender	Male	0.470	0.040	0.400	0.550	1.600
Gender	Female (reference)
Number of lanes per direction	Two lanes	0.120	0.050	0.020	0.210	1.127
Number of lanes per direction	One Lane (reference)
Day of the week	Weekdays	0.260	0.050	0.150	0.360	1.297
Day of the week	Weekend (reference)
Locations	Urban	0.430	0.050	0.340	0.530	1.537
Locations	Rural (reference)
Vehicle type	Van	−0.320	0.050	−0.420	−0.220	0.726
	SUV	−0.490	0.090	−0.670	−0.320	0.613
	Pick-up truck	0.250	0.050	0.160	0.340	1.284
	Passenger car (reference)
Vehicle registration	Wyoming	0.350	0.040	0.270	0.430	1.419
Vehicle registration	Other state
Road type	Secondary road	0.310	0.060	0.190	0.430	1.363
	Local road	0.540	0.210	0.140	0.960	1.716
	Interstate (reference)
WAIC		21,772.4	175.4

Note: SE = standard error; OR = odds ratio; WAIC = Watanabe–Akaike information criterion.

Table 4.

Fatality Analysis Reporting System Data Model Results model

	Random-effects model					Fixed-effects model
			Credible intervals					Credible intervals
Random effect	Estimate	SE	2.50%	97.50%	ICC	Estimate	SE	2.50%	97.50%	OR
Intercept standard deviation	2.04	0.18	1.7	2.4	0.57	N/A	N/A	N/A	N/A	N/A
	Estimate	SE	2.50%	97.50%	OR	Estimate	SE	2.50%	97.50%	OR
Intercept	−1.83	0.33	−2.49	−1.19	6.23	−1.27	0.2	−1.66	−0.87	3.56
County
Big horn	1.02	0.48	0.11	1.97	2.77	0.62	0.3	0.03	1.2	1.86
Crook	1.4	0.46	0.5	2.31	4.06	1.16	0.3	0.57	1.75	3.19
Sheridan	1	0.45	0.11	1.91	2.72	0.77	0.28	0.23	1.31	2.16
Sweetwater	0.68	0.33	0.05	1.33	1.97	0.56	0.2	0.18	0.95	1.75
Time
Midnight (midnight–5 a.m.)	0.46	0.27	−0.07	0.99	1.58	0.36	0.15	0.06	0.65	1.43
Day time (5 a.m.–5 p.m.)
Age
Children (9 years and younger)	−1.07	0.32	−1.71	−0.45	2.92	−0.8	0.22	−1.23	−0.36	2.23
Age (9–18 years)	0.48^*	0.26	−0.02	0.98	1.62	0.32	0.16	0.01	0.63	1.38
Age (18–30 years)	0.42	0.2	0.03	0.81	1.52	0.32	0.13	0.07	0.57	1.38
Older people (65 years and older)	−0.71	0.25	−1.2	−0.22	2.03	−0.57	0.17	−0.92	−0.23	1.77
Age (50–65 years)
Seating position
Rear seat	0.88	0.18	0.52	1.23	2.41	0.52	0.12	0.29	0.76	1.68
Front seat (reference)
Alcohol
Yes	2.24	0.2	1.87	2.64	9.39	1.56	0.11	1.34	1.79	4.76
No (reference)
Functional classification of roads
Collectors and local roads	0.41*	0.22	−0.01	0.84	1.51	0.15	0.12	−0.09	0.39	1.16
Interstates (reference)
WAIC	2,853.9	48.8				3,269.9	44.4

Variables significant at the 90% credible interval.

Note: SE = standard error; OR = odds ratio; ICC = intra-class correlation.

Figure 4.

Sample variable trace plots.

Remarks on the Observational Survey Data Model Results

The results of the binary logit model, presented in Table 3, reveal remarkable insights into seat belt use choice. It should be noted that incorporating the vehicle identification as a variable to capture correlations among occupants in the same vehicle was worth the attempt. However, this variable was not available in the data set. Nevertheless, all estimated parameters were identified as influential since their Bayesian 95th percentile CIs did not contain 0. The impact of each parameter on the response was interpreted assuming all else was controlled. The survey results demonstrated that vehicle occupants in Campbell, Carbon, Converse, Johnson, Park, Sheridan, and Sweetwater counties were likely to be unrestrained by 2.97, 3.90, 2.34, 1.30, 2.05, 1.20, and 3.94 times, respectively compared with those of which data were sampled from Albany County. On the contrary, occupants observed in Crook were less likely to be unrestrained by 0.34 times. These findings were consistent with those of previous studies ( 5 , 22 ). Furthermore, the oil and gas industry is the single largest economic driver in Wyoming, while Sweetwater, Campbell, Carbon, Park, Converse, Sheridan, Johnson, and other counties are characterized by mining activities. From Table 3, it is evident that occupants observed in these counties were more likely to be unrestrained, which led to the inference that oil and gas industry workers might have been unwilling to use the seat belts. According to the Wyoming Department of Workforce Services, 38% of fatally injured workers were unbuckled throughout the years 2012 to 2018 ( 57 ). Statistics also suggested that oil and gas industry workers experience six times the rate of motor vehicle-related death compared with workers in other industries ( 58 ). Another study also found a substantial increase in crash severity and frequency given the expanding oil industry in North Dakota ( 59 ). However, observational survey data do not have information with regard to occupants’ occupations. Future studies should explore additional evidence if there is an association between low seat belt use and oil/gas industry workers’ trips. In this study, oil and gas production data from 2019 were collected from the Wyoming Oil and Gas Conservation Commission ( 44 ), and analyses were conducted to assess the correlations between seat belt non-use and oil/gas production (Table 5). As shown in the table, such correlations existed.

Table 5.

Correlation Matrix of Seat Belt Non-Use Proportions and Oil and Gas Production

	Proportion of unbelted occupants	Oil production	Gas production
Proportion of unbelted occupants	1	0.472	0.619
Oil production	0.472	1	0.396
Gas production	0.619	0.396	1

The difference in seat belt use habits among the counties may also be a result of the level of urbanization, differences in occupant attitudes toward enforcement and seat belt use awareness programs across the different counties. Appropriate interventions for each county ought to be considered to increase seat belt usage rates.

In the case of the occupant gender, the results indicated that the estimated odds of not using the seat belt increased by 1.6 times for males, on average. This finding was consistent with several previous studies ( 5 , 19 , 60 ).

Significant differences in seat belt use choice were also found among occupants of different vehicle types. According to the results, pick-up truck occupants were more likely to refrain from buckling up compared with passenger car occupants (OR = 1.28). On the other hand, when riding in vans and SUVs, the estimated odds of seat belt non-use decreased by 0.27 and 0.39 times, respectively. In multiple past studies, lower seat belt usage rates among occupants of pick-up trucks were observed compared with those of other vehicle types including sedans, SUVs, and minivans ( 19 , 60 , 61 ). Combining media campaigns with high visibility programs, such as “Buckle up in your truck,” were demonstrated to be effective in prompting pick-up truck occupants to fasten their seat belts as per ( 5 ).

As shown in Table 3, vehicle occupants traversing within urban areas were more likely to be unbuckled by 1.54 times compared with those traversing within rural areas. The finding is in line with the research carried out by Lipovac et al. ( 62 ) and Huang et al. ( 63 ). They inferred that drivers were unwilling to wear seat belts when traveling short distances on low-speed roads in urban areas ( 62 , 63 ). Although these results, low seat belt use in urban areas, contradicted most of the previous studies ( 50 , 51 ), it might be a result of a compensation effect in which travelers perceived less risk of not using the seat belt when traveling shorter distances in urban areas. On the other hand, given geographically dispersed/sparsely located amenities in rural areas ( 64 ), trip-takers traveled long distances to reach their destinations ( 65 ) and, therefore, would be inclined to buckle up. On a different note, the notion of rural and urban is very different in this least populated state relative to the more populated and geographically smaller states which may also be a reason for the opposing finding ( 66 ).

When it comes to the influence of the day of the week, the estimated odds of not using the seat belt were 1.30 times higher on weekdays than on weekends. A similar result was found in a previous study ( 60 ). This finding might also lead to the interpretation that oil and gas industry workers were likely to be unrestrained while commuting short distances to their job sites during weekdays.

Other than the day of the week’s influence on seat belt use habits, the findings revealed that occupants of vehicles registered in Wyoming had a lower chance of buckling up than those of out-of-state vehicles (OR = 1.42). Plausibly, this might be because Wyoming’s residents perceive lower risks of not using seat belts when traveling on familiar roads which consequently makes them more susceptible to crashes. Historically, Wyoming residents are two to three times more likely to be unrestrained and killed in a crash than nonresidents ( 67 ). According to the Wyoming DOT, two-thirds of all drivers involved in crashes had a Wyoming driver’s license in 2020 ( 68 ). Further studies should be carried out to investigate the reasons behind the unwillingness to use seat belts among Wyoming residents.

When it comes to the road type, the likelihood of not using seat belts increased on local and collector roads compared with interstates. As a result of the low speed, short distances, and road facilities, travelers perhaps perceived lower risks on local and collector roads, discouraging the use of seat belts. The number of lanes per direction also induced effects on seat belt use choice. As shown in Table 3, the odds ratio was estimated as 1.13 when having two lanes relative to one, a finding in line with that of Rezapour and Ksaibati ( 69 ).

Remarks on the Fatality Analysis Reporting System Data Model Results

Table 4 presents the comparison between the results of the random-effects Bayesian model and of the fixed-effects model, which were developed to investigate the factors associated with seat belt non-use in Wyoming using the FARS data from 2010 to 2019. The WAIC of the random-effects model was found to be lower than that of the fixed-effects model, indicating that the random-effects model demonstrated better performance than the fixed-effects model. Descriptive statistics of the observational survey were incorporated as weakly informative priors into the model. As previously stated, studies have demonstrated the existence of correlations among vehicle occupants affecting seat belt use habits ( 5 , 14 ). In this study, the ICC value was computed to measure such correlations. The variance of the occupants’ seat belt use choices (estimated as 4.16) was used to calculate the ICC as follows:

ICC = \frac{4.16}{4.16 + \frac{π^{2}}{3}} = 0.558

The ICC value indicated that 55.8% of unexplained variations were a result of the presence of vehicle-specific unobserved factors associated with seat belt non-use ( 70 ). McElreath ( 48 ) recommended the use of hierarchical models when unobserved heterogeneity effects were observed in the model. Therefore, given the existence of vehicle-specific unobserved factors associated with seat belt non-use, the hierarchical logit model was employed to account for this unobserved heterogeneity.

The model parameters were interpreted provided that all else was unchanged. As shown in Table 4, vehicle occupants in Big Horn, Crook, Sheridan, and Sweetwater counties had 2.77, 4.06, 2.72 and 1.97 higher odds of being unrestrained, respectively, compared with those observed in Albany County. These findings were consistent with those of previous studies ( 8 , 22 ).

Other than the seat belt use rates by county, it could be inferred from the model results that occupants were more likely to be unrestrained during midnight (midnight–5 a.m.) conditions compared with when riding during daytime conditions. This result was in line with that of ( 5 ) who interpreted that refraining from using the seat belt during the evening might be a result of the misconception of the improbability of receiving a citation at night among other possible reasons. Several other studies have also shown that the time of day has a substantial impact on road users’ seat belt usage ( 61 , 71 –73). Another study found that single-vehicle occupants are less likely to be restrained in the morning while multi-vehicle occupants are less likely to be restrained in the afternoon ( 60 ).

In the case of occupant ages, it was found that the age groups, 10 to 18 years and 19 to 30 years, had 1.62 and 1.52 higher odds, respectively, of being unrestrained compared with occupants of the age group, 51 to 60 years. By contrast, the estimated odds of being unrestrained decreased by 0.66 and 0.51 times for children (9 years and younger) and elderly occupants (66 years and older), respectively, compared with occupants aged 51 to 65. These inferences were consistent with other studies ( 14 , 21 ). Furthermore, NHTSA reported that, in 2017, 54% of unbuckled car occupants involved in crashes, aged between 13 and 15, were fatally injured in road crashes ( 2 ).

The seat location was another factor that was found to have an impact on seat belt use habits. The odds of rear seat passengers not using the seat belt were 2.41 times higher than those of front-seat passengers. The false perception that rear seats would be safer than front seats might make rear seat occupants less likely to buckle up. According to Findley et al. ( 74 ), buckled rear seat occupants not only protected themselves but also encouraged front-seat occupants to use their seat belts. As such, appropriate mitigation measures ought to be implemented to increase seat belt use rates among rear seat occupants. Alcohol involvement was also found to play a significant role in seat belt use choice. From the analysis results, the estimated odds of not using the seat belt increased by 9.39 times as a consequence of alcohol intoxication. Han ( 14 ) concluded that passengers were likely to refrain from using their seat belts when the drivers were driving under the influence of alcohol and Afghari et al. ( 26 ) reported that passengers would be less willing to buckle up when they themselves were riding under the influence.

Other than alcohol involvement, the roadway functional classification was also found to influence seat belt use choice. The likelihood of not using the seat belt was larger for collectors and local roads compared with interstates. This finding was in line with that of the observational survey model results and of previous studies ( 5 , 60 ). Occupant gender and crash dates were other variables that were investigated in this analysis. However, those parameters were not found to have an impact on seat belt use habits (Bayesian CI included 0).

Incorporating random parameters into the model to capture unobserved heterogeneity effects is another approach worth selecting ( 75 ). The authors attempted the model with the age groups of the occupants as random parameters. However, the WAIC value of the random parameters model indicated that the random-effects model exhibited a better fit than the random parameters model (3,305 and 2,854 for the random parameters model and the random-effects model, respectively).

Conclusions

Previous seat belt use studies have focused either on the analysis of fatal crash data or on the use of observational survey data. This study offers a contribution to the occupant protection literature in that prior information about seat belt use choice, obtained from an observational survey, was introduced in the investigation of the factors associated with seat belt use in fatal crashes. Also, unobserved heterogeneity effects among vehicles were considered. Two Bayesian binary logistic regression models were developed. One was estimated using records of daytime seat belt use data collected from a survey disseminated in Wyoming in 2019. The second was developed for fatal crash records, collected from NHTSA’s FARS data on crashes from the years 2010 to 2019 in the state.

The results of the observational survey model indicated that males were less likely to be restrained than females and occupants traveling in pick-up trucks had a lower chance of using their seat belt compared with occupants of passenger cars. Furthermore, urban area trips were characterized by a lower probability of seat belt use relative to rural area trips. Similarly, the results indicated that weekday trips were less likely to be characterized by restrained occupants as opposed to weekend trips. In addition, occupants of vehicles having Wyoming license plates were less inclined to use their seat belts than occupants of out-of-state vehicles. The results also showed that counties with vibrant oil and gas industries such as Campbell, Carbon, Converse, Johnson, Park, Sheridan, and Sweetwater had higher seat belt non-use rates. It is thus inferred that workers/employees of the oil and gas industries may refrain from using seat belts. This finding is crucial for legislatures since they may target this specific group.

The results of the second model revealed insights into the seat belt use choices of occupants involved in fatal crashes based on a variety of factors and unobserved heterogeneity effects among vehicles. The ICC value signaled a strong correlation in seat belt non-use among occupants of the same vehicle. It was also inferred that vehicle occupants, on whom records were sampled from Big Horn, Crook, Sheridan, and Sweetwater, were more likely to be unrestrained compared with those on whom records were sampled from Albany County. In addition, the results indicated that the occupants, representing the fatal crash data, were more likely to refrain from using the seat belts during the late night hours of midnight to 5 a.m. compared with those traveling during the daytime (5 a.m. to 5 p.m.). It was also interpreted that young occupants (10 to 18 years and 19 to 30 years) had a higher chance of being unrestrained relative to occupants, aged 51 to 65. On the contrary, children (9 years and younger) and elderly occupants (66 years and older) were less likely to be unrestrained in relation to occupants aged 51 to 65. Also, rear seat occupants were found to be less disposed to buckle up compared with front-seat occupants. In addition, it was inferred from the findings that alcohol impairment considerably raised the probability of seat belt non-use. The roadway functional classification was another factor that was found to influence seat belt use habits. That is, occupants were less likely to wear their seat belts when traveling along collector and local roads.

Other than the inferences of the individual explanatory variables from the models’ results, seat belts use choices may also vary across the occupant population. As previously stated, observation survey results are those of front-seat occupants traveling during the day in 2019 while the NHTSA’s FARS data represent occupants involved in fatal crashes during 2010 to 2019. Given the variability among occupant populations and time frame, inferences drawn from the observational survey data modeling results differed from those of the FARS data modeling results. For instance, in the results of the daytime observational survey data model, vehicle occupants observed in Campbell, Carbon, Converse, Johnson, Park, Sheridan, and Sweetwater counties were more likely to be unrestrained. Yet, in the results of the fatal crash data model, vehicle occupants observed in Big Horn, Crook, Carbon, Sheridan, and Sweetwater counties were more likely to be unrestrained. Specifically, we can see inverse results of seat belt use rate in the crook county in two models. In addition, the former model’s results indicated that male occupants were more likely to be unbuckled whereas the occupant gender variable was not found to have an impact on seat belt use choice in the results of the latter model. Likewise, the day of the week variable (weekday versus weekend) was identified as an influential parameter only in the former model. Previous studies also confirmed the variability of occupant seat belt use policy compliance rates across the population (5, 76 , 77 ). Therefore, considering the variability of occupants across the population is a crucial factor when proposing mitigation strategies and seat belt use campaigns.

Wyoming has one of the lowest seat belt use rates in the United State. The findings of this study highlighted noteworthy trends in seat belt use habits in the state. This will aid policymakers in their decision-making processes with regard to the development and implementation of road safety programs aimed at increasing seat belt use rates. Several management measures including upgrading to primary seat belt laws, raising penalties for violating seat belt laws, and nighttime enforcement programs might be effective in reducing the seat belt non-compliance rate in Wyoming ( 78 ). Past studies have demonstrated strong evidence to support upgrading seat belt laws from secondary to primary laws by showing that primary seat belt law is more effective in lowering seat belt non-compliance rates compared with secondary laws ( 8 , 9 , 20 ). In addition, imposing higher fines along with standard enforcement has demonstrated significant impacts on seat belt use rates. Since the FARS data model revealed that vehicle occupants were more likely to be unrestrained during midnight, it could be helpful to implement nighttime seat belt law enforcement programs to increase seat belt use. Also, counties with thriving oil and gas industries exhibit higher seat belt non-use rates in the observational survey data model. Effective outreach and education interventions for those counties such as targeted educational programs to point out the benefit of seat belt use, and the risks associated with not using seat belts ought to be considered. It was also inferred that young occupants and impairment were positively associated with seat belt non-use. Combining law enforcement and media coverage is proved to be effective in increasing seat belt use rates for target populations such as men, teens, and young adults ( 78 ).

Finally, there were several limitations to this study worth discussing. First, the unobserved heterogeneity effects of the factors that contribute to seat belt use choice from the observational survey data were not examined. Incorporating random parameters into the model to capture those unobserved heterogeneity effects is an approach worth selecting ( 75 ). Furthermore, various spatial autocorrelation techniques, such as geographical weighted regression (GWR), spatial econometric modeling, and the Besag–York–Mollié structure have been employed to model traffic crashes ( 79 , 80 ). The spatial autocorrelation ought to be considered in investigating geographical factors that relate to seat belt use choice in future studies.

Footnotes

Author Contributions

The authors confirm contribution to the paper as follows: study conception and design: S. Nazneen, A. Farid, K. Ksaibati; data collection: S. Nazneen, A. Farid, K. Ksaibati; data analysis and interpretation of results: S. Nazneen, A. Farid, K. Ksaibati; draft manuscript preparation: S. Nazneen. All authors reviewed the results and approved the final version of the manuscript.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The authors would like to thank the Wyoming Department of Transportation (DOT) for sponsoring this study and for providing the seat belt use survey data. This work is part of Project #RS02220 funded by the Wyoming DOT. The authors would also appreciate the effort of the Mountain Plains Consortium for providing matching funds under grant 69A3551747108 (FAST Act).

ORCID iD

Khaled Ksaibati

All opinions, stated in this manuscript, are solely those of the authors. The subject matter, all figures, tables, and equations, not previously copyrighted by outside sources, are copyrighted by Wyoming DOT, the State of Wyoming, and the University of Wyoming. All rights reserved copyrighting in 2021.

References

Deutermann

Characteristics of Fatal Rollover Crashes. Report No. DOT HS 809 438. National Highway Traffic Safety Administration, U.S. Department of Transportation, Washington, D.C., 2002.

National Highway Traffic Safety Administration. Seat Belts. 2020. https://www.nhtsa.gov/risky-driving/seat-belts. Accessed March 7, 2021.

National Highway Traffic Safety Administration. Research Note: Seat Belt Use in 2019—Overall Results. Report No. DOT HS 812 875. National Highway Traffic Safety Administration, U.S. Department of Transportation, Washignton, D.C., 2019.

Insurance Institute for Highway Safety. Fatality Facts 2018: State by State. 2019. https://www.iihs.org/topics/fatality-statistics/detail/state-by-state. Accessed February 18, 2021.

Boakye

Khattak

Everett

Nambisan

Correlates of Front-Seat Passengers’ Non-Use of Seat belts at Night. Accident Analysis & Prevention, Vol. 130, 2019, pp. 30–37. https://doi.org/10.1016/j.aap.2018.04.006.

McCartt

Northrup

Factors Related to Seat Belt Use Among Fatally Injured Teenage Drivers. Journal of Safety Research, Vol. 35, No. 1, 2004, pp. 29–38. https://doi.org/10.1016/j.jsr.2003.09.016.

Lee

Monuteaux

Burghardt

Fleegler

Nigrovic

Meehan

Schutzman

Mannix

Motor Vehicle Crash Fatalities in States With Primary Versus Secondary Seat Belt Laws: A Time-Series Analysis. Annals of Internal Medicine, Vol. 163, No. 3, 2015, pp. 184–190. https://doi.org/10.7326/M14-2368.

Boakye

Nambisan

Seat belt Laws and Seat belt Use Among Front- and Rear-Seat Vehicle Occupants in Fatal Crashes in the United States. Case Studies on Transport Policy, Vol. 8, No. 3, 2020, pp. 1030–1037. https://doi.org/10.1016/j.cstp.2020.04.003.

Harper

Would Stronger Seat Belt Laws Reduce Motor Vehicle Crash Deaths? A Semi-Bayesian Analysis. Epidemiology, Vol. 30, No. 3, 2019, pp. 380–387. https://doi.org/10.1097/EDE.0000000000000990.

10.

Enriquez

Pickrell

T. M.

Seat Belt Use in 2018 – Overall Results. Traffic Safety Facts Research Note. Report No. DOT HS 812 662. National Highway Traffic Safety Administration, Washington, D.C., 2019.

11.

Nichols

J. L.

Tippetts

A. S.

Fell

J. C.

Eichelberger

A. H.

Haseltine

P. W.

The Effects of Primary Enforcement Laws and Fine Levels on Seat Belt Usage in the United States. Taylor & Francis, Vol. 15, No. 6, 2014, pp. 640–644. https://doi.org/10.1080/15389588.2013.857017.

12.

Pickrell

Liu

Occupant Restraint Use in 2013: Results From the National Occupant Protection Use Survey Controlled Intersection Study. US Department of Transportation, National Highway Traffic Safety Administration, Washington, D.C., 2015.

13.

Bhat

Beck

Bergen

Kresnow

Predictors of Rear Seat Belt Use Among US Adults, 2012. Journal of Safety Research, Vol. 53, 2015, pp. 103–106.

14.

Han

Non-Seat belt Use and Associated Factors Among Passengers. International Journal of Injury Control and Safety Promotion, Vol. 24, No. 2, 2017, pp. 251–255. https://doi.org/10.1080/17457300.2016.1170042.

15.

Cooper

Atkins

Gillen

Measuring the Impact of Passenger Restrictions on New Teenage Drivers. Accident Analysis & Prevention, Vol. 37, No. 1, 2005, pp. 19–23. https://doi.org/10.1016/J.AAP.2004.02.003.

16.

Williams

A. F.

Shabanova

V. I.

Situational Factors in Seat Belt Use by Teenage Drivers and Passengers. Traffic Injury Prevention, Vol. 3, No. 3, 2002, pp. 201–204. https://doi.org/10.1080/15389580213650.

17.

Şimşekoǧlu

Ö.

Lajunen

Environmental and Psychosocial Factors Affecting Seat Belt Use Among Turkish Front-Seat Occupants in Ankara: Two Observation Studies. Traffic Injury Prevention, Vol. 9, No. 3, 2008, pp. 264–267. https://doi.org/10.1080/15389580801966508.

18.

Nambisan

S. S.

Vasudevan

Is Seat Belt Usage by Front Seat Passengers Related to Seat Belt Usage by Their Drivers?

Journal of Safety Research, Vol. 38, No. 5, 2007, pp. 545–555. https://doi.org/10.1016/J.JSR.2007.06.002.

19.

Chakraborty

Singh

Savolainen

Gates

Examining Correlation and Trends in Seat belt Use Among Occupants of the Same Vehicle. Transportation Research Record: Journal of the Transportation Research Board, 2021. 2675: 288–298.

20.

Houston

D. J.

Richardson

L. E.

Kweon

Y.-J.

Kockelman

K. M.

Reagan

I. J.

McClafferty

J. A.

Berlin

S. P.

Hankey

J. M.

Driver Attitudes and Choices: Speed Limits, Seat Belt Use, and Drinking-and-Driving. Accident Analysis and Prevention, Vol. 45, No. 6, 2010, pp. 600–607. https://doi.org/10.5399/osu/jtrf.45.3.611.

21.

Lerner

Jehle

Billittier

IV Moscati

Connery

Stiller

The Influence of Demographic Factors on Seat belt Use by Adults Injured in Motor Vehicle Crashes. Accident Analysis & Prevention, Vol. 33, No. 5, 2001, pp. 659–662. https://doi.org/10.1016/S0001-4575(00)00080-4.

22.

Beck

Downs

Stevens

Sauber-Schatz

Rural and Urban Differences in Passenger-Vehicle–Occupant Deaths and Seat Belt Use Among Adults — United States, 2014. Morbidity and Mortality Weekly Report Surveillance Summaries, Vol. 66, No. 17, 2017, pp. 1–13. https://doi.org/10.15585/mmwr.ss6617a1.

23.

Briggs

N. C.

Lambert

E. W.

Goldzweig

I. A.

Levine

R. S.

Warren

R. C.

Driver and Passenger Seat belt Use Among US High School Students. American Journal of Preventive Medicine, Vol. 35, 2008, pp. 224–229.

24.

Shinar

Schechtman

Compton

Trends in Safe Driving Behaviors and in Relation to Trends in Health Maintenance Behaviors in the USA: 1985-1995. Accident Analysis and Prevention, Vol. 31, 1999, pp. 497–503.

25.

Glassbrenner

Carra

Nichols

Recent Estimates of Safety Belt Use. Journal of Safety Research, Vol. 35, No. 2, 2004. https://doi.org/10.1016/j.jsr.2004.03.010.

26.

Afghari

Imani

Papadimitriou

Gelder

Hezaveh

A. M.

Disentangling the effects of unobserved factors on seatbelt use choices in multi-occupant vehicles. Journal of Choice Modelling, Vol. 41, 2021. https://doi.org/10.1016/j.jocm.2021.100324.

27.

Nambisan

Boakye

Seat belt Use Among Vehicle Occupants in Fatal Crashes in the United States: Does Vehicle Type and Age Affect Injury Outcome?

Presented at 100th Annual Meeting of the Transportation Research Board, Washington, D.C., 2021.

28.

Nukenine

S. K. L.

Daniel

Impact of Belt Use by Rear-Seat Occupants on Injury Severity of Belted Drivers in New Jersey. Transportation Research Record: Journal of the Transportation Research Board, 2011. 2265: 91–99.

29.

Jermakian

J. S.

Weast

R. A.

Passenger Use of and Attitudes Toward Rear Seat Belts. Journal of Safety Research, Vol. 64, 2018, pp. 113–119. https://doi.org/10.1016/j.jsr.2017.12.006.

30.

Boyle

Lampkin

2007 Motor Vehicle Occupant Safety Survey Volume 2: Seat Belt Report. Report No. DOT HS 810 975. National Highway Traffic Safety Administration, U.S. Department of Transportation, Washington, D.C., 2008.

31.

Kidd

D. G.

McCartt

A. T.

Drivers’ Attitudes Toward Front or Rear Child Passenger Belt Use and Seat Belt Reminders at These Seating Positions. Traffic Injury Prevention, Vol. 15, No. 3, 2014, pp. 278–286. https://doi.org/10.1080/15389588.2013.810333.

32.

Eluru

Bhat

A Joint Econometric Analysis of Seat Belt Use and Crash-Related Injury Severity. Accident Analysis & Prevention, Vol. 39, No. 5, 2007, pp. 1037–1049. https://doi.org/10.1016/j.aap.2007.02.001.

33.

Chen

Zhang

Huang

Wang

Tarefder

Examining Driver Injury Severity Outcomes in Rural Non-Interstate Roadway Crashes Using a Hierarchical Ordered Logit Model. Accident Analysis & Prevention, Vol. 96, 2016, pp. 79–87. https://doi.org/10.1016/j.aap.2016.06.015.

34.

Huang

Abdel-Aty

Multilevel Data and Bayesian Analysis in Traffic Safety. Accident Analysis & Prevention, Vol. 42, 2010, pp. 1556–1565.

35.

Huang

Chor

Haque

Severity of Driver Injury and Vehicle Damage in Traffic Crashes at Intersections: A Bayesian Hierarchical Analysis. Accident Analysis & Prevention, Vol. 40, No. 1, 2008, pp. 45–54. https://doi.org/10.1016/j.aap.2007.04.002.

36.

Haque

Chin

Huang

Applying Bayesian Hierarchical Models to Examine Motorcycle Crashes at Signalized Intersections. Accident Analysis & Prevention, Vol. 42, No. 1, 2010, pp. 203–212. https://doi.org/10.1016/j.aap.2009.07.022.

37.

Ahmed

I. U.

Gaweesh

S. M.

Ahmed

M. M.

Exploration of Hazardous Material Truck Crashes on Wyoming’s Interstate Roads Using a Novel Hamiltonian Monte Carlo Markov Chain Bayesian Inference. Transportation Research Record: Journal of the Transportation Research Board, 2020. 2674: 661–675.

38.

Haq

Zlatkovic

Ksaibati

Assessment of Tire Failure Related Crashes and Injury Severity on a Mountainous Freeway: Bayesian Binary Logit Approach. Accident Analysis & Prevention, Vol. 145, 2020, p. 105693. https://doi.org/10.1016/j.aap.2020.105693.

39.

Afghari

Hezaveh

Haque

Cherry

A Home-Based Approach to Understanding Seat belt Use in Single-Occupant Vehicles in Tennessee: Application of a Latent Class Binary Logit Model. Accident Analysis & Prevention, Vol. 146, 2020, p. 105743. https://doi.org/10.1016/j.aap.2020.105743.

40.

Fernsler

Wyoming Seat Belt Survey. Wyoming State Government, Cheyenne, WY, 2019.

41.

Seat Belt Survey Regulation. Final Rule for Section 157 Surveys. 2000. https://icsw.nhtsa.gov/nhtsa/whatsup/tea21/GrantMan/HTML/19c_StBeltSurvyReg23CFR1340.html. Accessed April 17, 2022.

42.

National Highway Traffic Safety Administration. Traffic Safety Facts 2017 Data Older Population. U.S. Department of Transportation, Washington, D.C., 2019.

43.

Laramie County. Child Restraint Enforcement Guide. 2020. https://www.laramiecountywy.gov/_officials/Sheriff/Restraint.aspx. Accessed June 10, 2021.

44.

Ahmed

Franke

Ksaibati

Shinstine

Effects of Truck Traffic on Crash Injury Severity on Rural Highways in Wyoming Using Bayesian Binary Logit Models. Accident Analysis & Prevention, Vol. 117, 2018, pp. 106–113. https://doi.org/10.1016/j.aap.2018.04.011.

45.

Lemoine

Moving Beyond Noninformative Priors: Why and How to Choose Weakly Informative Priors in Bayesian Analyses. OIKOS, Vol. 128, No. 7, 2019, pp. 912–928. https://doi.org/10.1111/oik.05985.

46.

Brooks

Gelman

Jones

Meng

X.-L.

Handbook of Markov Chain Monte Carlo. Chemical Rubber Company (CRC) Press, Taylor & Francis Group, Boca Raton, FL, 2011.

47.

Bürkner

Brms: An R Package for Bayesian Multilevel Models Using Stan. Journal of Statistical Software, Vol. 80, No. 1, 2017, pp. 1–28. https://doi.org/10.18637/jss.v080.i01.

48.

McElreath

Statistical Rethinking: A Bayesian Course With Examples in R and Stan, 2nd ed. Chemical Rubber Company (CRC) Press, Taylor & Francis Group, Boca Raton, FL, 2020.

49.

Gelman

Carlin

Stern

Dunson

Vehtari

Rubin

Bayesian Data Analysis, 3rd ed. Chemical Rubber Company (CRC) Press, Taylor & Francis Group, Boca Raton, FL, 2014.

50.

Kutner

M. H.

Nachtsheim

C. J.

Neter

Applied Linear Statistical Models. McGraw-Hill/Irwin, New York, NY, 2005.

51.

Glen

Tolerance Level / Tolerance Statistics: Definition, Examples. Statistics How to. com., 2018. https://www.statisticshowto.com/tolerance-level-statistics/. Accessed June 10, 2021.

52.

Watanabe

A Widely Applicable Bayesian Information Criterion. Journal of Machine Learning Research, Vol. 14, 2013, pp. 867–897.

53.

Gabry

Simpson

Vehtari

Betancourt

Gelman

Visualization in Bayesian Workflow. Journal of the Royal Statistical Society: Series A (Statistics in Society), Vol. 182, No. 2, 2019, pp. 389–402. https://doi.org/10.1111/rssa.12378.

54.

Haq

M. T.

Zlatkovic

Ksaibati

Assessment of Commercial Truck Driver Injury Severity as a Result of Driving Actions. Transportation Research Record: Journal of the Transportation Research Board, 2021. 2675: 1707–1719.

55.

Haq

M. T.

Zlatkovic

Ksaibati

Investigating Occupant Injury Severity of Truck-Involved Crashes Based on Vehicle Types on a Mountainous Freeway: A Hierarchical Bayesian Random Intercept Approach. Accident Analysis & Prevention, Vol. 144, 2020, p. 105654. https://doi.org/10.1016/J.AAP.2020.105654.

56.

Gelman

Rubin

Inference From Iterative Simulation Using Multiple Sequences. Statistical Science, Vol. 7, No. 4, 1992, pp. 457–472. https://doi.org/10.1214/ss/1177011136.

57.

Wyoming Department of Workforce Services. Seat Belts & Worker Safety in Wyoming. Wyoming Department of Workforce Services, Cheyenne, WY, 2020.

58.

Retzer

Identifying and Preventing Oil and Gas Fatalities in Wyoming and US. Cheyenne, WY. 2017. Presented at the Wyoming Workforce Safety Summit. 2017.

59.

Kubas

Vachal

Oil County Traffic Safety Survey, 2012. North Dakota Department of Transportation. http://www.ugpti.org/pubs/pdf/DP253.pdf.

60.

Gkritza

Mannering

Mixed Logit Analysis of Safety-Belt Use in Single- and Multi-Occupant Vehicles. Accident Analysis & Prevention, Vol. 40, No. 2, 2008, pp. 443–451. https://doi.org/10.1016/j.aap.2007.07.013.

61.

Solomon

Chaudhary

Preusser

Daytime and Nighttime Seat Belt Use at Selected Sites in New Mexico. Final Report No. DOT-HS-810-705. U.S. Department of Transportation, National Highway Traffic Safety Administration, Washington, D.C., 2007.

62.

Lipovac

Tešić

Marić

Đerić

Self-Reported and Observed Seat Belt Use – A Case Study: Bosnia and Herzegovina. Accident Analysis & Prevention, Vol. 84, 2015, pp. 74–82. https://doi.org/10.1016/J.AAP.2015.08.010.

63.

Huang

Y. H.

Zhang

Murphy

Shi

Lin

Attitudes and Behavior of Chinese Drivers Regarding Seat belt Use. Accident Analysis & Prevention, Vol. 43, No. 3, 2011, pp. 889–897. https://doi.org/10.1016/J.AAP.2010.11.009.

64.

Marshall

W. E.

Ferenchak

N. N.

Assessing Equity and Urban/Rural Road Safety Disparities in the US. Journal of Urbanism: International Research on Placemaking and Urban Sustainability, Vol. 10, No. 4, 2017, pp. 422–441. https://doi.org/10.1080/17549175.2017.1310748.

65.

Schulte Gary

S. L.

Aultman-Hall

McCourt

Stamatiadis

Consideration of Driver Home County Prohibition and Alcohol-Related Vehicle Crashes. Accident Analysis and Prevention, Vol. 35, No. 5, 2003, pp. 641–648. https://doi.org/10.1016/S0001-4575(02)00042-8.

66.

2019 Seat Belt Survey Analysis In Wyoming. Wyoming Department of Transportation, 2019, https://www.dot.state.wy.us/files/live/sites/wydot/files/shared/Highway_Safety/Annual%20Report/2019%20Survey%20Analysis%20of%20Seat%20Belt%20Use%20in%20Wyoming.pdf

67.

Coulter

Senate Committee Advances Bill Allowing Law Enforcement to Stop Drivers Who Don’t Wear Seat Belts and Local News and Wyomingnews. Com. https://www.wyomingnews.com/news/local_news/senate-committee-advances-bill-allowing-law-enforcement-to-stop-drivers-who-dont-wear-seat-belts/article_8ebff856-c413-509a-9ba7-26e7a7ec266f.html. Accessed April 19, 2022.

68.

Wyoming Report on Traffic Crashes 2021. Wyoming Department of Transportation. 2022. https://www.dot.state.wy.us/files/live/sites/wydot/files/shared/Highway_Safety/_Crash%20Data/Publications/Report%20on%20Traffic%20Crashes/Report%20on%20Traffic%20Crashes%202021.pdf

69.

Rezapour

Ksaibati

Factors Impacting the Chaoice of Seatbelt Use,Accounting for Complexity of Traveler's Behavior. Future Transportation. Vol. 2, 2022, pp. 237-248. https://doi.org/10.3390/futuretrans2010012.

70.

Koo

A Guideline of Selecting and Reporting Intraclass Correlation Coefficients for Reliability Research. Journal of Chiropractic Medicine, Vol. 15, No. 2, 2016, pp. 155–163. https://doi.org/10.1016/j.jcm.2016.02.012.

71.

Vivoda

J. M.

Eby

D. W.

st. Louis

R. M.

Kostyniuk

L. P.

A Direct Observation Study of Nighttime Safety Belt Use in Indiana. Journal of Safety Research, Vol. 38, No. 4, 2007, pp. 423–429. https://doi.org/10.1016/J.JSR.2007.04.003.

72.

Chaudhary

N. K.

Alonge

Preusser

D. F.

Evaluation of the Reading, PA Nighttime Safety Belt Enforcement Campaign: September 2004. Journal of Safety Research, Vol. 36, No. 4, 2005, pp. 321–326. https://doi.org/10.1016/J.JSR.2005.06.009.

73.

Hezaveh

A. M.

Cherry

C. R.

Neighborhood-Level Factors Affecting Seat Belt Use. Accident Analysis & Prevention, Vol. 122, 2019, pp. 153–161. https://doi.org/10.1016/J.AAP.2018.10.005.

74.

Findley

Sanchez

Nye

Estimating the Effect of Standard Enforcement of a Rear Seat Belt Law for Rear Seat Fatality Prevention. Transportation Research Record: Journal of the Transportation Research Board, 2018. 2672: 67–77.

75.

Park

Lee

Development of Accident Frequency Models With Random Parameters on Interstate Roadway Segments With and Without Lighting Systems. Journal of Advanced Transportation, Vol. 2017, 2017, p. 8. https://doi.org/10.1155/2017/4740532.

76.

Tison

Williams

A. F.

Analyzing the First Years of the Click it or Ticket Mobilizations. Final Report No. DOT HS 811 232. National Highway Traffic Safety Administration, Washington, D.C., 2010, pp. 1–100.

77.

Salzberg

Yamada

Saibel

Moffat

Predicting Seat Belt Use in Fatal Motor Vehicle Crashes From Observation Surveys of Belt Use. Accident Analysis and Prevention, Vol. 34, No. 2, 2002, pp. 139–148. https://doi.org/10.1016/S0001-4575(01)00007-0.

78.

Venkatraman

Richard

C. M.

Magee

, Battelle Memorial Institute, Johnson

Countermeasures That Work: A Highway Safety Countermeasure Guide for State Highway Safety Offices, 10nth ed. Final Report No. DOT HS 813 097. National Highway Traffic Safety Administration, Washington, D.C., 2020.

79.

Boulieri

Liverani

de Hoogh

Blangiardo

A Space–Time Multivariate Bayesian Model to Analyse Road Traffic Accidents by Severity. Journal of the Royal Statistical Society Series A: Statistics in Society, Vol. 180, No. 1, 2017, 119–139.

80.

Pirdavani

Bellemans

Brijs

Wets

Application of Geographically Weighted Regression Technique in Spatial Analysis of Fatal and Injury Crashes. Journal of Transportation Engineering, Vol. 140, No. 8, 2014, p. 04014032. https://doi.org/10.1061/(ASCE)TE.1943-5436.0000680.