- Email: [email protected]
Personal, temporal and spatial characteristics of seriously injured crash-involved seat belt non-users in Hawaii
Accident Analysis and Prevention 35 (2003) 121–130
Personal, temporal and spatial characteristics of seriously injured crash-involved seat belt non-users in Hawaii Sungyop Kim a,1 , Karl Kim b,∗ a
Interdisciplinary Ph.D. Program in Urban Design and Planning, University of Washington, 410 Gould, Box 355740, Seattle, WA 98195-5740, USA b Department of Urban and Regional Planning, University of Hawaii at Manoa, 2424 Maile Way, #107, Honolulu, HI 96822, USA Received 17 July 2001; accepted 22 October 2001
Abstract The characteristics of crash-involved seat belt non-users in a high use state (Hawaii) are examined in order to better design enforcement and education programs. Using police crash report data over a 10-year period (1986–1995), we compare belted and unbelted drivers and front seat occupants, who were seriously injured in crashes, in terms of personal (age, gender, alcohol involvement, etc.) and crash characteristics (time, location, roadway factors, etc.). A logistic regression model combined with the spline method is used to analyze and categorize the salient differences between users and non-users. We find that unbelted occupants are more likely to be male, younger, unlicensed, intoxicated and driving pickup trucks versus other vehicles. Moreover, non-users are more likely than users to be involved in speed-related crashes in rural areas during the nighttime. Passengers are 70 times more likely to be unbelted if the driver is also unbelted than passengers of vehicles with belted drivers. While our general findings are similar to other seat belt studies, the contribution of this paper is in terms of a deeper understanding of the relative importance of various factors associated with non-use among seriously injured occupants as well as demonstrating a powerful methodology for analyzing safety problems entailing the categorization of various groups. While the former has implication for seat belt enforcement and education programs, the latter is relevant to a host of other research questions. © 2002 Elsevier Science Ltd. All rights reserved. Keywords: Seriously injured occupants; Seat belt use; z-Tests; Logistic regression; Spline method; Hawaii
1. Introduction The seat belt is one of the most effective devices for saving lives and reducing injuries to the occupants of vehicles involved in crashes. Despite mandatory seat belt laws and numerous safety programs that have been introduced to increase belt use nationwide, use rates remain at approximately 69% (NHTSA, 1998a). Seat belt use rates in the US are significantly lower than Canada, Australia and several Western European countries that have achieved compliance rates of more than 90% (NHTSA, 1997). Front seat occupants in Victoria, Australia had use rates of over 97% and rear seat passengers had use rates of 94% (Diamantopoulou et al., 1997). Even drivers in their 20s with the lowest use rates had a 95% compliance rate. The benefits of seat belt use have been well documented. Lap/shoulder belts reduce the risk of fatal injury to front seat passenger occupants by 45% and reduce the risk of ∗ Corresponding author. Tel.: +1-808-956-7381; fax: +1-808-956-6870. E-mail addresses: [email protected] (S. Kim), [email protected] (K. Kim). 1 Tel.: +1-206-221-2931.
moderate–critical injury by 50% (NHTSA, 1998a). The crash outcome data evaluation system (CODES) study, moreover, has demonstrated that the average inpatient charge for unbelted drivers involved in crashes is more than 55% higher than the average charge for belted drivers (NHTSA, 1996). Laws and their enforcement may affect seat belt use rates. At present, 11 states have primary enforcement laws that enable police to stop and cite motorists in violation of belt use laws. Secondary enforcement laws, in which motorists must violate some other traffic law before they can be stopped and cited, exist in 38 states. Secondary enforcement law states have use rates that are on the average 15% lower than primary enforcement states (NHTSA, 1998a). Enforcement practices, moreover, may also affect belt use. In Hawaii, for example, it has been shown that cumulative number of citations issued explained more than half of the monthly variation in seat belt use (Kim, 1991). In order to plan and implement effective programs for increasing seat belt use, more reliable data on belt use are needed. Most jurisdictions use observational studies to determine belt use rates. These methods typically involve observing and recording seat belt use of occupants as well
0001-4575/02/$ – see front matter © 2002 Elsevier Science Ltd. All rights reserved. PII: S 0 0 0 1 - 4 5 7 5 ( 0 1 ) 0 0 0 9 7 - 5
122
S. Kim, K. Kim / Accident Analysis and Prevention 35 (2003) 121–130
as vehicle type, location and roadway characteristics. Observational studies are costly to implement. It is difficult to achieve control over sampling. In addition to geographic sampling considerations, a variety of other temporal and roadway factors (road class, speed limit, volume, etc.) may also influence belt use. The problem, for example, of unbelted drivers on rural roads has been documented (Li et al., 1999). Belt use and attitudes towards the use of seat belts (Hamed and Easa, 1998) may also be influenced by trip purpose, distance from home, accident experience or other factors, such as the number of occupants in the vehicle. While some people may “belt up” all the time, there is also evidence of selective belt use (NHTSA, 1997). Another approach to gathering data on seat belt use is through attitudinal surveys. However, there are validity problems with self-reported use studies. Self-reported seat belt use is typically 12–25% higher than observed use (NHTSA, 1998b). As documented in earlier studies, there are significant differences between observed and self-reported seat belt use (Kim, 1999). One study found that 10% of motorists who described themselves as using belts “all the time” did not use belts at least once while driving during the past week (NHTSA, 1997). In spite of these difficulties associated with data collection, more information about seat belt non-users is needed. One way to better identify the characteristics and behaviors of non-users is to more closely examine police crash report data. However, there are problems and limitations with police crash data. Estimates of seat belt use from police crash data can be unreliable, particularly for minor crashes (O’Day, 1993; Hunter et al., 1993). Police crash investigators typically rely on interviews with drivers and passengers to determine whether or not belts were used in minor crashes (Figgers and Nash, 1989). In many states, admitting to police that drivers or occupants were not belted could lead to traffic fines or increases in auto insurance premiums. This is the so-called “lie factor” in traffic safety (Kim, 1999). According to the police data in Hawaii, over the period 1986–1995, 96.8% of all crash involved drivers and front seat passengers age 5 and over used seat belts. This rate is much higher than the annual average of 76.9% based on observational studies conducted during the same period. However, police crash data includes a valuable source of information for seat belt use studies. Crash data are readily available. Crash data are also populationbased, with detailed information on occupants and the crash environment. Police officers derive information on restraint use by examining the condition of the vehicle and the extent of the occupants’ injuries in serious crashes (O’Day, 1993; Hunter et al., 1993; Figgers and Nash, 1989). A previous study has demonstrated that severely injured motorists are less likely to “lie” to the police (Kim, 1999). Based on data consisting of 369 drivers for which crash and hospital records were linked in 1990, the police reported a belt use rate of 89.9% among the drivers with minor injuries. However, hospital
records indicated that only 58.3% of the drivers were actually belted. The discrepancy between police and hospital records narrowed significantly as injury levels increased. The police reported a belt use rate of 76.6% and hospital records showed a rate of 73.4% among drivers who had incapacitating or fatal injuries (Kim, 1999). By restricting our analysis of police data to seriously injured occupants, that is only those who received KABC0 scores of K (fatal injury) or A (incapacitating injury) and removing cases with B (non-incapacitating injury), C (possible injury) and 0 (no injury) scores, the reliability of belt use data increases greatly. Our analysis, therefore, focuses on the characteristics of unbelted drivers and front seat occupants involved in serious, injury producing crashes in which a police report has been filed. This is the critical population for which injury prevention strategies and law enforcement efforts should be targeted. The purpose of this paper is to examine the characteristics of seriously injured occupants seat belt non-users. There have been many studies on seat belt use. Those studies generally provide descriptive statistics on seat belt use rates by characteristics, such as age group, gender, race, vehicle type, etc. This paper provides models of seat belt non-use as a function of temporal and spatial factors as well as the personal characteristics of occupants (Kim et al., 1995). We begin by comparing the characteristics of users and non-users in terms of demographic variables and crash, roadway, temporal and spatial factors. We use a spline method to better control for the age variable and then build a logistic regression model for explaining the likelihood of being a non-user. We conclude with some observations on how relationships between age, demographic, crash and other variables might be useful in the design of programs to increase belt use.
2. Data and methods 2.1. Data sources Data for this paper were collected as part of the Hawaii Crash Outcome Data Evaluation System (CODES) project funded by the US Department of Transportation, National Highway Traffic Safety Administration (NHTSA). The CODES project facilitated the acquisition, cleaning, editing and linkage of crash and injury outcome (EMS transport and hospital records) files. The analysis is based on police reports of vehicle crashes over a 10-year period between 1986 and 1995. The quality of police-reported crash data in Hawaii is better than in many other states. Hawaii is a geographically isolated island state with only four counties and four police departments. Hawaii also has a standardized and centralized data collection system using a common crash report form. Specially trained accident investigators are used for serious injury crashes. The crash forms are sent to the State Department of Transportation for entry and analysis. We imported the crash files and
S. Kim, K. Kim / Accident Analysis and Prevention 35 (2003) 121–130
123
Fig. 1. Seat belt use rates in Hawaii (1986–1995).
used SAS (a statistical analysis software package) to build and analyze the data in this paper. Only occupants who had fatal (K) and incapacitating (A) injuries were included in the analysis. Over the period 1986–1995, 3457 drivers (age 15 or older) and 926 right front seat passengers (age 5 or older) had incapacitating or fatal injuries. We separately analyzed drivers and passengers in order to examine the association between driver seat belt use and passenger seat belt use. Interestingly, we found that the seat belt use rate among seriously injured occupants is close to the observed seat belt use rate in Hawaii. Fig. 1 shows the seat belt use rates for incapacitated and fatally injured occupants (from the police crash data) and the observed seat belt use rates over the period 1986–1995. Between 1988 and 1992, there are slight differences between rates based on police-reported data and the rates based on the observational studies. Police crash reports also include detailed data on occupants involved in crashes. Data on the age and gender of occupants, time and location are among the most reliable fields (Li et al., 1999). Personal, temporal and spatial attributes from the police crash data were used in our analysis. 2.2. Statistical analysis Descriptive statistics are used to characterize the differences between seriously injured seat belt users and non-users. Using z-tests, these two groups are compared in terms of personal, temporal and spatial variables. A spline method is employed to show the association between occupant age and seat belt use. The spline method uses simple polynomials to construct a smooth curve. This method has been extensively used in inteporlation, smoothing and approximation. Standardization and quadratic or cubic polynomial regression techniques are often used to examine a linear relationship between two continuous variables. However, these methods tend to provide only a rough estimation of
the relationship between variables. The spline method uses “knots” for smoothing the model curve. This improves the approximation of the model curve to observed values. Five-year “knots” in age were used with the spline methods to smooth the seat belt use rate curve. Using spline methods, we show the “smoothed” variation of expected seat belt use rates by occupant age. Then we used logistic regression to examine the probability of seat belt non-use as a function of age transformed by the spline method. Further discussion of the spline technique and regression analysis can be found in Schumaker (1993) and Eubank (1999). The logistic regression model is one of most widely used analytical tools in traffic safety research. It does not assume equal distribution of the dependent variable for each level of the independent variable, nor necessarily a linear relationship between the independent and dependent variables. Moreover, the logistic regression model does not assume a normal distribution of the variables. As such, it is a particularly robust model for various traffic safety analyses. Our research question focuses on the characteristics of seat belt non-users among seriously injured occupants in crashes. We use a binary variable to indicate seat belt use among occupants as the dependent variable with 1 = no seat belt use and 0 = seat belt use. The logistic regression model applies maximum likelihood estimation after transforming the categorical dependent variable into a logit variable. A logit is the log of the odds ratio (Appendix A). The logistic regression model estimates the probability of a certain event occurring, that is seat belt non-use in our study, defined as: 1 (1) 1 + e−U where Y is the estimated probability of seat belt non-use and U the linear regression equation:
Prob(Y = 1) =
U = β0 + β1 X1 + β2 X2 + · · · + βk Xk
(2)
where β is the parameter estimate, X the independent predictor and k the number of independent predictor.This linear
124
S. Kim, K. Kim / Accident Analysis and Prevention 35 (2003) 121–130
regression equation creates the logit or the log of the odds ratio: Y ln (3) = β 0 + β 1 X 1 + β 2 X 2 + · · · + β k Xk 1−Y The linear regression equation is the natural log of the probability of one outcome (seat belt non-use) over the other (seat belt use). We built two separate logistic regression models to explain seat belt use for drivers and front seat passengers as a function of various personal, temporal and spatial factors. The modeling process involves fitting various predictors related to seat belt non-use. The −2 log likelihood statistic, which has a χ 2 distribution under the null hypothesis that all regression coefficients of the model is zero, was used to assess the overall model fit. To assess the quality of the logistic model, the percentages of concordant, discordant and tied total pairs were examined. Total numbers of pairs are derived from multiplication of the frequencies of two response values. For all observation pairs, the procedure involves determining which cases are unbelted and which are belted. A pair is “concordant” if the observed unbelted case has a higher predicted event probability (of non-use) than does the observed belted case. A pair is “discordant” when the observed seat belt non-use case has a lower predicted event probability (of non-use) than does the observed belted case. Pairs that are neither concordant nor discordant are classified as “tied”. The two logistic models (for seat belt non-use for drivers and passengers) have significant −2 log likelihood statistic values at the 0.05 significance level and relatively high concordant rates.
3. Results 3.1. Personal characteristics of serious crash-involved seat belt non-users Table 1 shows the comparison of personal characteristics between two groups, seat belt non-users and users who were seriously injured in crashes. Drivers and front seat passengers are analyzed separately. The significance of the difference between seat belt non-users and users is identified by z-test statistics. Table 1 shows the proportional distribution of seat belt use by driver and passenger gender. Males are over-represented among unbelted drivers. More than 76% of unbelted drivers are males, while only 23.9% of unbelted drivers are females. Male drivers, however, represent 59.2% of belted drivers and female drivers represent 40.8% of belted drivers. Males are also over-represented among unbelted passengers. The differences are statistically significant at the 0.05 level. Males have about 1.87 and 1.83 times the odds of females to be unbelted for drivers and for passengers, respectively. Table 1 also shows a strong relationship between seat belt use and age. The proportion of younger age groups (age below 30)
among both unbelted drivers and passengers is greater than the proportion of unbelted older occupants. The mean age of belted drivers and passengers is greater than the mean age of unbelted occupants. These differences are statistically significant at the 0.05 level. Other studies have also confirmed that young drivers and males are less likely to use seat belts (NHTSA, 1995, 1997; Preusser et al., 1991). According to the NHTSA observational surveys, moreover, female belt use is 10% higher than male use (NHTSA, 1997). Fig. 2 shows a positive association between age and seat belt use among seriously injured drivers and front seat passengers. Logistic regression with the spline method was used to smooth the curves. Fig. 2 shows that the probability of seat belt use increases as driver age increases. Drivers between the age of 15 and 25 are the least likely to use a seat belt. From age 25 to 60, the graph reveals a constant increase in belt use. Passenger belt use also tends to increase with age. Occupants age 15 and young adults in their mid 20s are less likely to use seat belts. Children between age 5 and 6 may have had the lowest use rates because they are not covered by child safety seat laws and often do not sit in the front seat which is covered by the mandatory seat belt law. A national occupant protection use survey (NOPUS) conducted in 1998 found that pickup truck drivers (60.2%) and light truck drivers (67.1%) have lower use rates than drivers of vans (71.4%) and cars (72.4%) (NHTSA, 1998a). We found similar results in our analysis. Drivers of cars (79.8%) and vans (77.7%) have a higher seat belt use rate than drivers of pickup trucks (70.0%) or trucks (59.2%). As shown in Table 1, the proportion of unbelted occupants in pickup trucks is greater than in other vehicles. The test statistics are significant at the 0.05 level for all types of vehicles except for vans. License status of the driver has a strong association with seat belt use. The proportion of unlicensed drivers (16.3%) among unbelted drivers is greater than among belted drivers (10.5%). Unlicensed drivers are 1.46 times more likely than licensed drivers to be unbelted. Our police-reported data does not distinguish between drivers with revoked licenses and those with no license. Unlicensed drivers include drivers who had revoked licenses. High-risk behaviors, such as alcohol involvement and speeding have also been found to be associated with seat belt non-use (Preusser, 1988). Our analysis of seriously injured occupants found unbelted drivers have higher levels of alcohol involvement and speeding rates than belted drivers. Alcohol involved drivers have approximately 2.78 times the odds of non-alcohol involved drivers and speeding drivers have 2.18 times the odds of non-speeding drivers with respect to non-use of seat belts. Among belted drivers, only 6.7% were intoxicated. Whereas, approximately 25.4% of unbelted drivers were intoxicated. While 34.4% of unbelted drivers were involved in speed-related crashes, only 15% of belted drivers were speeding at the time of the crash. Similar results are observed among passengers. Passengers with
S. Kim, K. Kim / Accident Analysis and Prevention 35 (2003) 121–130
125
Table 1 Characteristics of seat belt non-users involved in serious crashes in Hawaii (1986–1995) Driver
Gender Male Female Age Children (age 5–9) Early teens (age 10–14) Late teens (age 15–19) Young adults (age 20–29) Adults (age 30–64) Seniors (age 65 and over)
Passenger
Unbelted (%) (N = 777)
Belted (%) (N = 2680)
Odds ratio
95% CL Low
High
76.1∗ 23.9∗
59.2∗ 40.8∗
1.87
1.56
2.24
1.12 1.50 0.75 0.52
Odds ratio
95% CL Low
High
52.3∗ 47.7∗
33.4∗ 66.6∗
1.83
1.33
2.51
4.0 4.5 25.4∗ 34.8∗ 27.9∗ 3.5∗
2.6 3.0 13.8∗ 25.5∗ 40.7∗ 14.3∗
1.39 1.34 1.74 1.40 0.63 0.27
0.60 0.61 1.19 1.00 0.45 0.12
3.23 2.94 2.55 1.96 0.89 0.58
27.7∗ 14.6
37.6∗ 20.0
73.1∗ 4.5 19.4∗ 3.0
80.4∗ 5.4 12.7∗ 1.5
0.73 0.86 1.46 1.66
0.51 0.41 0.97 0.61
1.05 1.80 2.20 4.56
11.2 29.9∗ 50.1∗ 8.8∗
Mean age S.D.
31.4∗ 13.2
36.9∗ 16.4
Vehicle type Passenger Van Pickup Truck
67.4∗ 5.0 23.6∗ 4.0∗
77.3∗ 5.1 16.0∗ 1.7∗
0.77 0.12 1.51 1.91
0.64 0.04 1.24 1.20
0.92 0.34 1.83 3.04
Driver license Unlicensed Licensed
16.3∗ 83.7∗
10.5∗ 89.5∗
1.46
1.16 1.00
1.83 1.00
Alcohol involvement (driver) Yes No
25.4∗ 74.6∗
6.7∗ 93.3∗
2.78
2.23
3.47
16.9∗ 83.1∗
4.7∗ 95.3∗
2.56
1.55
4.25
Speeding Yes No
34.4∗ 65.6∗
15.0∗ 85.0∗
2.18
1.82
2.62
27.4∗ 72.6∗
10.6∗ 89.4∗
2.27
1.54
3.35
82.0∗ 18.0∗
70.5∗ 29.5∗
1.69
1.14
2.50
23.4∗ 36.5∗
10.5∗ 30.0∗
1.99 1.26
1.33 0.90
2.98 1.74
36.5∗ 3.6∗
45.6∗ 13.9∗
0.74 0.28
0.54 0.13
1.03 0.61
57.6∗ 42.4∗
1.6∗ 98.4∗
8.52
4.47
16.22
Driver age Teen drivers (age 15–19) Young adult drivers (age 20–29) Adult drivers (age 30–64) Senior drivers (age 65 and over) Driver seat belt use Unbelted drivers Belted drivers ∗
1.42 1.77 0.88 0.76
Belted (%) (N = 725)
12.7 42.1∗ 40.9∗ 4.2∗
Driver gender Male drivers Female drivers
0.88 1.27 0.64 0.36
Unbelted (%) (N = 201)
Significant at the 0.05 level.
intoxicated drivers and in speeding vehicles are less likely to be belted. Passengers with intoxicated drivers are 2.56 times more likely than passengers with sober drivers to be unbelted. Also, passengers with speeding drivers are 2.27 times more likely than passengers with non-speeding drivers to be unbelted. These differences are statistically significant at the 0.05 level. We also examined the association between passenger belt use and driver behavior and characteristics in our analysis. Higher proportions of unbelted passengers were observed
among those riding with intoxicated and speeding drivers. When drivers were intoxicated and speeding, passengers were 2.6 times and 2.3 times more likely to be unbelted, respectively. Passengers of high-risk drivers’ vehicles are less likely to use seat belts. Driver gender and age are also significantly associated with passenger seat belt use. As shown in Table 1, there are proportionately more unbelted passengers when the driver was unbelted. Male drivers are 1.69 times more likely to have unbelted passengers than female drivers. Younger drivers, moreover, are also more likely to have
126
S. Kim, K. Kim / Accident Analysis and Prevention 35 (2003) 121–130
Fig. 2. Probability of seat belt use by age, among seriously injured occupants in Hawaii (1986–1995).
unbelted passengers. Among teen (age 15–19) and young adult (age 20–29) drivers, the proportion of unbelted passengers exceed that of belted passengers. Among adult (age 30–64) and senior (age 65 and over) drivers, however, the proportion of unbelted passengers is smaller than that of belted passengers. These differences are all significant at the 0.05 level. In Hawaii, the driver is responsible for a passenger’s seat belt use regardless of the passenger’s age. In order to measure the association between driver belt use and passenger belt use, we compared drivers and front seat passengers. We found 926 seriously injured right front seat passengers. Among these, 201 were unbelted and 725 were belted. Approximately 58% of passengers were unbelted when drivers were unbelted. However, only 1.6% of belted passengers were observed in the same cases. Unbelted drivers were 8.52 times more likely to have unbelted passengers than belted drivers. Among seriously injured occupants, there is a strong association between driver seat belt use and passenger seat belt use. 3.2. Temporal and spatial factors associated with seat belt non-use Table 2 shows several environmental factors related to belt use of occupants in serious crashes. We divided road type into two categories, freeway/highway and local road. There is no significant difference in belt use by road type for either drivers or passengers. There is, however, a significant difference between urban and rural areas in terms of belt use. In rural area drivers and passengers are 1.71 and 1.45 times more likely to be unbelted, respectively. Both drivers and passengers are less likely to be belted on weekend. Yet, time of the day has a strong association with seat belt use. Table 2 shows that during the nighttime
(6:00 p.m. to 6:00 a.m.) proportionally more both drivers and passengers are unbelted than during the daytime (6:00 a.m. to 6:00 p.m.), 67.8 and 55.7% respectively. These differences are all statistically significant at the 0.05 level. During the night, drivers are 2.24 times more likely to be unbelted than during the day, while passengers are 1.78 times more likely to be unbelted. Fig. 3 shows that as time approaches late night, belt use decreases. Fig. 4 also shows a similar typical pattern for passengers. On rainy days, there is a higher proportion of belted drivers (20.1%) than unbelted drivers, 20.1 versus 9.6% on clear days. This is significant at the 0.05 level. There is some evidence to suggest that drivers may be more cautious on rainy days or of other hazardous driving conditions (Hamed and Easa, 1998). 3.3. Logistic regression models explaining seat belt non-use among seriously injured drivers and passengers in crashes Because of the differences we observed between drivers and passengers, we built two separate logistic regression models for analysis of seat belt non-use: one for drivers and another for passengers. The dependent variable, seat belt non-use, is explained by a combination of several explanatory variables including personal characteristics and temporal and spatial factors. The parameter estimates, standard error, P-value, odds ratio and odds ratios with a 95% confidence interval (CI) are provided to explain the model. A log-odds ratio, exponentiation of parameter estimate, is used to express the magnitude of effect of a predictor in the model. Table 3 shows the logistic regression equation for unbelted crash involved drivers. As noted earlier, personal characteristics of the drivers are significantly associated with seat belt non-use among seriously injured drivers in crashes. As shown in Table 3, driver age is negatively related to non-use.
S. Kim, K. Kim / Accident Analysis and Prevention 35 (2003) 121–130
127
Table 2 Temporal and spatial factors related to seat belt use among seriously injured occupants in Hawaii (1986–1995) Driver
Passenger
Unbelted (%) (N = 777)
Belted (%) (N = 2680)
Odds ratio
95% CL Low
High
Road type Local road Freeway/highway
47.2 52.8
46.7 53.3
1.02
0.87
Urban/rural Rural Urban
53.4∗ 46.6∗
36.2∗ 63.8∗
1.71
Day of the week Monday Tuesday Wednesday Thursday Friday Saturday Sunday
12.5 13.6 13.0 11.7 14.9 18.3 16.0∗
13.3 12.4 13.7 12.5 17.4 17.1 13.6∗
67.8∗
42.9∗
32.2∗
57.1∗
90.4∗ 9.6∗
79.9∗ 20.1∗
Time of the day Night (6:00 p.m. to 6:00 a.m.) Day (6:00 a.m. to 6:00 p.m.) Weather Not rainy Rainy ∗
Unbelted (%) (N = 201)
Belted (%) (N = 725)
Odds ratio ‘
95% CL Low
High
1.19
45.3 54.7
45.8 54.2
0.98
0.72
1.35
1.46
2.01
49.3∗ 50.7∗
37.5∗ 62.5∗
1.45
1.06
1.99
0.95 1.08 0.95 0.94 0.86 1.07 1.16
0.74 0.86 0.75 0.74 0.69 0.87 0.93
1.20 1.37 1.21 1.21 1.08 1.31 1.44
13.9 8.5∗ 10.9 12.9 13.9 23.4∗ 16.4
12.4 13.9∗ 15.0 11.7 13.4 17.4∗ 16.1
1.11 0.64 0.75 1.09 1.03 1.33 1.01
0.70 0.37 0.46 0.68 0.66 0.91 0.66
1.75 1.09 1.21 1.75 1.63 1.94 1.54
2.24
1.89
2.65
55.7∗
37.5∗
1.78
1.29
2.44
44.3∗
62.5∗
85.8 14.2
89.3 10.7
0.78
0.49
1.24
2.03
1.57
2.63
Significant at the 0.05 level.
This indicates younger drivers are less likely to use seat belts. Male drivers are 1.6 times more likely than female drivers to be unbelted. Unlicensed drivers are 1.4 times more likely than licensed drivers to be unbelted. Intoxicated drivers are 2.6 times more likely to be unbelted and speeding drivers are 1.8 times more likely than non-speeding drivers to be unbelted. The
regression results that high-risk behaviors including alcohol use, speeding and seat belt non-use are all strongly and significantly associated. Vehicle type also affects likelihood of seat belt use. Pickup truck or truck drivers are 1.4 times more likely to be unbelted than drivers of passenger cars and vans. Temporal and spatial factors are also strong predictors of seat belt use. A driver at night is 1.8 times more likely to
Fig. 3. Odds ratio for seat belt non-use among seriously injured drivers by hour of the day in Hawaii (1986–1995).
128
S. Kim, K. Kim / Accident Analysis and Prevention 35 (2003) 121–130
Fig. 4. Odds ratio for seat belt non-use among seriously injured passengers by hour of the day in Hawaii (1986–1995).
be unbelted compared to driver in day. Drivers are 1.7 times more likely to be unbelted in rural areas than drivers in urban areas. Weather also is related to seat belt use. On rainy days, drivers are less likely to be unbelted. The log-odds ratio of not using seat belts on rainy days is 0.8.
Table 4 shows the log linear relationship between seriously injured front seat passenger belt use and various explanatory variables. Passenger age is significantly associated with belt use. A negative parameter estimate shows that non-users are most likely to be young and belt use increases
Table 3 Logistic regression analysis of seat belt non-use: driver modela,b Variable
Intercept Driver age Male Unlicensed Alcohol use Speeding Pickup truck/truck Nighttime (6:00 p.m. to 6:00 a.m.) Rural Rainy a b
Parameter −2.0644 −0.0126 0.4688 0.3344 0.9543 0.6113 0.3165 0.5845 0.5443 −0.2648
S.E.
0.1616 0.0033 0.0999 0.1257 0.1226 0.1027 0.1034 0.0947 0.0883 0.1437
Prob(X2 ) <0.0001 0.0001 <0.0001 0.0078 <0.0001 <0.0001 0.0022 <0.0001 <0.0001 0.0654
Odds ratio
0.987 1.598 1.397 2.597 1.843 1.372 1.794 1.723 0.767
95% CI Lower
Upper
0.981 1.314 1.092 2.042 1.507 1.121 1.490 1.450 0.579
0.994 1.944 1.787 3.302 2.253 1.681 2.160 2.049 1.017
Likelihood ratio χ 2 = 430.5751 (with 9 d.f., P ≤ 0.0001). Association of predicted probability and observed responses: concordant = 73.5%; discordant = 26.1%; tied = 0.4%.
Table 4 Logistic regression analysis of seat belt non-use: passenger modela,b,c Variable
Intercept Passenger age Male passenger Nighttime (6:00 p.m. to 6:00 a.m.) Rural Unbelted driver a
Parameter −2.0106 −0.0199 0.4093 0.6324 0.3658 4.2568
S.E.
0.3038 0.0065 0.2205 0.2216 0.2159 0.3448
Prob(X2 ) <0.0001 0.0023 0.0635 0.0043 0.0903 <0.0001
Odds ratio
0.980 1.506 1.882 1.442 70.583
For right front seat passenger only. Likelihood ratio χ 2 = 359.8888 (with 5 d.f., P ≤ 0.0001). c Association of predicted probability and observed responses: concordant = 85.2%; discordant = 14.4%; tied = 0.4%. b
95% CI Lower
Upper
0.968 0.977 1.219 0.944 35.911
0.993 2.320 2.906 2.201 138.732
S. Kim, K. Kim / Accident Analysis and Prevention 35 (2003) 121–130
as age increases. Male passengers are 1.5 times more likely than female passengers to be unbelted. However, the male passenger variable is not significant at the 0.05 level. Early we showed that there was a strong association between temporal circumstances and seat belt use. In this logistic regression model, nighttime (6:00 p.m. to 6:00 a.m.) is a statistically significant explanatory variable. The odds ratio of nighttime is 1.9. The odds ratio of rural area as compared to urban area is 1.4, however, it is not significant at the 0.05 level. The strongest predictor for passenger belt use is of driver belt use. Approximately 89.3% of the passengers were belted when drivers were belted and 90.9% of passengers were not belted when drivers were not belted. The logistic regression model shows that passengers of vehicles with unbelted drivers are 70.6 times more likely to be unbelted than passengers of vehicles with belted drivers.
4. Discussion This paper investigated the relationships between seat belt use and personal, temporal and spatial characteristics. However, our analysis is limited to the examination of seriously injured crash-involved drivers and front seat passengers. Since, seriously injured occupants are more likely to be unbelted, our analysis may have some selectivity bias if we were to apply our results to the entire motorist population. Seriously injured occupants, however, need special attention since they should be a target group in promoting roadway traffic safety. In order to achieve the greatest public health benefits, education and enforcement programs need to be focused on seriously injured unbelted motorists. Our findings are similar to previous studies regarding the associations between seat belt use and related factors. Despite the belief that unbelted drivers tend to be more involved in crashes than belted drivers, the average annual belt use rate of the population in our study is actually quite close to the annual observed belt use rates in Hawaii. The average annual rate from the crash data was slightly higher than the average annual observed seat belt use rate during the same period. This suggests that crash-involved occupants are not significantly less likely to use belts than the general population. Our study provides a comprehensive use analysis of belt use that takes into account specific personal and temporal and spatial characteristics. A strong association between seat belt use and various personal factors, such as age, gender, vehicle type, driver license status, alcohol intoxication and speeding was found. In particular, we found that occupant age is a strong factor that is associated with seat belt use. Both driver and passenger belt use increases as age increases as shown in Fig. 2. This suggests that education should particularly pay attention to young population group. Our study results show that drivers and passengers in pickup trucks and trucks are far more likely to be unbelted than those in passenger cars and vans. More attention should
129
be paid to the relationship between unlicensed drivers and seat belt non-use. Unlicensed drivers are 1.4 times more likely than licensed drivers to be unbelted as shown in Table 3. The results of our model show that risk-taking behaviors, such as speeding and alcohol are strongly associated with non-use. We feel that these associations have a synergistic effect and may intensify occupant injury levels in vehicle crashes. This would seem to dispel notions regarding risk compensation or that unbelted drivers are more cautious than belted ones. Preusser et al. (1991) found that unbelted drivers has 20% more traffic accidents on their records than belted drivers. They also found that personal injury from previous accidents does not strongly affect seat belt use positively. These facts suggest that enforcement of speeding, alcohol and belt use laws may have cumulative positive effects. Perhaps drivers cited for speeding should also be more routinely scrutinized for belt use or alcohol involvement. Perhaps, the lack of belt use is an indicator of other potential violations. An integrated approach to seat belt use and other risk taking behaviors needs to be introduced in education and law enforcement. The part time seat belt user problem has drawn attention amidst rising overall national seat belt use. According to a seat belt survey and telephone interview study done in New York in 1988, 26% of 300 belted drivers said they did not always wear seat belts (Preusser et al., 1991). It is important to consider driving environments including spatial and temporal circumstances affecting part time seat belt users. Our study shows that there is salient difference between daytime and nighttime in seat belt use as shown in Figs. 3 and 4. We also found that drivers and front seat passengers in rural areas are more likely to be unbelted than those in urban areas. Police enforcement and other strategies for increasing seat belt use need to be focused on rural areas and during the nighttime as compared to the daytime. We also found a strong correlation between driver and passenger seat belt use. Overall, unbelted passengers have similar personal characteristics as the drivers of the vehicles they ride with. Drivers and front seat passengers tend to be of similar age. Yet, we need to know look more closely at how driver’s seat belt use affects passenger’s seat belt use.
5. Conclusions The analysis reveals that the characteristics of seat belt non-users can not only be identified, but also used to develop a statistical model for understanding the relative effects of age, gender, risk-taking behaviors and other crash variables. Our study, therefore, contributes to a more detailed understanding of the complex relationships between driver, vehicle, roadway and environmental factors and seat belt non-use among seriously injured occupants. We think that the logistic regression technique is particularly robust in that it allows us to handle categorical data and to compare
130
S. Kim, K. Kim / Accident Analysis and Prevention 35 (2003) 121–130
the relative odds or likelihood of being in a particular cross-classified category. The spline method is an enhancement to the technique, which increases the predictive power of the overall model. We think that the technique can be applied to many other traffic safety research problems. There are some limitations to our study. We hesitate to transfer our results to all drivers in all locales in part because we focused more narrowly on crash-involved, seriously injured motorists. They may not, necessarily, be identical to the overall motoring public. For instance, the proportion of drivers who are intoxicated or commit speeding offences in our analysis may be higher than the average driver because drunk drivers are more likely to be seriously injured in crashes. We need additional analysis of less seriously injured as well as unbelted occupants involved in minor crashes. Our study demonstrates that police-reported crash data are a useful resource in investigating seat belt use. However, the crash data does not contain strong potential factors that may affect seat belt use. Those factors include, income, education level, marital status, occupation, presence of children in household, detailed vehicle type information (sports car, sedan, coupe, etc.), vehicle crash experience, road environment (speed limit, congestion level, etc.), belief in the effectiveness of seat belt and the level of perceived injury risk. Perhaps future analysis will focus on citation data linked to crash data. There is a need to combine analysis of observational, crash, citation and attitudinal data for a single jurisdiction, over the same period of time. We think that Hawaii is the ideal place to examine these different sources of data on non-use. Hopefully our future research endeavors will enable us to do so. Appendix A The odds ratio, that is the odds of being an unbelted occupant, relative to another associated factor (such as gender) can be expressed as follows: OR =
a/(a + b) , c/(c + d)
where OR is the odds ratio, a the unbelted male frequency, b the belted male frequency, c the unbelted female frequency and d the belted female frequency. To calculate the 95% CI using Woolf’s method, 1 1 1 1 S.E.(ln OR) = + + + , a b c d ln (OR) = ±1.96 × S.E.(ln OR)
The confidence limit (CL) can be derived by taking antilogs (source: Kahn and Sempos (1989)). References Diamantopoulou, K., Dyte, D., Cameron, M., 1997. Seat belt wearing rates in Victoria (1994). Monash University Accident Research Centre-Report #89. Eubank, R.L., 1999. Non-Parametric Regression and Spline Smoothing. Marcel Dekker, New York. Figgers, R., Nash, C.E., 1989. Accurately reporting belt use in automobile crashes. The Police Chief 07/89, 50. Hamed, M.M., Easa, S.M., 1998. Ordered probability modeling of seat belt usage. J. Transport. Eng. ASCE 124 (3), 271–276. Hunter, W.W., Stewart, J.R., Stutts, J.C., Rodgman, E.A., 1993. Observed and self-reported seat belt wearing as related to prior accidents and convictions. Accid. Anal. Prev. 25, 545–554. Kahn, H., Sempos, C., 1989. Statistical Methods in Epidemiology. Oxford University Press, Oxford. Kim, K., 1991. Effects of enforcement on seat belt use in Hawaii. Transport. Res. Rec. 1325, 51–56. Kim, K., 1999. The lie factor in traffic safety: comparison of police and hospital reporting of seat belt and alcohol use in Hawaii. Transport. Res. Rec. 1665, 141–146. Kim, K., Nitz, L., Richardson, J., Li, L., 1995. Personal and behavioral predictors of automobile crash and injury severity. Accid. Anal. Prev. 27, 469–481. Li, L., Kim, K., Nitz, L., 1999. Predictors of seat belt use among crash-involved drivers and front seat passengers: adjusting for over-reporting. Accid. Anal. Prev. 31, 631–638. National Highway Traffic Safety Administration (NHTSA), US Department of Transportation, 1995. Research Note: National Occupant Protection Use Survey. Controlled Intersection Study. National Highway Traffic Safety Administration (NHTSA), US Department of Transportation, 1996. Report to Congress: Benefits of Safety Belts and Motorcycle Helmets. National Highway Traffic Safety Administration (NHTSA), US Department of Transportation, 1997. Presidential Initiative for Increasing Seat Belt Use Nationwide. National Highway Traffic Safety Administration (NHTSA), US Department of Transportation, 1998a. Buckle Up America: First Report to Congress. The Presidential Initiative for Increasing Seat Belt Use Nationwide. National Highway Traffic Safety Administration (NHTSA), US Department of Transportation, 1998b. Research Note: Observed Seat belt Use in 1998. O’Day, J., 1993. Accident Data Quality: A Synthesis of Highway Practice, National Cooperative Highway Research Program 192, Transportation Research Board. National Academy Press, Washington, DC. Preusser, D.F., 1988. Belt use by high-risk drivers before and after New York’s seat belt use law. Accid. Anal. Prev. 20, 245–250. Preusser, D.F., Williams, A.F., Lund, A.K., 1991. Characteristics of belted and unbelted drivers. Accid. Anal. Prev. 23, 475–482. Schumaker, L.L., 1993. Spline Functions: Basic Theory. Krieger Publishing Co., Melbourne.
Personal, temporal and spatial characteristics of seriously injured crash-involved seat belt non-users in Hawaii Sungyop Kim a,1 , Karl Kim b,∗ a
Interdisciplinary Ph.D. Program in Urban Design and Planning, University of Washington, 410 Gould, Box 355740, Seattle, WA 98195-5740, USA b Department of Urban and Regional Planning, University of Hawaii at Manoa, 2424 Maile Way, #107, Honolulu, HI 96822, USA Received 17 July 2001; accepted 22 October 2001
Abstract The characteristics of crash-involved seat belt non-users in a high use state (Hawaii) are examined in order to better design enforcement and education programs. Using police crash report data over a 10-year period (1986–1995), we compare belted and unbelted drivers and front seat occupants, who were seriously injured in crashes, in terms of personal (age, gender, alcohol involvement, etc.) and crash characteristics (time, location, roadway factors, etc.). A logistic regression model combined with the spline method is used to analyze and categorize the salient differences between users and non-users. We find that unbelted occupants are more likely to be male, younger, unlicensed, intoxicated and driving pickup trucks versus other vehicles. Moreover, non-users are more likely than users to be involved in speed-related crashes in rural areas during the nighttime. Passengers are 70 times more likely to be unbelted if the driver is also unbelted than passengers of vehicles with belted drivers. While our general findings are similar to other seat belt studies, the contribution of this paper is in terms of a deeper understanding of the relative importance of various factors associated with non-use among seriously injured occupants as well as demonstrating a powerful methodology for analyzing safety problems entailing the categorization of various groups. While the former has implication for seat belt enforcement and education programs, the latter is relevant to a host of other research questions. © 2002 Elsevier Science Ltd. All rights reserved. Keywords: Seriously injured occupants; Seat belt use; z-Tests; Logistic regression; Spline method; Hawaii
1. Introduction The seat belt is one of the most effective devices for saving lives and reducing injuries to the occupants of vehicles involved in crashes. Despite mandatory seat belt laws and numerous safety programs that have been introduced to increase belt use nationwide, use rates remain at approximately 69% (NHTSA, 1998a). Seat belt use rates in the US are significantly lower than Canada, Australia and several Western European countries that have achieved compliance rates of more than 90% (NHTSA, 1997). Front seat occupants in Victoria, Australia had use rates of over 97% and rear seat passengers had use rates of 94% (Diamantopoulou et al., 1997). Even drivers in their 20s with the lowest use rates had a 95% compliance rate. The benefits of seat belt use have been well documented. Lap/shoulder belts reduce the risk of fatal injury to front seat passenger occupants by 45% and reduce the risk of ∗ Corresponding author. Tel.: +1-808-956-7381; fax: +1-808-956-6870. E-mail addresses: [email protected] (S. Kim), [email protected] (K. Kim). 1 Tel.: +1-206-221-2931.
moderate–critical injury by 50% (NHTSA, 1998a). The crash outcome data evaluation system (CODES) study, moreover, has demonstrated that the average inpatient charge for unbelted drivers involved in crashes is more than 55% higher than the average charge for belted drivers (NHTSA, 1996). Laws and their enforcement may affect seat belt use rates. At present, 11 states have primary enforcement laws that enable police to stop and cite motorists in violation of belt use laws. Secondary enforcement laws, in which motorists must violate some other traffic law before they can be stopped and cited, exist in 38 states. Secondary enforcement law states have use rates that are on the average 15% lower than primary enforcement states (NHTSA, 1998a). Enforcement practices, moreover, may also affect belt use. In Hawaii, for example, it has been shown that cumulative number of citations issued explained more than half of the monthly variation in seat belt use (Kim, 1991). In order to plan and implement effective programs for increasing seat belt use, more reliable data on belt use are needed. Most jurisdictions use observational studies to determine belt use rates. These methods typically involve observing and recording seat belt use of occupants as well
0001-4575/02/$ – see front matter © 2002 Elsevier Science Ltd. All rights reserved. PII: S 0 0 0 1 - 4 5 7 5 ( 0 1 ) 0 0 0 9 7 - 5
122
S. Kim, K. Kim / Accident Analysis and Prevention 35 (2003) 121–130
as vehicle type, location and roadway characteristics. Observational studies are costly to implement. It is difficult to achieve control over sampling. In addition to geographic sampling considerations, a variety of other temporal and roadway factors (road class, speed limit, volume, etc.) may also influence belt use. The problem, for example, of unbelted drivers on rural roads has been documented (Li et al., 1999). Belt use and attitudes towards the use of seat belts (Hamed and Easa, 1998) may also be influenced by trip purpose, distance from home, accident experience or other factors, such as the number of occupants in the vehicle. While some people may “belt up” all the time, there is also evidence of selective belt use (NHTSA, 1997). Another approach to gathering data on seat belt use is through attitudinal surveys. However, there are validity problems with self-reported use studies. Self-reported seat belt use is typically 12–25% higher than observed use (NHTSA, 1998b). As documented in earlier studies, there are significant differences between observed and self-reported seat belt use (Kim, 1999). One study found that 10% of motorists who described themselves as using belts “all the time” did not use belts at least once while driving during the past week (NHTSA, 1997). In spite of these difficulties associated with data collection, more information about seat belt non-users is needed. One way to better identify the characteristics and behaviors of non-users is to more closely examine police crash report data. However, there are problems and limitations with police crash data. Estimates of seat belt use from police crash data can be unreliable, particularly for minor crashes (O’Day, 1993; Hunter et al., 1993). Police crash investigators typically rely on interviews with drivers and passengers to determine whether or not belts were used in minor crashes (Figgers and Nash, 1989). In many states, admitting to police that drivers or occupants were not belted could lead to traffic fines or increases in auto insurance premiums. This is the so-called “lie factor” in traffic safety (Kim, 1999). According to the police data in Hawaii, over the period 1986–1995, 96.8% of all crash involved drivers and front seat passengers age 5 and over used seat belts. This rate is much higher than the annual average of 76.9% based on observational studies conducted during the same period. However, police crash data includes a valuable source of information for seat belt use studies. Crash data are readily available. Crash data are also populationbased, with detailed information on occupants and the crash environment. Police officers derive information on restraint use by examining the condition of the vehicle and the extent of the occupants’ injuries in serious crashes (O’Day, 1993; Hunter et al., 1993; Figgers and Nash, 1989). A previous study has demonstrated that severely injured motorists are less likely to “lie” to the police (Kim, 1999). Based on data consisting of 369 drivers for which crash and hospital records were linked in 1990, the police reported a belt use rate of 89.9% among the drivers with minor injuries. However, hospital
records indicated that only 58.3% of the drivers were actually belted. The discrepancy between police and hospital records narrowed significantly as injury levels increased. The police reported a belt use rate of 76.6% and hospital records showed a rate of 73.4% among drivers who had incapacitating or fatal injuries (Kim, 1999). By restricting our analysis of police data to seriously injured occupants, that is only those who received KABC0 scores of K (fatal injury) or A (incapacitating injury) and removing cases with B (non-incapacitating injury), C (possible injury) and 0 (no injury) scores, the reliability of belt use data increases greatly. Our analysis, therefore, focuses on the characteristics of unbelted drivers and front seat occupants involved in serious, injury producing crashes in which a police report has been filed. This is the critical population for which injury prevention strategies and law enforcement efforts should be targeted. The purpose of this paper is to examine the characteristics of seriously injured occupants seat belt non-users. There have been many studies on seat belt use. Those studies generally provide descriptive statistics on seat belt use rates by characteristics, such as age group, gender, race, vehicle type, etc. This paper provides models of seat belt non-use as a function of temporal and spatial factors as well as the personal characteristics of occupants (Kim et al., 1995). We begin by comparing the characteristics of users and non-users in terms of demographic variables and crash, roadway, temporal and spatial factors. We use a spline method to better control for the age variable and then build a logistic regression model for explaining the likelihood of being a non-user. We conclude with some observations on how relationships between age, demographic, crash and other variables might be useful in the design of programs to increase belt use.
2. Data and methods 2.1. Data sources Data for this paper were collected as part of the Hawaii Crash Outcome Data Evaluation System (CODES) project funded by the US Department of Transportation, National Highway Traffic Safety Administration (NHTSA). The CODES project facilitated the acquisition, cleaning, editing and linkage of crash and injury outcome (EMS transport and hospital records) files. The analysis is based on police reports of vehicle crashes over a 10-year period between 1986 and 1995. The quality of police-reported crash data in Hawaii is better than in many other states. Hawaii is a geographically isolated island state with only four counties and four police departments. Hawaii also has a standardized and centralized data collection system using a common crash report form. Specially trained accident investigators are used for serious injury crashes. The crash forms are sent to the State Department of Transportation for entry and analysis. We imported the crash files and
S. Kim, K. Kim / Accident Analysis and Prevention 35 (2003) 121–130
123
Fig. 1. Seat belt use rates in Hawaii (1986–1995).
used SAS (a statistical analysis software package) to build and analyze the data in this paper. Only occupants who had fatal (K) and incapacitating (A) injuries were included in the analysis. Over the period 1986–1995, 3457 drivers (age 15 or older) and 926 right front seat passengers (age 5 or older) had incapacitating or fatal injuries. We separately analyzed drivers and passengers in order to examine the association between driver seat belt use and passenger seat belt use. Interestingly, we found that the seat belt use rate among seriously injured occupants is close to the observed seat belt use rate in Hawaii. Fig. 1 shows the seat belt use rates for incapacitated and fatally injured occupants (from the police crash data) and the observed seat belt use rates over the period 1986–1995. Between 1988 and 1992, there are slight differences between rates based on police-reported data and the rates based on the observational studies. Police crash reports also include detailed data on occupants involved in crashes. Data on the age and gender of occupants, time and location are among the most reliable fields (Li et al., 1999). Personal, temporal and spatial attributes from the police crash data were used in our analysis. 2.2. Statistical analysis Descriptive statistics are used to characterize the differences between seriously injured seat belt users and non-users. Using z-tests, these two groups are compared in terms of personal, temporal and spatial variables. A spline method is employed to show the association between occupant age and seat belt use. The spline method uses simple polynomials to construct a smooth curve. This method has been extensively used in inteporlation, smoothing and approximation. Standardization and quadratic or cubic polynomial regression techniques are often used to examine a linear relationship between two continuous variables. However, these methods tend to provide only a rough estimation of
the relationship between variables. The spline method uses “knots” for smoothing the model curve. This improves the approximation of the model curve to observed values. Five-year “knots” in age were used with the spline methods to smooth the seat belt use rate curve. Using spline methods, we show the “smoothed” variation of expected seat belt use rates by occupant age. Then we used logistic regression to examine the probability of seat belt non-use as a function of age transformed by the spline method. Further discussion of the spline technique and regression analysis can be found in Schumaker (1993) and Eubank (1999). The logistic regression model is one of most widely used analytical tools in traffic safety research. It does not assume equal distribution of the dependent variable for each level of the independent variable, nor necessarily a linear relationship between the independent and dependent variables. Moreover, the logistic regression model does not assume a normal distribution of the variables. As such, it is a particularly robust model for various traffic safety analyses. Our research question focuses on the characteristics of seat belt non-users among seriously injured occupants in crashes. We use a binary variable to indicate seat belt use among occupants as the dependent variable with 1 = no seat belt use and 0 = seat belt use. The logistic regression model applies maximum likelihood estimation after transforming the categorical dependent variable into a logit variable. A logit is the log of the odds ratio (Appendix A). The logistic regression model estimates the probability of a certain event occurring, that is seat belt non-use in our study, defined as: 1 (1) 1 + e−U where Y is the estimated probability of seat belt non-use and U the linear regression equation:
Prob(Y = 1) =
U = β0 + β1 X1 + β2 X2 + · · · + βk Xk
(2)
where β is the parameter estimate, X the independent predictor and k the number of independent predictor.This linear
124
S. Kim, K. Kim / Accident Analysis and Prevention 35 (2003) 121–130
regression equation creates the logit or the log of the odds ratio: Y ln (3) = β 0 + β 1 X 1 + β 2 X 2 + · · · + β k Xk 1−Y The linear regression equation is the natural log of the probability of one outcome (seat belt non-use) over the other (seat belt use). We built two separate logistic regression models to explain seat belt use for drivers and front seat passengers as a function of various personal, temporal and spatial factors. The modeling process involves fitting various predictors related to seat belt non-use. The −2 log likelihood statistic, which has a χ 2 distribution under the null hypothesis that all regression coefficients of the model is zero, was used to assess the overall model fit. To assess the quality of the logistic model, the percentages of concordant, discordant and tied total pairs were examined. Total numbers of pairs are derived from multiplication of the frequencies of two response values. For all observation pairs, the procedure involves determining which cases are unbelted and which are belted. A pair is “concordant” if the observed unbelted case has a higher predicted event probability (of non-use) than does the observed belted case. A pair is “discordant” when the observed seat belt non-use case has a lower predicted event probability (of non-use) than does the observed belted case. Pairs that are neither concordant nor discordant are classified as “tied”. The two logistic models (for seat belt non-use for drivers and passengers) have significant −2 log likelihood statistic values at the 0.05 significance level and relatively high concordant rates.
3. Results 3.1. Personal characteristics of serious crash-involved seat belt non-users Table 1 shows the comparison of personal characteristics between two groups, seat belt non-users and users who were seriously injured in crashes. Drivers and front seat passengers are analyzed separately. The significance of the difference between seat belt non-users and users is identified by z-test statistics. Table 1 shows the proportional distribution of seat belt use by driver and passenger gender. Males are over-represented among unbelted drivers. More than 76% of unbelted drivers are males, while only 23.9% of unbelted drivers are females. Male drivers, however, represent 59.2% of belted drivers and female drivers represent 40.8% of belted drivers. Males are also over-represented among unbelted passengers. The differences are statistically significant at the 0.05 level. Males have about 1.87 and 1.83 times the odds of females to be unbelted for drivers and for passengers, respectively. Table 1 also shows a strong relationship between seat belt use and age. The proportion of younger age groups (age below 30)
among both unbelted drivers and passengers is greater than the proportion of unbelted older occupants. The mean age of belted drivers and passengers is greater than the mean age of unbelted occupants. These differences are statistically significant at the 0.05 level. Other studies have also confirmed that young drivers and males are less likely to use seat belts (NHTSA, 1995, 1997; Preusser et al., 1991). According to the NHTSA observational surveys, moreover, female belt use is 10% higher than male use (NHTSA, 1997). Fig. 2 shows a positive association between age and seat belt use among seriously injured drivers and front seat passengers. Logistic regression with the spline method was used to smooth the curves. Fig. 2 shows that the probability of seat belt use increases as driver age increases. Drivers between the age of 15 and 25 are the least likely to use a seat belt. From age 25 to 60, the graph reveals a constant increase in belt use. Passenger belt use also tends to increase with age. Occupants age 15 and young adults in their mid 20s are less likely to use seat belts. Children between age 5 and 6 may have had the lowest use rates because they are not covered by child safety seat laws and often do not sit in the front seat which is covered by the mandatory seat belt law. A national occupant protection use survey (NOPUS) conducted in 1998 found that pickup truck drivers (60.2%) and light truck drivers (67.1%) have lower use rates than drivers of vans (71.4%) and cars (72.4%) (NHTSA, 1998a). We found similar results in our analysis. Drivers of cars (79.8%) and vans (77.7%) have a higher seat belt use rate than drivers of pickup trucks (70.0%) or trucks (59.2%). As shown in Table 1, the proportion of unbelted occupants in pickup trucks is greater than in other vehicles. The test statistics are significant at the 0.05 level for all types of vehicles except for vans. License status of the driver has a strong association with seat belt use. The proportion of unlicensed drivers (16.3%) among unbelted drivers is greater than among belted drivers (10.5%). Unlicensed drivers are 1.46 times more likely than licensed drivers to be unbelted. Our police-reported data does not distinguish between drivers with revoked licenses and those with no license. Unlicensed drivers include drivers who had revoked licenses. High-risk behaviors, such as alcohol involvement and speeding have also been found to be associated with seat belt non-use (Preusser, 1988). Our analysis of seriously injured occupants found unbelted drivers have higher levels of alcohol involvement and speeding rates than belted drivers. Alcohol involved drivers have approximately 2.78 times the odds of non-alcohol involved drivers and speeding drivers have 2.18 times the odds of non-speeding drivers with respect to non-use of seat belts. Among belted drivers, only 6.7% were intoxicated. Whereas, approximately 25.4% of unbelted drivers were intoxicated. While 34.4% of unbelted drivers were involved in speed-related crashes, only 15% of belted drivers were speeding at the time of the crash. Similar results are observed among passengers. Passengers with
S. Kim, K. Kim / Accident Analysis and Prevention 35 (2003) 121–130
125
Table 1 Characteristics of seat belt non-users involved in serious crashes in Hawaii (1986–1995) Driver
Gender Male Female Age Children (age 5–9) Early teens (age 10–14) Late teens (age 15–19) Young adults (age 20–29) Adults (age 30–64) Seniors (age 65 and over)
Passenger
Unbelted (%) (N = 777)
Belted (%) (N = 2680)
Odds ratio
95% CL Low
High
76.1∗ 23.9∗
59.2∗ 40.8∗
1.87
1.56
2.24
1.12 1.50 0.75 0.52
Odds ratio
95% CL Low
High
52.3∗ 47.7∗
33.4∗ 66.6∗
1.83
1.33
2.51
4.0 4.5 25.4∗ 34.8∗ 27.9∗ 3.5∗
2.6 3.0 13.8∗ 25.5∗ 40.7∗ 14.3∗
1.39 1.34 1.74 1.40 0.63 0.27
0.60 0.61 1.19 1.00 0.45 0.12
3.23 2.94 2.55 1.96 0.89 0.58
27.7∗ 14.6
37.6∗ 20.0
73.1∗ 4.5 19.4∗ 3.0
80.4∗ 5.4 12.7∗ 1.5
0.73 0.86 1.46 1.66
0.51 0.41 0.97 0.61
1.05 1.80 2.20 4.56
11.2 29.9∗ 50.1∗ 8.8∗
Mean age S.D.
31.4∗ 13.2
36.9∗ 16.4
Vehicle type Passenger Van Pickup Truck
67.4∗ 5.0 23.6∗ 4.0∗
77.3∗ 5.1 16.0∗ 1.7∗
0.77 0.12 1.51 1.91
0.64 0.04 1.24 1.20
0.92 0.34 1.83 3.04
Driver license Unlicensed Licensed
16.3∗ 83.7∗
10.5∗ 89.5∗
1.46
1.16 1.00
1.83 1.00
Alcohol involvement (driver) Yes No
25.4∗ 74.6∗
6.7∗ 93.3∗
2.78
2.23
3.47
16.9∗ 83.1∗
4.7∗ 95.3∗
2.56
1.55
4.25
Speeding Yes No
34.4∗ 65.6∗
15.0∗ 85.0∗
2.18
1.82
2.62
27.4∗ 72.6∗
10.6∗ 89.4∗
2.27
1.54
3.35
82.0∗ 18.0∗
70.5∗ 29.5∗
1.69
1.14
2.50
23.4∗ 36.5∗
10.5∗ 30.0∗
1.99 1.26
1.33 0.90
2.98 1.74
36.5∗ 3.6∗
45.6∗ 13.9∗
0.74 0.28
0.54 0.13
1.03 0.61
57.6∗ 42.4∗
1.6∗ 98.4∗
8.52
4.47
16.22
Driver age Teen drivers (age 15–19) Young adult drivers (age 20–29) Adult drivers (age 30–64) Senior drivers (age 65 and over) Driver seat belt use Unbelted drivers Belted drivers ∗
1.42 1.77 0.88 0.76
Belted (%) (N = 725)
12.7 42.1∗ 40.9∗ 4.2∗
Driver gender Male drivers Female drivers
0.88 1.27 0.64 0.36
Unbelted (%) (N = 201)
Significant at the 0.05 level.
intoxicated drivers and in speeding vehicles are less likely to be belted. Passengers with intoxicated drivers are 2.56 times more likely than passengers with sober drivers to be unbelted. Also, passengers with speeding drivers are 2.27 times more likely than passengers with non-speeding drivers to be unbelted. These differences are statistically significant at the 0.05 level. We also examined the association between passenger belt use and driver behavior and characteristics in our analysis. Higher proportions of unbelted passengers were observed
among those riding with intoxicated and speeding drivers. When drivers were intoxicated and speeding, passengers were 2.6 times and 2.3 times more likely to be unbelted, respectively. Passengers of high-risk drivers’ vehicles are less likely to use seat belts. Driver gender and age are also significantly associated with passenger seat belt use. As shown in Table 1, there are proportionately more unbelted passengers when the driver was unbelted. Male drivers are 1.69 times more likely to have unbelted passengers than female drivers. Younger drivers, moreover, are also more likely to have
126
S. Kim, K. Kim / Accident Analysis and Prevention 35 (2003) 121–130
Fig. 2. Probability of seat belt use by age, among seriously injured occupants in Hawaii (1986–1995).
unbelted passengers. Among teen (age 15–19) and young adult (age 20–29) drivers, the proportion of unbelted passengers exceed that of belted passengers. Among adult (age 30–64) and senior (age 65 and over) drivers, however, the proportion of unbelted passengers is smaller than that of belted passengers. These differences are all significant at the 0.05 level. In Hawaii, the driver is responsible for a passenger’s seat belt use regardless of the passenger’s age. In order to measure the association between driver belt use and passenger belt use, we compared drivers and front seat passengers. We found 926 seriously injured right front seat passengers. Among these, 201 were unbelted and 725 were belted. Approximately 58% of passengers were unbelted when drivers were unbelted. However, only 1.6% of belted passengers were observed in the same cases. Unbelted drivers were 8.52 times more likely to have unbelted passengers than belted drivers. Among seriously injured occupants, there is a strong association between driver seat belt use and passenger seat belt use. 3.2. Temporal and spatial factors associated with seat belt non-use Table 2 shows several environmental factors related to belt use of occupants in serious crashes. We divided road type into two categories, freeway/highway and local road. There is no significant difference in belt use by road type for either drivers or passengers. There is, however, a significant difference between urban and rural areas in terms of belt use. In rural area drivers and passengers are 1.71 and 1.45 times more likely to be unbelted, respectively. Both drivers and passengers are less likely to be belted on weekend. Yet, time of the day has a strong association with seat belt use. Table 2 shows that during the nighttime
(6:00 p.m. to 6:00 a.m.) proportionally more both drivers and passengers are unbelted than during the daytime (6:00 a.m. to 6:00 p.m.), 67.8 and 55.7% respectively. These differences are all statistically significant at the 0.05 level. During the night, drivers are 2.24 times more likely to be unbelted than during the day, while passengers are 1.78 times more likely to be unbelted. Fig. 3 shows that as time approaches late night, belt use decreases. Fig. 4 also shows a similar typical pattern for passengers. On rainy days, there is a higher proportion of belted drivers (20.1%) than unbelted drivers, 20.1 versus 9.6% on clear days. This is significant at the 0.05 level. There is some evidence to suggest that drivers may be more cautious on rainy days or of other hazardous driving conditions (Hamed and Easa, 1998). 3.3. Logistic regression models explaining seat belt non-use among seriously injured drivers and passengers in crashes Because of the differences we observed between drivers and passengers, we built two separate logistic regression models for analysis of seat belt non-use: one for drivers and another for passengers. The dependent variable, seat belt non-use, is explained by a combination of several explanatory variables including personal characteristics and temporal and spatial factors. The parameter estimates, standard error, P-value, odds ratio and odds ratios with a 95% confidence interval (CI) are provided to explain the model. A log-odds ratio, exponentiation of parameter estimate, is used to express the magnitude of effect of a predictor in the model. Table 3 shows the logistic regression equation for unbelted crash involved drivers. As noted earlier, personal characteristics of the drivers are significantly associated with seat belt non-use among seriously injured drivers in crashes. As shown in Table 3, driver age is negatively related to non-use.
S. Kim, K. Kim / Accident Analysis and Prevention 35 (2003) 121–130
127
Table 2 Temporal and spatial factors related to seat belt use among seriously injured occupants in Hawaii (1986–1995) Driver
Passenger
Unbelted (%) (N = 777)
Belted (%) (N = 2680)
Odds ratio
95% CL Low
High
Road type Local road Freeway/highway
47.2 52.8
46.7 53.3
1.02
0.87
Urban/rural Rural Urban
53.4∗ 46.6∗
36.2∗ 63.8∗
1.71
Day of the week Monday Tuesday Wednesday Thursday Friday Saturday Sunday
12.5 13.6 13.0 11.7 14.9 18.3 16.0∗
13.3 12.4 13.7 12.5 17.4 17.1 13.6∗
67.8∗
42.9∗
32.2∗
57.1∗
90.4∗ 9.6∗
79.9∗ 20.1∗
Time of the day Night (6:00 p.m. to 6:00 a.m.) Day (6:00 a.m. to 6:00 p.m.) Weather Not rainy Rainy ∗
Unbelted (%) (N = 201)
Belted (%) (N = 725)
Odds ratio ‘
95% CL Low
High
1.19
45.3 54.7
45.8 54.2
0.98
0.72
1.35
1.46
2.01
49.3∗ 50.7∗
37.5∗ 62.5∗
1.45
1.06
1.99
0.95 1.08 0.95 0.94 0.86 1.07 1.16
0.74 0.86 0.75 0.74 0.69 0.87 0.93
1.20 1.37 1.21 1.21 1.08 1.31 1.44
13.9 8.5∗ 10.9 12.9 13.9 23.4∗ 16.4
12.4 13.9∗ 15.0 11.7 13.4 17.4∗ 16.1
1.11 0.64 0.75 1.09 1.03 1.33 1.01
0.70 0.37 0.46 0.68 0.66 0.91 0.66
1.75 1.09 1.21 1.75 1.63 1.94 1.54
2.24
1.89
2.65
55.7∗
37.5∗
1.78
1.29
2.44
44.3∗
62.5∗
85.8 14.2
89.3 10.7
0.78
0.49
1.24
2.03
1.57
2.63
Significant at the 0.05 level.
This indicates younger drivers are less likely to use seat belts. Male drivers are 1.6 times more likely than female drivers to be unbelted. Unlicensed drivers are 1.4 times more likely than licensed drivers to be unbelted. Intoxicated drivers are 2.6 times more likely to be unbelted and speeding drivers are 1.8 times more likely than non-speeding drivers to be unbelted. The
regression results that high-risk behaviors including alcohol use, speeding and seat belt non-use are all strongly and significantly associated. Vehicle type also affects likelihood of seat belt use. Pickup truck or truck drivers are 1.4 times more likely to be unbelted than drivers of passenger cars and vans. Temporal and spatial factors are also strong predictors of seat belt use. A driver at night is 1.8 times more likely to
Fig. 3. Odds ratio for seat belt non-use among seriously injured drivers by hour of the day in Hawaii (1986–1995).
128
S. Kim, K. Kim / Accident Analysis and Prevention 35 (2003) 121–130
Fig. 4. Odds ratio for seat belt non-use among seriously injured passengers by hour of the day in Hawaii (1986–1995).
be unbelted compared to driver in day. Drivers are 1.7 times more likely to be unbelted in rural areas than drivers in urban areas. Weather also is related to seat belt use. On rainy days, drivers are less likely to be unbelted. The log-odds ratio of not using seat belts on rainy days is 0.8.
Table 4 shows the log linear relationship between seriously injured front seat passenger belt use and various explanatory variables. Passenger age is significantly associated with belt use. A negative parameter estimate shows that non-users are most likely to be young and belt use increases
Table 3 Logistic regression analysis of seat belt non-use: driver modela,b Variable
Intercept Driver age Male Unlicensed Alcohol use Speeding Pickup truck/truck Nighttime (6:00 p.m. to 6:00 a.m.) Rural Rainy a b
Parameter −2.0644 −0.0126 0.4688 0.3344 0.9543 0.6113 0.3165 0.5845 0.5443 −0.2648
S.E.
0.1616 0.0033 0.0999 0.1257 0.1226 0.1027 0.1034 0.0947 0.0883 0.1437
Prob(X2 ) <0.0001 0.0001 <0.0001 0.0078 <0.0001 <0.0001 0.0022 <0.0001 <0.0001 0.0654
Odds ratio
0.987 1.598 1.397 2.597 1.843 1.372 1.794 1.723 0.767
95% CI Lower
Upper
0.981 1.314 1.092 2.042 1.507 1.121 1.490 1.450 0.579
0.994 1.944 1.787 3.302 2.253 1.681 2.160 2.049 1.017
Likelihood ratio χ 2 = 430.5751 (with 9 d.f., P ≤ 0.0001). Association of predicted probability and observed responses: concordant = 73.5%; discordant = 26.1%; tied = 0.4%.
Table 4 Logistic regression analysis of seat belt non-use: passenger modela,b,c Variable
Intercept Passenger age Male passenger Nighttime (6:00 p.m. to 6:00 a.m.) Rural Unbelted driver a
Parameter −2.0106 −0.0199 0.4093 0.6324 0.3658 4.2568
S.E.
0.3038 0.0065 0.2205 0.2216 0.2159 0.3448
Prob(X2 ) <0.0001 0.0023 0.0635 0.0043 0.0903 <0.0001
Odds ratio
0.980 1.506 1.882 1.442 70.583
For right front seat passenger only. Likelihood ratio χ 2 = 359.8888 (with 5 d.f., P ≤ 0.0001). c Association of predicted probability and observed responses: concordant = 85.2%; discordant = 14.4%; tied = 0.4%. b
95% CI Lower
Upper
0.968 0.977 1.219 0.944 35.911
0.993 2.320 2.906 2.201 138.732
S. Kim, K. Kim / Accident Analysis and Prevention 35 (2003) 121–130
as age increases. Male passengers are 1.5 times more likely than female passengers to be unbelted. However, the male passenger variable is not significant at the 0.05 level. Early we showed that there was a strong association between temporal circumstances and seat belt use. In this logistic regression model, nighttime (6:00 p.m. to 6:00 a.m.) is a statistically significant explanatory variable. The odds ratio of nighttime is 1.9. The odds ratio of rural area as compared to urban area is 1.4, however, it is not significant at the 0.05 level. The strongest predictor for passenger belt use is of driver belt use. Approximately 89.3% of the passengers were belted when drivers were belted and 90.9% of passengers were not belted when drivers were not belted. The logistic regression model shows that passengers of vehicles with unbelted drivers are 70.6 times more likely to be unbelted than passengers of vehicles with belted drivers.
4. Discussion This paper investigated the relationships between seat belt use and personal, temporal and spatial characteristics. However, our analysis is limited to the examination of seriously injured crash-involved drivers and front seat passengers. Since, seriously injured occupants are more likely to be unbelted, our analysis may have some selectivity bias if we were to apply our results to the entire motorist population. Seriously injured occupants, however, need special attention since they should be a target group in promoting roadway traffic safety. In order to achieve the greatest public health benefits, education and enforcement programs need to be focused on seriously injured unbelted motorists. Our findings are similar to previous studies regarding the associations between seat belt use and related factors. Despite the belief that unbelted drivers tend to be more involved in crashes than belted drivers, the average annual belt use rate of the population in our study is actually quite close to the annual observed belt use rates in Hawaii. The average annual rate from the crash data was slightly higher than the average annual observed seat belt use rate during the same period. This suggests that crash-involved occupants are not significantly less likely to use belts than the general population. Our study provides a comprehensive use analysis of belt use that takes into account specific personal and temporal and spatial characteristics. A strong association between seat belt use and various personal factors, such as age, gender, vehicle type, driver license status, alcohol intoxication and speeding was found. In particular, we found that occupant age is a strong factor that is associated with seat belt use. Both driver and passenger belt use increases as age increases as shown in Fig. 2. This suggests that education should particularly pay attention to young population group. Our study results show that drivers and passengers in pickup trucks and trucks are far more likely to be unbelted than those in passenger cars and vans. More attention should
129
be paid to the relationship between unlicensed drivers and seat belt non-use. Unlicensed drivers are 1.4 times more likely than licensed drivers to be unbelted as shown in Table 3. The results of our model show that risk-taking behaviors, such as speeding and alcohol are strongly associated with non-use. We feel that these associations have a synergistic effect and may intensify occupant injury levels in vehicle crashes. This would seem to dispel notions regarding risk compensation or that unbelted drivers are more cautious than belted ones. Preusser et al. (1991) found that unbelted drivers has 20% more traffic accidents on their records than belted drivers. They also found that personal injury from previous accidents does not strongly affect seat belt use positively. These facts suggest that enforcement of speeding, alcohol and belt use laws may have cumulative positive effects. Perhaps drivers cited for speeding should also be more routinely scrutinized for belt use or alcohol involvement. Perhaps, the lack of belt use is an indicator of other potential violations. An integrated approach to seat belt use and other risk taking behaviors needs to be introduced in education and law enforcement. The part time seat belt user problem has drawn attention amidst rising overall national seat belt use. According to a seat belt survey and telephone interview study done in New York in 1988, 26% of 300 belted drivers said they did not always wear seat belts (Preusser et al., 1991). It is important to consider driving environments including spatial and temporal circumstances affecting part time seat belt users. Our study shows that there is salient difference between daytime and nighttime in seat belt use as shown in Figs. 3 and 4. We also found that drivers and front seat passengers in rural areas are more likely to be unbelted than those in urban areas. Police enforcement and other strategies for increasing seat belt use need to be focused on rural areas and during the nighttime as compared to the daytime. We also found a strong correlation between driver and passenger seat belt use. Overall, unbelted passengers have similar personal characteristics as the drivers of the vehicles they ride with. Drivers and front seat passengers tend to be of similar age. Yet, we need to know look more closely at how driver’s seat belt use affects passenger’s seat belt use.
5. Conclusions The analysis reveals that the characteristics of seat belt non-users can not only be identified, but also used to develop a statistical model for understanding the relative effects of age, gender, risk-taking behaviors and other crash variables. Our study, therefore, contributes to a more detailed understanding of the complex relationships between driver, vehicle, roadway and environmental factors and seat belt non-use among seriously injured occupants. We think that the logistic regression technique is particularly robust in that it allows us to handle categorical data and to compare
130
S. Kim, K. Kim / Accident Analysis and Prevention 35 (2003) 121–130
the relative odds or likelihood of being in a particular cross-classified category. The spline method is an enhancement to the technique, which increases the predictive power of the overall model. We think that the technique can be applied to many other traffic safety research problems. There are some limitations to our study. We hesitate to transfer our results to all drivers in all locales in part because we focused more narrowly on crash-involved, seriously injured motorists. They may not, necessarily, be identical to the overall motoring public. For instance, the proportion of drivers who are intoxicated or commit speeding offences in our analysis may be higher than the average driver because drunk drivers are more likely to be seriously injured in crashes. We need additional analysis of less seriously injured as well as unbelted occupants involved in minor crashes. Our study demonstrates that police-reported crash data are a useful resource in investigating seat belt use. However, the crash data does not contain strong potential factors that may affect seat belt use. Those factors include, income, education level, marital status, occupation, presence of children in household, detailed vehicle type information (sports car, sedan, coupe, etc.), vehicle crash experience, road environment (speed limit, congestion level, etc.), belief in the effectiveness of seat belt and the level of perceived injury risk. Perhaps future analysis will focus on citation data linked to crash data. There is a need to combine analysis of observational, crash, citation and attitudinal data for a single jurisdiction, over the same period of time. We think that Hawaii is the ideal place to examine these different sources of data on non-use. Hopefully our future research endeavors will enable us to do so. Appendix A The odds ratio, that is the odds of being an unbelted occupant, relative to another associated factor (such as gender) can be expressed as follows: OR =
a/(a + b) , c/(c + d)
where OR is the odds ratio, a the unbelted male frequency, b the belted male frequency, c the unbelted female frequency and d the belted female frequency. To calculate the 95% CI using Woolf’s method, 1 1 1 1 S.E.(ln OR) = + + + , a b c d ln (OR) = ±1.96 × S.E.(ln OR)
The confidence limit (CL) can be derived by taking antilogs (source: Kahn and Sempos (1989)). References Diamantopoulou, K., Dyte, D., Cameron, M., 1997. Seat belt wearing rates in Victoria (1994). Monash University Accident Research Centre-Report #89. Eubank, R.L., 1999. Non-Parametric Regression and Spline Smoothing. Marcel Dekker, New York. Figgers, R., Nash, C.E., 1989. Accurately reporting belt use in automobile crashes. The Police Chief 07/89, 50. Hamed, M.M., Easa, S.M., 1998. Ordered probability modeling of seat belt usage. J. Transport. Eng. ASCE 124 (3), 271–276. Hunter, W.W., Stewart, J.R., Stutts, J.C., Rodgman, E.A., 1993. Observed and self-reported seat belt wearing as related to prior accidents and convictions. Accid. Anal. Prev. 25, 545–554. Kahn, H., Sempos, C., 1989. Statistical Methods in Epidemiology. Oxford University Press, Oxford. Kim, K., 1991. Effects of enforcement on seat belt use in Hawaii. Transport. Res. Rec. 1325, 51–56. Kim, K., 1999. The lie factor in traffic safety: comparison of police and hospital reporting of seat belt and alcohol use in Hawaii. Transport. Res. Rec. 1665, 141–146. Kim, K., Nitz, L., Richardson, J., Li, L., 1995. Personal and behavioral predictors of automobile crash and injury severity. Accid. Anal. Prev. 27, 469–481. Li, L., Kim, K., Nitz, L., 1999. Predictors of seat belt use among crash-involved drivers and front seat passengers: adjusting for over-reporting. Accid. Anal. Prev. 31, 631–638. National Highway Traffic Safety Administration (NHTSA), US Department of Transportation, 1995. Research Note: National Occupant Protection Use Survey. Controlled Intersection Study. National Highway Traffic Safety Administration (NHTSA), US Department of Transportation, 1996. Report to Congress: Benefits of Safety Belts and Motorcycle Helmets. National Highway Traffic Safety Administration (NHTSA), US Department of Transportation, 1997. Presidential Initiative for Increasing Seat Belt Use Nationwide. National Highway Traffic Safety Administration (NHTSA), US Department of Transportation, 1998a. Buckle Up America: First Report to Congress. The Presidential Initiative for Increasing Seat Belt Use Nationwide. National Highway Traffic Safety Administration (NHTSA), US Department of Transportation, 1998b. Research Note: Observed Seat belt Use in 1998. O’Day, J., 1993. Accident Data Quality: A Synthesis of Highway Practice, National Cooperative Highway Research Program 192, Transportation Research Board. National Academy Press, Washington, DC. Preusser, D.F., 1988. Belt use by high-risk drivers before and after New York’s seat belt use law. Accid. Anal. Prev. 20, 245–250. Preusser, D.F., Williams, A.F., Lund, A.K., 1991. Characteristics of belted and unbelted drivers. Accid. Anal. Prev. 23, 475–482. Schumaker, L.L., 1993. Spline Functions: Basic Theory. Krieger Publishing Co., Melbourne.