- Email: [email protected]
Car insurance and the risk of car crash injury
Accident Analysis and Prevention 35 (2003) 987–990
Car insurance and the risk of car crash injury Stephanie Blows a,∗ , Rebecca Q. Ivers a , Jennie Connor b , Shanthi Ameratunga b , Robyn Norton a a
Institute For International Health, University of Sydney, P.O. Box 576, 144 Burren Street, Newton 2042, NSW, Australia b Division of Community Health, University of Auckland, Auckland, New Zealand Received 8 August 2002; received in revised form 26 September 2002; accepted 12 October 2002
Abstract Despite speculation about the role of vehicle insurance in road traffic accidents, there is little research estimating the direction or extent of the risk relationship. Data from the Auckland Car Crash Injury Study (1998–1999) were used to examine the association between driving an uninsured motor vehicle and car crash injury. Cases were all cars involved in crashes in which at least one occupant was hospitalized or killed anywhere in the Auckland region. Controls were 588 drivers of randomly selected cars on Auckland roads. Participants completed a structured interview. Uninsured drivers had significantly greater odds of car crash injury compared to insured drivers after adjustment for age, sex, level of education, and driving exposure (odds ratio 4.77, 95% confidence interval 2.94–7.75). The causal mechanism for insurance and car crash injury is not easily determined. Although we examined the effects of multiple potential confounders in our analysis including socioeconomic status and risk-taking behaviours, both of which have been previously observed to be associated with both insurance status and car crash injury, residual confounding may partly explain this association. The estimated proportion of drivers who are uninsured is between 5 and 15% in developed countries, representing a significant public health problem worthy of further investigation. © 2003 Elsevier Science Ltd. All rights reserved. Keywords: Traffic accidents; Risky driving; Insurance; Case-control study
1. Introduction The proportion of drivers who do not purchase insurance for their vehicle is estimated to be up to 15% in developed countries (Insurance Research Council, 2000; OECD, 1990). The insurance status of a vehicle is likely to be indicative of certain characteristics of its driver, and insurance status is therefore potentially associated with the risk of car crash injury. Several hypotheses have been proposed on the role of vehicle insurance in motor vehicle crashes. These are based mostly on insurance market theories and descriptive data linking insurance status with lifestyle. Some of these (Evans, 1991; Hemenway, 1993; Bruce and Weissenberger, 2001) suggest uninsured drivers are at increased risk of car crash. Others (Wilde, 1994; Evans, 1991) suggest insured drivers are at increased risk. Despite the uncertainty about the relationship between car insurance and car crash injury, no epidemiological studies directly examining this association were found in an extensive literature search. We examined the association between vehicle insurance and car crash injury using data from the Auckland Car Crash Injury Study, a population-based case-control study. Comprehensive or third-party property motor vehicle insurance ∗
Corresponding author. Tel.: +61-2-9351-0007; fax: +61-2-9351-0008.
is not compulsory in New Zealand. The study aimed to examine the associations between human, vehicle and environmental factors and car crash injury. A large number of car crash related variables were measured, allowing us to control for a range of potential confounding variables.
2. Methods The Auckland Car Crash Injury Study was conducted between 1998 and 1999 in the Auckland region of New Zealand. Details of the methodology of the study have been published elsewhere (Connor et al., 2002). Cases were all cars involved in crashes in which at least one occupant of the car was hospitalised with injuries or killed, during the study period, anywhere in the Auckland region. Cases were recruited by identifying all individuals (both passengers and drivers) with injuries from a car crash and admitted within 24 h of the crash to any of the four hospitals that serve the Auckland region. Fatally injured occupants were identified through the Auckland Coroner’s office. Controls were selected to represent a sample of all driving time in the study region during the study period. A list of all non-local roads was obtained from New Zealand road-controlling authorities and 69 random road sites were
0001-4575/$ – see front matter © 2003 Elsevier Science Ltd. All rights reserved. PII: S 0 0 0 1 - 4 5 7 5 ( 0 2 ) 0 0 1 0 6 - 9
988
S. Blows et al. / Accident Analysis and Prevention 35 (2003) 987–990
selected. A day of week, time of day, and direction of travel were randomly assigned to each site. A random cluster sampling method was used to select clusters of control cars from these sites. Control cars were randomly identified and stopped as close as possible to the selected points, where they were asked for permission to make contact by telephone for an interview. Controls were recruited at approximately the same rate as cases. Drivers of case and control vehicles completed a face-to-face or telephone questionnaire-based structured interview. The next of kin or other suitable person (proxy respondent) was interviewed when a driver was unable to complete the interview or when the driver was fatally injured. The interview included demographics, circumstances of the crash, personal characteristics, vehicle factors and environmental factors. Participants were asked if the vehicle was insured with third party or replacement insurance at the time of the crash or survey; response categories were ‘yes’, ‘no’, or ‘don’t know’. Different types of vehicle insurance were not separated in this study. 2.1. Study population During the study period 615 eligible case vehicles were identified. Of these, interviews were completed for 571 drivers (92.8%). Non-responders included 30 (4.9%) case vehicles for which the driver declined to participate in an interview and 14 (2.3%) who could not be contacted. From the roadside surveys, 746 cars were identified as control vehicles. Of these, interviews were completed for 588 drivers (78.8%). Non-responders included 92 (12.3% of total) who declined to participate, 60 (8.0%) that could not be contacted, and 6 (0.8%) who could not participate for other reasons. Interviews were conducted by telephone for 204 (35.7%) case interviews compared to 576 (98.0%) control interviews. The remainder of interviews were conducted in person, except for three controls (0.5%) who self-completed a mailed questionnaire. Proxy respondents completed 57 (10%) case interviews and 2 (0.3%) control interviews.
potential confounders were eliminated in a backwards manner from the full model in order of significance (P-value). Variables were considered confounders and retained in the model if they produced at least a 10% change in the odds ratio for the relationship between insurance and car crash injury (Rothman and Greenland, 1998).
3. Results Table 1 shows the distribution of age and sex amongst cases and controls. The mean age of cases was 36.6 years and of controls 40.8 years. There were more males than females in both groups. The question about vehicle insurance was answered ‘yes’ or ‘no’ by 577 (98.1%) controls. Of these, 10.4% were driving an uninsured vehicle at the time they completed the survey. The mean age of uninsured drivers was 31.6 years and of insured control drivers was 41.7 years. More males than females were uninsured: 12.7% of male control drivers in the study were uninsured compared to 3.8% of females. More drivers who had studied past secondary school were insured. The proportion of uninsured control drivers was lowest amongst white-collar, high-skilled workers and highest amongst blue-collar, low-skilled workers. The question about vehicle insurance was answered ‘yes’ or ‘no’ by 528 (92.5%) cases. Of these, 39.0% were driving an uninsured vehicle at the time of the crash. Cases who did not answer this question tended to be younger, were less likely to be employed, and were more likely to have a blue collar, low skill occupation, compared to those who did answer the question. The proportion of participants who were driving a car that was registered in their own name was similar for both groups (56.0% of cases and 63.5% of controls). Table 2 shows the distribution of insurance status and confounding variables. Other potential confounding variables that were examined but not included in the final model were ethnicity, household income, employment status, occupation, time of day, sleep in the past 24 h, self-reported speed, number of passengers, acute sleepiness, duration of driving, registration status of the vehicle, current vehicle inspection
2.2. Analysis Logistic regression was performed using Sudaan software to account for intracluster correlation of data from the same site. Potential confounding variables, defined as those that were associated with the outcome and/or the exposure variable (Rothman and Greenland, 1998), were ascertained from the literature review. Unadjusted and age and sex adjusted associations between each potential confounding variable and car crash injury were initially examined. Potential confounding variables were included in the full model if they were significantly associated with car crash injury after adjustment for age and sex. Age and sex were included in each model regardless of their significance because of their a priori status as confounding variables (McDonald, 1994). Other
Table 1 Participants in the Auckland Car Crash Injury Study, by age group and sex, Auckland, New Zealand, 1998–1999 Age group (years)
<25 25–34 35–44 45–54 55–64 65+ Total
Cases (n = 571)
Controls (n = 588)
Females
Males
Females
Males
No.
%
No.
%
No.
No.
53 42 36 17 21 29
26.8 21.2 18.2 8.6 10.6 14.6
142 91 49 44 18 29
38.1 24.4 13.1 11.8 4.8 7.8
37 50 61 36 28 14
16.4 22.1 27.0 15.9 12.4 6.2
54 75 93 71 52 17
14.9 20.7 25.7 19.6 14.4 4.7
198
34.7
373
65.3
226
38.4
362
61.6
%
%
S. Blows et al. / Accident Analysis and Prevention 35 (2003) 987–990 Table 2 Distribution of vehicle insurance status and confounders for cases and controls, Auckland Car Crash Injury Study, New Zealand, 1998–1999a Cases
Controls
No.
%
No.
%
Vehicle insurance status Insured Not insured Don’t know
322 206 42
56.5 36.1 7.4
514 63 11
88.3 10.2 1.5
Age group (years) <25 25–34 35–44 45–54 55–64 65+
195 133 85 61 39 58
34.2 23.3 14.9 10.7 6.8 10.2
91 125 154 107 80 31
13.7 21.1 25.9 18.2 13.6 5.6
Sex Female Male
198 373
34.7 38.4
226 362
65.3 61.6
Education level Post secondary Secondary >3 years <3 years secondary
178 137 252
31.4 24.2 44.4
276 154 157
49.3 25.1 25.6
Average driving hours per week ≤5 219 6–10 205 11–20 63 21–30 11 >30 22
42.1 39.4 12.1 2.1 4.2
171 216 135 32 27
30.5 39.3 22.3 3.9 4.1
a
Proportions of controls are adjusted for cluster sampling design.
status of the vehicle, vehicle age, engine size, usual speeding habits, usual seatbelt use, self-reported and objective alcohol consumption, marijuana use, usual drink-driving habits, wearing a seatbelt at the time of the crash, licence type, and being a smoker. Table 3 shows the unadjusted, age and sex adjusted, and multivariate associations between insurance status and car crash injury. In the full model for insurance and car crash injury, before elimination of any potential confounders, uninsured drivers had significantly increased odds of car crash injury compared to insured drivers (odds ratio 3.16, 95% confidence interval 1.38–7.26). Confounding variables in the final model were age, sex, level of education, and driving exposure (hours spent driving per week). After controlling Table 3 Unadjusted, age and sex adjusted and multivariate adjusteda associations between vehicle insurance and car crash injury, Auckland Car Crash Injury Study, New Zealand, 1998–1999 Univariate OR
95% CI+
Vehicle insured Yes 1.00 No 5.52 3.57–8.53 a
Age and sex adjusted
Multivariatea
OR
95% CI
OR
95% CI
1.00 5.62
3.33–9.46
1.00 4.77
2.94–7.75
Adjusted for age, sex, education level, and driving exposure (hours per week).
989
for these confounders, uninsured drivers had significantly increased odds of car crash injury compared to insured drivers (odds ratio 4.77, 95% confidence interval 2.94–7.75). Controlling for the mode of interview (by telephone or in person) and excluding proxy respondents from the analysis did not significantly alter the effect estimates. Using a value of P < 0.01 as the criterion for significance, there was no evidence of interaction between the main effect and age or sex in these data.
4. Discussion This population-based case-control study provides new evidence that uninsured motorists have a significantly increased risk of being injured or killed in a car crash compared to insured motorists. There is limited prior research on the role of insurance in motor vehicle injury that has been carried out using sound research techniques. The case-control design of this study allowed measurement of multiple acute factors related to car crash injury, providing an opportunity to control for a range of important confounding factors. The relationship between uninsured motorists and car crash injury persisted after adjustment for multiple confounding variables. Furthermore, the differential response rate to the question on insurance between cases (7.5% non-responders) and controls (1.9% non-responders) may indicate that the estimated 39% of crash-involved vehicles being uninsured is an underestimate, and the real effect of insurance on car crash injury might be larger. In New Zealand, an accident compensation scheme provides no-fault personal injury insurance for all citizens and residents, and third-party property and replacement insurance are optional. ‘Uninsured’ drivers in this study were therefore those who chose not to purchase either of these non-compulsory forms of insurance. The results of this study to should therefore be generalised only to drivers who choose not to purchase non-compulsory vehicle insurance. For example, Australia, Canada, and parts of the United States operate a liability-based system that requires all drivers to purchase third-party personal injury insurance. Proof of this insurance is required for registration of the vehicle. Drivers without this compulsory insurance are driving an unregistered vehicle and are likely to have different characteristics to drivers who chose not to purchase third party or replacement insurance. In the control population of this study, conceptually representing the Auckland region driving population (Wacholder et al., 1992), 10% of drivers were uninsured. In countries that require compulsory third-party injury insurance, such as Australia, the proportion of drivers who chose not to purchase non-compulsory insurance is estimated to be about 9% (Productivity Commission Australia, 1998). These uninsured drivers have an estimated four times increased risk of car crash injury compared to insured drivers and they therefore represent a significant public health problem.
990
S. Blows et al. / Accident Analysis and Prevention 35 (2003) 987–990
The relationship between insurance and car crash injury may be subject to bias in our study. One potential source of bias is the mode of interview, as the majority of cases were interviewed in person whereas the majority of controls were interviewed by telephone. Furthermore, proxy respondents were interviewed for more cases than controls and their responses may have been less valid. However, controlling for mode of interview and proxy status did not significantly alter the effect estimate. The possibility of recall bias also has to be considered in all case-control studies using self-reported data. It is difficult, however, to predict the direction in which the estimated effect might be biased. The mechanism for a causal relationship between insurance and car crash injury is not clear, and the possibility of undetected confounding should be considered. For example, lower socioeconomic status is known to be associated with being uninsured, as well as with increased risk of car crash (Hunstead, 1999; Laflamme and Engstrom, 2002; Harper et al., 2000; Cubbin et al., 2000a,b; Loomis, 1991). An Australian survey found that of uninsured drivers, 32% were uninsured because they could not afford the insurance premium (Productivity Commission Australia, 1998). This study also found several socioeconomic status variables to be significantly associated with car crash injury, however, their inclusion in the model for insurance and car crash risk did not change the significant effect of insurance. Another possible explanation for the association between insurance and car crash injury is that drivers who are uninsured take more risks. An insurance market theory known as ‘propitious selection’ suggests that risk-averse individuals are more likely to take precautions to reduce risks when driving as well as purchase insurance (Hemenway, 1993). This hypothesis is supported by descriptive survey data indicating that people without insurance are more likely to engage in risky lifestyle behaviours and risky driving behaviours (Hemenway, 1992; Clyde et al., 1996; Preusser and Williams, 1991; Evans and Wasielewski, 1983). This study found that uninsured drivers are more likely than insured drivers to be injured in a car crash, supporting the suggestion that uninsured drivers engage in risky driving and raising the possibility that insurance is acting as a proxy variable for risk taking behaviour. Our analysis included many risky behaviour variables but they did not diminish the effect of insurance. Further research is needed to clarify the relationship between insurance status and car crash injury, as well as insurance status and risk taking behaviours and socio-economic status, to identify opportunities for injury prevention. This population-based case-control study provides new evidence that uninsured drivers are at higher risk of car crash injury than insured drivers after adjustment for multiple confounders. It is possible this relationship is subject to residual
confounding or bias. Despite this, the substantial proportion of uninsured motorists in the driving population makes this an important public health problem that warrants further investigation. References Bruce, C.J., Weissenberger, A.T., 2001. No-fault automobile insurance. Economica 6 (3), 1–16. Clyde, A.T., Hemenway, D., Nagurney, J.T., 1996. Seatbelt use, insurance status and hospital bad debt. J. Trauma 41 (1), 100–104. Connor, J., Norton, R., Ameratunga, S., et al., 2002. Driver sleepiness and risk of serious injury to car occupants: population-based case-control study. BMJ 324, 1–5. Cubbin, C., LeClere, F.B., Smith, G.S., 2000a. Socioeconomic status and the occurrence of fatal and non-fatal injury in the United States. Am. J. Public Health 90 (1), 70–77. Cubbin, C., LeClere, F.B., Smith, G.S., 2000b. Socioeconomic status and injury mortality: individual and neighbourhood determinants. J. Epidemiol. Commun. Health 54, 517–524. Evans, L., Wasielewski, P., 1983. Risky driving related to driver and vehicle characteristics. Accident Anal. Prev. 15 (2), 121–136. Evans L., 1991. Traffic Safety and The Driver. Van Nostrand Reinhold, New York. Harper, J.S., Marine, W.M., Garrett, C.J., Lezotte, D., Lowenstein, S.R., 2000. Motor vehicle crash fatalities: a comparison of hispanic and non-hispanic motorists in Colorado. Ann. Emergency Med. 36 (6), 589–596. Hemenway, D., 1992. Propitious selection in insurance. J. Risk Uncertainty 5, 247–251. Hemenway, D., 1993. Nervous Nellies and Dangerous Dans. J. Pol. Anal. Manage. 12 (2), 359–363. Hunstead L., 1999. Characteristics of the uninsured motorist. Californian Department of Insurance, Sacramento. Available at: http://www.insurance.ca.gov/PRP. Accessed March 2002. Insurance Research Council of America, 2000. Uninsured Motorists, 2000 ed. Available at: http://www.ircweb.org. Accessed March 2002. Laflamme, L., Engstrom, K., 2002. Socioeconomic differences in Swedish children and adolescents injured in road traffic accidents: cross sectional study. Br. Med. J. 324, 396–397. Loomis, D.P., 1991. Occupation, industry, and fatal motor vehicle crashes in 20 states, 1986–1987. Am. J. Public Health 81 (6), 733–735. McDonald W.A., 1994. Young driver research program: a review of information on young driver performance characteristics and capacities. Federal Government Road Safety Initiative CR 129, Federal Office of Road Safety, Canberra, Australia. Organisation for Economic Co-operation and Development (OECD), 1990. Road Transport Research: Automobile Insurance and Road Accident Prevention. OECD, Paris. Preusser, D.F., Williams, A.F., 1991. Characteristics of belted and unbelted drivers. Accident Anal. Prev. 23 (6), 475–482. Productivity Commission, 1998. Inquiry report into vehicle and recreational marine craft repair and insurance industries. Available at: http://www.pc.gov.au/ic/inquiry. Accessed May 2002. Rothman, K., Greenland, S., 1998. Modern Epidemiology. Lippencott-Raven, Philadelphia. Wacholder, S., McLaughlin, J.K., Silverman, D.T., Mandel, J.S., 1992. Selection of controls in case-control studies. I. Principles. Am. J. Epidemiol. 135 (9), 1019–1028. Wilde, G.J., 1994. Target Risk. PDE Publications, Canada.
Car insurance and the risk of car crash injury Stephanie Blows a,∗ , Rebecca Q. Ivers a , Jennie Connor b , Shanthi Ameratunga b , Robyn Norton a a
Institute For International Health, University of Sydney, P.O. Box 576, 144 Burren Street, Newton 2042, NSW, Australia b Division of Community Health, University of Auckland, Auckland, New Zealand Received 8 August 2002; received in revised form 26 September 2002; accepted 12 October 2002
Abstract Despite speculation about the role of vehicle insurance in road traffic accidents, there is little research estimating the direction or extent of the risk relationship. Data from the Auckland Car Crash Injury Study (1998–1999) were used to examine the association between driving an uninsured motor vehicle and car crash injury. Cases were all cars involved in crashes in which at least one occupant was hospitalized or killed anywhere in the Auckland region. Controls were 588 drivers of randomly selected cars on Auckland roads. Participants completed a structured interview. Uninsured drivers had significantly greater odds of car crash injury compared to insured drivers after adjustment for age, sex, level of education, and driving exposure (odds ratio 4.77, 95% confidence interval 2.94–7.75). The causal mechanism for insurance and car crash injury is not easily determined. Although we examined the effects of multiple potential confounders in our analysis including socioeconomic status and risk-taking behaviours, both of which have been previously observed to be associated with both insurance status and car crash injury, residual confounding may partly explain this association. The estimated proportion of drivers who are uninsured is between 5 and 15% in developed countries, representing a significant public health problem worthy of further investigation. © 2003 Elsevier Science Ltd. All rights reserved. Keywords: Traffic accidents; Risky driving; Insurance; Case-control study
1. Introduction The proportion of drivers who do not purchase insurance for their vehicle is estimated to be up to 15% in developed countries (Insurance Research Council, 2000; OECD, 1990). The insurance status of a vehicle is likely to be indicative of certain characteristics of its driver, and insurance status is therefore potentially associated with the risk of car crash injury. Several hypotheses have been proposed on the role of vehicle insurance in motor vehicle crashes. These are based mostly on insurance market theories and descriptive data linking insurance status with lifestyle. Some of these (Evans, 1991; Hemenway, 1993; Bruce and Weissenberger, 2001) suggest uninsured drivers are at increased risk of car crash. Others (Wilde, 1994; Evans, 1991) suggest insured drivers are at increased risk. Despite the uncertainty about the relationship between car insurance and car crash injury, no epidemiological studies directly examining this association were found in an extensive literature search. We examined the association between vehicle insurance and car crash injury using data from the Auckland Car Crash Injury Study, a population-based case-control study. Comprehensive or third-party property motor vehicle insurance ∗
Corresponding author. Tel.: +61-2-9351-0007; fax: +61-2-9351-0008.
is not compulsory in New Zealand. The study aimed to examine the associations between human, vehicle and environmental factors and car crash injury. A large number of car crash related variables were measured, allowing us to control for a range of potential confounding variables.
2. Methods The Auckland Car Crash Injury Study was conducted between 1998 and 1999 in the Auckland region of New Zealand. Details of the methodology of the study have been published elsewhere (Connor et al., 2002). Cases were all cars involved in crashes in which at least one occupant of the car was hospitalised with injuries or killed, during the study period, anywhere in the Auckland region. Cases were recruited by identifying all individuals (both passengers and drivers) with injuries from a car crash and admitted within 24 h of the crash to any of the four hospitals that serve the Auckland region. Fatally injured occupants were identified through the Auckland Coroner’s office. Controls were selected to represent a sample of all driving time in the study region during the study period. A list of all non-local roads was obtained from New Zealand road-controlling authorities and 69 random road sites were
0001-4575/$ – see front matter © 2003 Elsevier Science Ltd. All rights reserved. PII: S 0 0 0 1 - 4 5 7 5 ( 0 2 ) 0 0 1 0 6 - 9
988
S. Blows et al. / Accident Analysis and Prevention 35 (2003) 987–990
selected. A day of week, time of day, and direction of travel were randomly assigned to each site. A random cluster sampling method was used to select clusters of control cars from these sites. Control cars were randomly identified and stopped as close as possible to the selected points, where they were asked for permission to make contact by telephone for an interview. Controls were recruited at approximately the same rate as cases. Drivers of case and control vehicles completed a face-to-face or telephone questionnaire-based structured interview. The next of kin or other suitable person (proxy respondent) was interviewed when a driver was unable to complete the interview or when the driver was fatally injured. The interview included demographics, circumstances of the crash, personal characteristics, vehicle factors and environmental factors. Participants were asked if the vehicle was insured with third party or replacement insurance at the time of the crash or survey; response categories were ‘yes’, ‘no’, or ‘don’t know’. Different types of vehicle insurance were not separated in this study. 2.1. Study population During the study period 615 eligible case vehicles were identified. Of these, interviews were completed for 571 drivers (92.8%). Non-responders included 30 (4.9%) case vehicles for which the driver declined to participate in an interview and 14 (2.3%) who could not be contacted. From the roadside surveys, 746 cars were identified as control vehicles. Of these, interviews were completed for 588 drivers (78.8%). Non-responders included 92 (12.3% of total) who declined to participate, 60 (8.0%) that could not be contacted, and 6 (0.8%) who could not participate for other reasons. Interviews were conducted by telephone for 204 (35.7%) case interviews compared to 576 (98.0%) control interviews. The remainder of interviews were conducted in person, except for three controls (0.5%) who self-completed a mailed questionnaire. Proxy respondents completed 57 (10%) case interviews and 2 (0.3%) control interviews.
potential confounders were eliminated in a backwards manner from the full model in order of significance (P-value). Variables were considered confounders and retained in the model if they produced at least a 10% change in the odds ratio for the relationship between insurance and car crash injury (Rothman and Greenland, 1998).
3. Results Table 1 shows the distribution of age and sex amongst cases and controls. The mean age of cases was 36.6 years and of controls 40.8 years. There were more males than females in both groups. The question about vehicle insurance was answered ‘yes’ or ‘no’ by 577 (98.1%) controls. Of these, 10.4% were driving an uninsured vehicle at the time they completed the survey. The mean age of uninsured drivers was 31.6 years and of insured control drivers was 41.7 years. More males than females were uninsured: 12.7% of male control drivers in the study were uninsured compared to 3.8% of females. More drivers who had studied past secondary school were insured. The proportion of uninsured control drivers was lowest amongst white-collar, high-skilled workers and highest amongst blue-collar, low-skilled workers. The question about vehicle insurance was answered ‘yes’ or ‘no’ by 528 (92.5%) cases. Of these, 39.0% were driving an uninsured vehicle at the time of the crash. Cases who did not answer this question tended to be younger, were less likely to be employed, and were more likely to have a blue collar, low skill occupation, compared to those who did answer the question. The proportion of participants who were driving a car that was registered in their own name was similar for both groups (56.0% of cases and 63.5% of controls). Table 2 shows the distribution of insurance status and confounding variables. Other potential confounding variables that were examined but not included in the final model were ethnicity, household income, employment status, occupation, time of day, sleep in the past 24 h, self-reported speed, number of passengers, acute sleepiness, duration of driving, registration status of the vehicle, current vehicle inspection
2.2. Analysis Logistic regression was performed using Sudaan software to account for intracluster correlation of data from the same site. Potential confounding variables, defined as those that were associated with the outcome and/or the exposure variable (Rothman and Greenland, 1998), were ascertained from the literature review. Unadjusted and age and sex adjusted associations between each potential confounding variable and car crash injury were initially examined. Potential confounding variables were included in the full model if they were significantly associated with car crash injury after adjustment for age and sex. Age and sex were included in each model regardless of their significance because of their a priori status as confounding variables (McDonald, 1994). Other
Table 1 Participants in the Auckland Car Crash Injury Study, by age group and sex, Auckland, New Zealand, 1998–1999 Age group (years)
<25 25–34 35–44 45–54 55–64 65+ Total
Cases (n = 571)
Controls (n = 588)
Females
Males
Females
Males
No.
%
No.
%
No.
No.
53 42 36 17 21 29
26.8 21.2 18.2 8.6 10.6 14.6
142 91 49 44 18 29
38.1 24.4 13.1 11.8 4.8 7.8
37 50 61 36 28 14
16.4 22.1 27.0 15.9 12.4 6.2
54 75 93 71 52 17
14.9 20.7 25.7 19.6 14.4 4.7
198
34.7
373
65.3
226
38.4
362
61.6
%
%
S. Blows et al. / Accident Analysis and Prevention 35 (2003) 987–990 Table 2 Distribution of vehicle insurance status and confounders for cases and controls, Auckland Car Crash Injury Study, New Zealand, 1998–1999a Cases
Controls
No.
%
No.
%
Vehicle insurance status Insured Not insured Don’t know
322 206 42
56.5 36.1 7.4
514 63 11
88.3 10.2 1.5
Age group (years) <25 25–34 35–44 45–54 55–64 65+
195 133 85 61 39 58
34.2 23.3 14.9 10.7 6.8 10.2
91 125 154 107 80 31
13.7 21.1 25.9 18.2 13.6 5.6
Sex Female Male
198 373
34.7 38.4
226 362
65.3 61.6
Education level Post secondary Secondary >3 years <3 years secondary
178 137 252
31.4 24.2 44.4
276 154 157
49.3 25.1 25.6
Average driving hours per week ≤5 219 6–10 205 11–20 63 21–30 11 >30 22
42.1 39.4 12.1 2.1 4.2
171 216 135 32 27
30.5 39.3 22.3 3.9 4.1
a
Proportions of controls are adjusted for cluster sampling design.
status of the vehicle, vehicle age, engine size, usual speeding habits, usual seatbelt use, self-reported and objective alcohol consumption, marijuana use, usual drink-driving habits, wearing a seatbelt at the time of the crash, licence type, and being a smoker. Table 3 shows the unadjusted, age and sex adjusted, and multivariate associations between insurance status and car crash injury. In the full model for insurance and car crash injury, before elimination of any potential confounders, uninsured drivers had significantly increased odds of car crash injury compared to insured drivers (odds ratio 3.16, 95% confidence interval 1.38–7.26). Confounding variables in the final model were age, sex, level of education, and driving exposure (hours spent driving per week). After controlling Table 3 Unadjusted, age and sex adjusted and multivariate adjusteda associations between vehicle insurance and car crash injury, Auckland Car Crash Injury Study, New Zealand, 1998–1999 Univariate OR
95% CI+
Vehicle insured Yes 1.00 No 5.52 3.57–8.53 a
Age and sex adjusted
Multivariatea
OR
95% CI
OR
95% CI
1.00 5.62
3.33–9.46
1.00 4.77
2.94–7.75
Adjusted for age, sex, education level, and driving exposure (hours per week).
989
for these confounders, uninsured drivers had significantly increased odds of car crash injury compared to insured drivers (odds ratio 4.77, 95% confidence interval 2.94–7.75). Controlling for the mode of interview (by telephone or in person) and excluding proxy respondents from the analysis did not significantly alter the effect estimates. Using a value of P < 0.01 as the criterion for significance, there was no evidence of interaction between the main effect and age or sex in these data.
4. Discussion This population-based case-control study provides new evidence that uninsured motorists have a significantly increased risk of being injured or killed in a car crash compared to insured motorists. There is limited prior research on the role of insurance in motor vehicle injury that has been carried out using sound research techniques. The case-control design of this study allowed measurement of multiple acute factors related to car crash injury, providing an opportunity to control for a range of important confounding factors. The relationship between uninsured motorists and car crash injury persisted after adjustment for multiple confounding variables. Furthermore, the differential response rate to the question on insurance between cases (7.5% non-responders) and controls (1.9% non-responders) may indicate that the estimated 39% of crash-involved vehicles being uninsured is an underestimate, and the real effect of insurance on car crash injury might be larger. In New Zealand, an accident compensation scheme provides no-fault personal injury insurance for all citizens and residents, and third-party property and replacement insurance are optional. ‘Uninsured’ drivers in this study were therefore those who chose not to purchase either of these non-compulsory forms of insurance. The results of this study to should therefore be generalised only to drivers who choose not to purchase non-compulsory vehicle insurance. For example, Australia, Canada, and parts of the United States operate a liability-based system that requires all drivers to purchase third-party personal injury insurance. Proof of this insurance is required for registration of the vehicle. Drivers without this compulsory insurance are driving an unregistered vehicle and are likely to have different characteristics to drivers who chose not to purchase third party or replacement insurance. In the control population of this study, conceptually representing the Auckland region driving population (Wacholder et al., 1992), 10% of drivers were uninsured. In countries that require compulsory third-party injury insurance, such as Australia, the proportion of drivers who chose not to purchase non-compulsory insurance is estimated to be about 9% (Productivity Commission Australia, 1998). These uninsured drivers have an estimated four times increased risk of car crash injury compared to insured drivers and they therefore represent a significant public health problem.
990
S. Blows et al. / Accident Analysis and Prevention 35 (2003) 987–990
The relationship between insurance and car crash injury may be subject to bias in our study. One potential source of bias is the mode of interview, as the majority of cases were interviewed in person whereas the majority of controls were interviewed by telephone. Furthermore, proxy respondents were interviewed for more cases than controls and their responses may have been less valid. However, controlling for mode of interview and proxy status did not significantly alter the effect estimate. The possibility of recall bias also has to be considered in all case-control studies using self-reported data. It is difficult, however, to predict the direction in which the estimated effect might be biased. The mechanism for a causal relationship between insurance and car crash injury is not clear, and the possibility of undetected confounding should be considered. For example, lower socioeconomic status is known to be associated with being uninsured, as well as with increased risk of car crash (Hunstead, 1999; Laflamme and Engstrom, 2002; Harper et al., 2000; Cubbin et al., 2000a,b; Loomis, 1991). An Australian survey found that of uninsured drivers, 32% were uninsured because they could not afford the insurance premium (Productivity Commission Australia, 1998). This study also found several socioeconomic status variables to be significantly associated with car crash injury, however, their inclusion in the model for insurance and car crash risk did not change the significant effect of insurance. Another possible explanation for the association between insurance and car crash injury is that drivers who are uninsured take more risks. An insurance market theory known as ‘propitious selection’ suggests that risk-averse individuals are more likely to take precautions to reduce risks when driving as well as purchase insurance (Hemenway, 1993). This hypothesis is supported by descriptive survey data indicating that people without insurance are more likely to engage in risky lifestyle behaviours and risky driving behaviours (Hemenway, 1992; Clyde et al., 1996; Preusser and Williams, 1991; Evans and Wasielewski, 1983). This study found that uninsured drivers are more likely than insured drivers to be injured in a car crash, supporting the suggestion that uninsured drivers engage in risky driving and raising the possibility that insurance is acting as a proxy variable for risk taking behaviour. Our analysis included many risky behaviour variables but they did not diminish the effect of insurance. Further research is needed to clarify the relationship between insurance status and car crash injury, as well as insurance status and risk taking behaviours and socio-economic status, to identify opportunities for injury prevention. This population-based case-control study provides new evidence that uninsured drivers are at higher risk of car crash injury than insured drivers after adjustment for multiple confounders. It is possible this relationship is subject to residual
confounding or bias. Despite this, the substantial proportion of uninsured motorists in the driving population makes this an important public health problem that warrants further investigation. References Bruce, C.J., Weissenberger, A.T., 2001. No-fault automobile insurance. Economica 6 (3), 1–16. Clyde, A.T., Hemenway, D., Nagurney, J.T., 1996. Seatbelt use, insurance status and hospital bad debt. J. Trauma 41 (1), 100–104. Connor, J., Norton, R., Ameratunga, S., et al., 2002. Driver sleepiness and risk of serious injury to car occupants: population-based case-control study. BMJ 324, 1–5. Cubbin, C., LeClere, F.B., Smith, G.S., 2000a. Socioeconomic status and the occurrence of fatal and non-fatal injury in the United States. Am. J. Public Health 90 (1), 70–77. Cubbin, C., LeClere, F.B., Smith, G.S., 2000b. Socioeconomic status and injury mortality: individual and neighbourhood determinants. J. Epidemiol. Commun. Health 54, 517–524. Evans, L., Wasielewski, P., 1983. Risky driving related to driver and vehicle characteristics. Accident Anal. Prev. 15 (2), 121–136. Evans L., 1991. Traffic Safety and The Driver. Van Nostrand Reinhold, New York. Harper, J.S., Marine, W.M., Garrett, C.J., Lezotte, D., Lowenstein, S.R., 2000. Motor vehicle crash fatalities: a comparison of hispanic and non-hispanic motorists in Colorado. Ann. Emergency Med. 36 (6), 589–596. Hemenway, D., 1992. Propitious selection in insurance. J. Risk Uncertainty 5, 247–251. Hemenway, D., 1993. Nervous Nellies and Dangerous Dans. J. Pol. Anal. Manage. 12 (2), 359–363. Hunstead L., 1999. Characteristics of the uninsured motorist. Californian Department of Insurance, Sacramento. Available at: http://www.insurance.ca.gov/PRP. Accessed March 2002. Insurance Research Council of America, 2000. Uninsured Motorists, 2000 ed. Available at: http://www.ircweb.org. Accessed March 2002. Laflamme, L., Engstrom, K., 2002. Socioeconomic differences in Swedish children and adolescents injured in road traffic accidents: cross sectional study. Br. Med. J. 324, 396–397. Loomis, D.P., 1991. Occupation, industry, and fatal motor vehicle crashes in 20 states, 1986–1987. Am. J. Public Health 81 (6), 733–735. McDonald W.A., 1994. Young driver research program: a review of information on young driver performance characteristics and capacities. Federal Government Road Safety Initiative CR 129, Federal Office of Road Safety, Canberra, Australia. Organisation for Economic Co-operation and Development (OECD), 1990. Road Transport Research: Automobile Insurance and Road Accident Prevention. OECD, Paris. Preusser, D.F., Williams, A.F., 1991. Characteristics of belted and unbelted drivers. Accident Anal. Prev. 23 (6), 475–482. Productivity Commission, 1998. Inquiry report into vehicle and recreational marine craft repair and insurance industries. Available at: http://www.pc.gov.au/ic/inquiry. Accessed May 2002. Rothman, K., Greenland, S., 1998. Modern Epidemiology. Lippencott-Raven, Philadelphia. Wacholder, S., McLaughlin, J.K., Silverman, D.T., Mandel, J.S., 1992. Selection of controls in case-control studies. I. Principles. Am. J. Epidemiol. 135 (9), 1019–1028. Wilde, G.J., 1994. Target Risk. PDE Publications, Canada.