Accid. Anal. & Prev. Vol. 24. No. 4. pp. 329-338. Printed in Great Britain.
ARE RADAR
1992
KlOl-4575/92 8 1992 Pergamon
DETECTOR USERS LESS SAFE THAN NONUSERS?
PETER J. COOPER, YANLING Zuo, Insurance Corporation
$5.&l+ .OO Press Ltd.
and MARIO PINILI
of British Columbia, Vancouver, British Columbia, Canada
(Received 23 November
1990; in revised form 2 April 1991)
Abstract-One hundred and seventy-four drivers who had purchased special insurance coverage during 1988-1989 for in-vehicle equipment that included a radar detector were compared to a similarly sized and sociodemographically stratified driver population random sample. It was found that the radar detector owners had significantly more accident claims and speeding convictions during the period 1986-1989 than those representing the general driver population.
INTRODUCTION
Vehicular speed has a twofold influence on traffic accident risk. It affects both probability of involvement and severity of consequence. In terms of involvement, it was established over 20 years ago (Solomon 1964) that deviation from the mean travel speed of surrounding traffic carried an increased risk for the “offending” vehicle. This risk apparently works in both directions: travelling slower than surrounding traffic is just as bad as travelling faster. A vehicle travelling significantly faster or slower than surrounding traffic is either passing or being passed continually. Such overtaking creates opportunities for sideswipe and head-on crashes that would otherwise not exist. In addition to this, it can readily be appreciated that the faster a vehicle is travelling, the less time there is for a driver to react to an emergency-such as a child darting out across a road. Likewise, the unmitigated effect of speed on accident severity is clear. To bring a vehicle to a stop requires that it lose momentum proportional to the velocity at which it was initially travelling. If the stop is sudden, as in a frontal crash, then the occupant must undergo the same velocity change often in less than one half of one second. The greater the velocity change, the greater the probability of injury. In summary, there is little doubt as to the general direction in which speed influences the probability and severity of crashes. The issue that has concerned enforcement agencies is: does radar detector use encourage individual travel speeds sufficiently above the mean traffic speed that the probability and likely severity of accidents are significantly affected? Several studies have been conducted to assess the impact of detector use on speed. Typically, such studies have made use of nondetectable speed measuring devices such as in-pavement loop detectors or retuned radar in order to establish normal mean traffic speeds. Regular police radar is then activated and individual vehicle speed changes are observed. Those vehicles that suddenly slow by 8 km/h (5 mph) or more are assumed to be using radar detectors. Using such a methodology, studies in Texas, Virginia, and Maryland (Ciccone, Goodson, and Pollner 1987) have shown that likely radar-equipped vehicles make up an increasingly larger proportion of those vehicles exceeding the speed limit, as the size of this excess increases. For example, in Maryland where detectors were legal at the time of the study, about 20% of those travelling 10 mph or more over the highway speed limit were judged to have detectors, but over 40% of those exceeding the limit by 20 mph or more were likely detector users. A more recent study in Virginia and Maryland (Freedman, Williams, Teed, and Lund 1990) employed radar-detector detectors to provide a more reliable measure of usage for the devices. As with the previous study, the percentage of drivers using radar detectors who were exceeding the speed limit was significantly higher than was the percentage of those without detectors. 329
P. J. COOPERet al.
330
However, while the connection between use of radar detectors and excessive travel speed has been reasonably well established, it is not clear whether higher speeds are an expected outcome of detector use or whether those who own such devices would be fast drivers in any event, even in the face of possible apprehension. A recent survey of U.S. drivers found that over half of those who admitted using a radar detector suggested that they would probably not drive as fast without the device (Opinion Research Corporation 1988). But if 40% of those travelling 20 mph or more over the speed limit are detector users then, conversely, 60% are not. This latter group apparently chooses to speed even without the “protection” afforded by radar detectors. Presumably such drivers feel that the risk of apprehension and the subsequent consequences are outweighed by the momentary advantages in travel time that they gain. Regardless of such issues, the initial consideration from a safety viewpoint is not by how much the speed limit is exceeded by the detector users but to what extent they surpass the average speed of the surrounding traffic. In the Maryland and Virginia studies, for example, the speed limit in the study areas was 55 mph, but mean travel speed (prior to police radar activation) was 60 mph. The percentage of all vehicles represented by those apparently with detectors who were travelling at more than 10 mph over the mean traffic speed (the critical range on Soloman’s risk curve) was not specified by the authors but could be estimated from the figures presented to be probably under 5%. This is not a level of involvement which would suggest any great safety benefits to be derived from detector bans even if such were 100% effective in eliminating their use. On this latter subject, Ciccone et al. (1987) noted that the proportions of speeding vehicles in both Virginia and Maryland that responded suspiciously to police radar activation were essentially the same. This result was interesting since radar detectors were illegal in Virginia! They thus concluded that the effect of the detector ban was limited. So, why should we be concerned about who uses radar detectors? First comes the obvious question of legal or ethical consistency. If we wish to support a justification for setting speed limits in the first place, then we cannot logically sanction the purposeful circumvention of these same laws. Radar detectors exist primarily to allow drivers to exceed posted speed limits and avoid apprehension. Their presence has been compared by police to burglary tools. It can reasonably be assumed that if you carry burglary tools you intend, at some time, to commit a burglary. Second, and perhaps of more tangible consequence, is the possibility that radar detector use may represent an indicator or “flag” with which to identify high-risk drivers. In other words, while radar detectors may not themselves be accident-inducing, are those predisposed to buy them likely to be unsafe drivers-that is, overrepresented in the accident statistics? Interestingly, the radar detector manufacturers have been trying to defuse the antidetector argument by claiming that those who use detectors are actually better drivers and have fewer accidents per mile driven than the nonowners or those in the general population (Overend 1988). Is such a claim justified? There is a dearth of information in the literature concerning the connection between radar detector use and actual accident frequency. But, it is difficult to test the claims of the prodetector lobby without more solid data concerning safety impact. In 1988 the U.S. Federal Highway Administration denied a petition to ban detectors on vehicles used for interstate commerce, and the policy of a major insurance company of refusing to insure vehicle owners with radar detectors was declared by a Maryland court to be an unfair trade practice. The purpose of the research reported in this paper was to address the above question: are radar detector users less safe than nonusers or, at least, do they have worse driving records than a sample of the general driving population? METHODOLOGY
AND
DATA
REDUCTION
Sample selection
The only records kept by the Insurance Corporation of British Columbia in which radar detectors are specifically identified were, until 1990 when insurance of detectors
Are radar detector users less safe than nonusers?
331
was discontinued, applications for special coverage of additional in-car equipment not covered under the comprehensive policy limit on contents. All special coverage policies taken out during 1988 or 1989 with the Insurance Corporation and involving privately owned passenger vehicles were examined for this study. Those where the policy purchaser listed a radar detector among the additional equipment to be covered were extracted, and the vehicle/policy records were checked to see whether the vehicles had been registered to the same person who purchased the insurance for a period of at least four uninterrupted prior years, that there were no other principal drivers listed, and that the type of driving covered by the policy had not changed. This process resulted in a sample of 175 persons who owned a radar detector in either 1988 or 1989. Whether all of these drivers had owned detectors in all previous years could not be determined. The next step was to create a control group of drivers randomly selected from the general population. Using the SAS statistical software function RANUNI, we selected a random sample of 175 drivers from the 2.8 million drivers on the 1989 provincial drivers file. We then compared all the known sociodemographic characteristics of the two groups to test compatibility. Sample compatibility
Because of the somewhat unique class of drivers (i.e. having special insurance coverage) from which the radar detector sample was drawn, there was an anticipation in advance of the random selection process that the “radar” group might not prove to be representative of the driving population at large. Indeed, four variables were found to be significantly different among the two groups: 1. driver age group (x? = 73.1, d.f. = 4, p = .OOO); 2. driver gender (x2 = 34.3, d.f. = 1, p = .OOO); 3. insured vehicle rate group-an indicator of vehicle value (xz = 28.3, d.f. = 2, p = .OOO);and 4. insurance rate class-an indicator of amount of driving (xZ = 61.9, d.f. = 2, p = ,000).
The differences were probably not so much related to radar detector ownership as they were associated with the ownership of other, more expensive add-on equipment such as special stereos or cellular telephones. The people in our radar group were more likely to be younger (average age of 31.5 years as compared to 42.9 years for the population sample) and male (85.4% as compared to 56.5%). They were more likely to own expensive cars (11.5% had cars valued at over $20,000 as compared with only 4.5% for those in the population sample), and they were more likely to drive either to/from work or as part of their work (69.7% as compared to 48.9%). These differences were so marked between the two groups that, because the sample size was relatively modest, we felt any differences in accident or conviction rates would have been totally overshadowed in a multivariate analysis. The only viable solution was to control for these important variables in the first place. Therefore, we went back to the beginning and undertook a stratified random sampling exercise to produce a population sample with the identical distribution to the radar group for all four of the above variables. Care was taken to ensure that none of the drivers randomly selected from the population was also a member of the radar group. During this final matching process, one radar group driver had to be dropped due to missing driver record data and thus the stratified sample size was reduced to 174. The final distributions of the stratification variables in both the radar and population groups are shown in Table 1. The variable categories shown in Table 1 are somewhat collapsed for ease of presentation. In fact, sample matching was done using 8 age categories, 11 rate class categories and 19 rate group categories. Controlling for exposure
The “rate class” variable we used represented essentially three levels of vehicle use: pleasure only, to/from work, and business. Most insurance companies use a similar type AAP 24:4-0
332
P. J. COOPERet al. Table 1. Final distribution of stratification variables in both radar and population groups Gender Male: Female: Age 5 24 25-27 28-30 31-42 243 Rate class (type of driving) Pleasure only To!from work Business Rate group (current value of car) Low (<$S,OOO) Medium (%.5,000-$20.000) High (>$20,000)
n (%) 151 (86.8) 23 (13.2) 42 41 26 36 29
(24.1) (23.6) (14.9) (20.7) (16.7)
53 (30.5) 86 (49.4) 35 (20.1) 75 (43.1) 79 (45.4) 20 (11.5)
of classification as a surrogate for travel exposure. The more a person drives, the greater his/her risk of being involved in a crash (all other things being equal). But how valid is the assumption that such usage categories represent real exposure distinctions? Just because a person drives to and from work each day does not necessarily mean that he will put more kilometers on the vehicle than does someone who drives only for “pleasure.” We needed to test such assumptions in order to ensure that differences in accident frequency between the radar and control groups were not due to different levels of exposure. In order to accomplish this, we made use of another aspect of the Insurance Corporation’s operation. When a person brings a vehicle in to make a damage claim, the estimator generally records the odometer reading. As with the special coverage (radar) group, we considered only cases where the same owner had been insured for an uninterrupted period but, in this case, six months to four years following purchase of the vehicle when new. Again, cases where the insured use (rate class) had changed over the life of the vehicle were dropped. In this way, 823 cases were collected out of about 10,000 claims submitted during a one-month period. The odometer readings were divided by the number of months since purchase to arrive at an average monthly kilometers of travel. Analysis of variance techniques for unbalanced cell size (SAS GLM procedure) were used to test whether the average monthly kilometers of travel were different among the three rate classes. The results (Table 2) showed a significant difference in travel by
Table 2. Relation between insured rate class and actual amount of driving Analysis of Variance Procedure [Dependent
Variable: Exposure (km/month)]
Source of variation
Degrees of freedom
Sum of squares
Mean square
F value
Prob. value
Rate class Error Total
2 821 823
25412764.52 426531469.74 451944234.26
12706382.26 519526.76
24.46
0.0001
Comparing
pairs of rate class means [Student-Newman-Keuls ~km/month~l
Test for Variable:
Grouping
Mean
n
Rate class
A A B
1368.16 1476.17 1947.91
270 452 102
Pleasure only Work & pleasure Business
Note: (Means with the same letter are not significantly different at alpha = 0.05)
Exposure
Are radar detector users less safe than nonusers?
333
rate class. Pairwise comparisons of the means of the monthly kilometers for the three rate classes were carried out. The results in Table 2 indicated that the business group had a significantly greater amount of monthly kilometers of travel then the other two groups. “Work and pleasure” drivers had a greater mean travel amount than did “pleasure only” drivers but the difference was not significant. As a confi~ation of likely equivalence of travel exposure between radar and control groups, we were able to find 42 out of the 348 drivers in our study (174 x 2) who had, in the past, submitted claims on their vehicles and who had purchased their vehicles new within the previous four years, and for whom an estimator’s form with odometer reading could be found in the old document records. Of these 42, 30 were in the radar group and 12 were in the control group. After calculating the average number of kilometers per month for all 42 cases and normalizing using an inverse transformation, we calculated means and standard deviations for both groups and compared the results in a students’ t test. The results showed a slightly higher average monthly travel for the control group (2,058 km per month vs. 1,711 for the radar group) but the difference was not significant (t = 1,037, p = .282). Thus the results of a specific travel exposure comparison on a subset (12%) of the sample confirmed the assumption based on stratification by rate class; that is, the amount of travel undertaken by our radar group drivers was probably not different from that undertaken by those in the stratified population sample. The stratification by rate class seemed to be a valid method to control for travel exposure in samples of this size. Analysis procedure
The foregoing methodology allowed us to arrive at two compatible samples of 174 drivers each whose only known difference was that the members of one group all had insured radar detectors while none of the members of the other group had. This did not, of course, mean that no one in the population sample had a radar detector, but it seemed reasonable to assume that ownership of the devices in the population sample would be substantially less than 100% given recent estimates from other North American jurisdictions of about 5% radar detector use among drivers of specialty or sport cars (Freedman et al. 1990). The two groups of drivers were then compared on the basis of total past accident and conviction experience using the nonparametric chi-square statistic. Ten categorical traffic safety-related variables (Table 3) were defined, and the data were cross-classified into ten two-dimensional tables using variable categories and group (radar or control). For each of the tables, a chi-square test was performed to examine the probabiIity that group membership was related to variable level. Where significant (p < .Ol) associations were found, plots of proportions were drawn in order to examine the nature of the relationships. Three sources of data were utilized in defining the safety-related variables: 1. police-attended accidents where computerized accident forms were available on a historical basis from the provincial Motor Vehicle Branch (MVB); 2. computerized insurance claim records at the Insurance Corporation concerning traffic accident-related events (there are many more such claims than there are police-attended crashes) and assignment of responsibihty (fault); and 3. computerized driver record information concerning traffic law violation convictions (also available from the MVB). Linkage among the three data sources for both radar and control groups was accomplished using driver license number and policy (plate) number of the insured vehicle. We also ensured that the age and gender of each driver in question were consistent across databases. Only cases where there was a consistent historical connection between driver and vehicle were accepted. For example, the driver whose record was enumerated had to be the registered owner of the same insured vehicle during the entire period of record
334
P. J. COOPERet al. Table 3. Definitions of variables employed Variable Variable Variable Variable Variable
1 (Vl): 2 (V2): 3 (V3): 4 (V4): 5 (V5):
Variable Variable Variable Variable Variable
6 (V6): 7 (V7): 8 (V8): 9 (V9): 10 (VlO):
Police-attended accidents (police report data) Police-attended accidents (Vl) where speed was reported as a factor Police-attended accidents (Vl) where alcohol was reported as a factor Insurance claims (BI or PD) resulting from motor vehicle accidents Accident claims (V4) where the driver was assigned 50% or more responsibility Accident claims (V4) on weekends and involving no other vehicles* Accident claims (V4) involving casualty (BI claims) Criminal code traffic conviction& Speeding convictions not resulting from accidents Other moving violation convictions not resulting from accidents
*Single-vehicle-weekend claims were chosen as an alternate indicator for alcohol involvement, since the numbers of such claims were much higher than those for V3. tPredominantly drinking-driving offenses.
examined, back to 1984. There could be no lapses in insurance coverage, no other principal drivers listed for that vehicle, and no change in the type of driving (rate class) covered. Such restrictions resulted in the rejection of nine out of every eleven drivers considered. Obtaining our 348 drivers thus entailed an examination of almost 2,000 individual records. Accident fault was established in each case from insurance records where a 50percent-or-higher responsibility assessment by adjusters was taken to be indicative of culpability on the part of the driver in question. RESULTS
AND
DISCUSSION
The results of the chi-square tests are shown in Table 4. Four of the ten variables were found to be significant at the p < .Ol level. These were: 1. 2. 3. 4.
V4 V5 V6 V9
-
Accident-related claims Culpable accident-related claims Single-vehicle, weekend, accident-related claims Nonaccident-related speeding convictions
Accident-related claims involving single vehicles and occurring on weekends were selected as a surrogate for alcohol-involved crashes. Definition of a surrogate for alcoholrelated accidents was considered worthwhile due to the fact that police in British Columbia tend to underreport alcohol involvement (Donelson, Beirness, Haas, and Walsh 1989) and also because the number of police-attended accidents overall was considerably less than the number of accident-related claims. With each of the above four variables, the radar detector owners were found to be Table 4. Results of chi-square tests between sample groups by variables Proportion of drivers with one or more occurrences Variables Group Group Group Group Group Group Group Group Group Group
and and and and and and and and and and
Vl V2 V3 V4 V5 V6 V7 V8 V9 VlO
d.f.
x2 value
Probability
Radar
Control
2 1 1 4 3 2 2 1 7 4
0.948 0.234 0.203 22.162 14.763 10.636 3.290 0.114 24.541 10.628
0.623 0.628 0.652 0.000 0.002 0.005 0.193 0.736 0.001 0.031
.293 .057 .Oll .638 ,391 .184 .190 ,023 .793 ,563
.247 .046 .017 .489 .213 .075 ,121 .029 ,655 ,391
Are radar detector users less safe than nonusers?
335
‘a.
x. .... l.
*a..
... ...
.I....
x.. Y. -.... -..
l.,
..
..--.-
..“.......“‘.“............................~
I
I
1
4 Accident Related Claims
Fig. 1. Proportion of drivers having 1 to 4 accident-related
claims for radar and control groups.
associated with higher values than their control-group counterparts. The proportions having different levels of claims and convictions by group membership are shown in Figs. 1 through 4. The last significant relationship (V9 vs. group) is interesting since it indicates that radar detector users may have significantly higher levels of speeding convictions than nonusers in spite of their ownership of a device purchased to avoid being caught speeding. Such a result does not mean that detectors are necessarily inefficient. The difference probably reflects a more prevalent speeding behaviour on the part of the radar detector owners who are then caught occasionally-perhaps with their detectors turned off.
I
1
I
I
2
3
Culpable Accident Related Claims Fig. 2. Proportion of drivers having 1 to 3 culpable accident-related
claims for radar and control groups.
P. J. COOPERet al.
336
15-
10-
.I..,...‘%.___ ......_. Q...... .I.....__ .....__. __ ....____ *+_.__ Y-e ..._.._ +*. ...___. ..%_.._, --...._____ _-.__. ...___ ......._....__ ...._._. ......____ ‘.
5-
OI 1
I 2 Single Vehicle Weekend Accident Related Ck4mS
Fig. 3. Proportion of drivers having f or 2 single vehicle, weekend, accident-related and control groups.
claims for radar
There is, of course, another possible explanation for the above result, and that is that most of the speeding convictions were obtained prior to purchase of detectors-in fact, these convictions may have led to acquisition of the devices in the first place. The reason such a possibility exists is that we were able to fix radar detector ownership only at a single point in time-towards the end of the four-year period of record employed. Thus, it is possible that some of the high accident claim as well as conviction experience occurred prior to radar detector use. If the above scenario were operative, we would expect a plot of speeding conviction
i 1
t 2
I
I
/
I
I
3
4
5
6
7
Non-accident Related Speed Conviction Fig. 4. Proportion
of drivers having 1 to 7 non-accident-related control groups.
speeding convictions for radar and
Are radar detector users less safe than nonusers?
331
-I 0
0
Fig. 5. Proportion
of drivers having non-accident-related speeding convictions by year from 1984 to 1988 for radar and control groups.
probability over time for the radar group members to show a descending characteristic during the last few years of record. After all, as we have previously suggested, it is unlikely that radar detectors themselves cause speeding. Rather, it is likely that they are utilized by persons who wish to purchase immunity from the legal consequences of their existing predilection to speed. If a substantial portion of our radar owners group had purchased a detector only at some time during rather than prior to the five years of record, then we might expect the annual total of speeding convictions to be higher near the beginning than towards the end of that period. To address this question, the speeding conviction data were cross-classified into a three-dimensional table according to year, group, and V9. However, since the number of observations in some cells were small, no statistical tests with adequate power could be used. So we changed the categorization of the variable into two levels, with “0” representing no convictions and “1” representing one or more convictions. The proportions of drivers having speeding convictions in different years for the radar group and the general population group are shown in Fig. 5. Figure 5 shows that the proportion having a speeding conviction by year for both those in the radar detector group and those in the population group has remained relatively stable with no downward trend. There was no significant variation by year (chi-square values of 0.62 and 3.86 respectively, d.f. = 4, p > .05) over the five years of record. No evidence was thus apparent to support a contention that conviction totals for radar detector owners may have been largely accumulated prior to purchase of the detectors. Likewise, when the three significant, claim-related variables were similarly crossclassified according to year, group, and variable, no significant downward trends over time were found. In other words, the higher total claim numbers for radar detector owners were not due to disproportionately higher counts in the early years of record when direct evidence of detector ownership was not available. CONCLUSIONS
While we cannot draw conclusions about the effects of radar detector use among the general population of drivers owing to the nonrepresentative nature of the detector
338
P. J. COOPERet al.
owners which we were able to identify through insurance records, we were able to isolate the effects associated with detector ownership for a specific sociodemographic class of drivers. These were predominantly male drivers between the ages of 21 and 42 who drove for business purposes or to/from work, typically in medium- or higher-priced cars. For such drivers, we found radar detector ownership to be associated with significantly higher rates of accident-related claims per year than were similarly classified drivers randomly selected from the general population. The detector owners also had significantly higher rates of accident claims for which the driver was judged at fault and significantly higher rates of accident claims occurring on weekends where only their own vehicles were involved (a possible indicator of alcohol involvement). Finally, we found that the use of radar detectors by these drivers did not make them immune from apprehension by the police. They had a significantly higher rate of speeding convictions (not resulting from accidents) over the four-year period of record than did the drivers selected from the general population. In obtaining the above results, every effort was made to control for travel distance. Within the limits of available data, we can state that the higher accident claims and convictions for the radar users were probably not a result of greater driving exposure. The overall conclusion must be, therefore, that drivers in the sociodemographic group that we investigated who own radar detectors were probably less safe than their counterparts in the general driving population as evidenced by accumulation of higher levels of accident-related claims. In stating this conclusion, we do not imply any direct cause-and-effect relationship between radar detector ownership and safety. Rather, we suggest that ownership of a detector may be indicative of a predisposition towards more risky driving behaviour on the part of those who avail themselves of the device’s perceived protection. Acknowledgemenis-The authors would like to acknowledge the contributions of staff in the Claims Division of the Insurance Corporation of British Columbia and also the assistance of John Pullyblank, who was involved in the initial project development. REFERENCES Ciccone, M. A.; Goodson. M.; Pollner. J. Radar detectors and speeds in Maryland and Virginia. Washington, DC: Insurance Institute for Highway Safety; 1987. Donelson. A. C.; Beirness. D. J.; Haas, G. C.: Walsh. P. J. The role of alcohol in fatal traffic crashes; British Columbia. 1985-1986. Ottawa, Ontario: Traffic Injury Research Foundation of Canada; 1989. Freedman, M.; Williams. A. F.; Teed. N.: Lund, A. K. Radar detector use and speeds in Maryland and Virginia. Washington, DC: Insurance Institute for Highway Safety; 1990. Opinion Research Corporation. A survey about radar detectors and driving behaviour. Washington. DC: Insurance Institute for Highway Safety; 1988. Overend. R. B. Radar detectors spark controversy at TRB meeting. Traffic Safety 88:21-23; 1988. Solomon, D. Accidents on main rural highways related to speed, driver and vehicle. Washington. DC: Bureau of Public Roads: 1964.