Risk Factors for All-Terrain Vehicle Injuries: A National Case-Control Study

American Journal of Epidemiology
Copyright © 2001 by The Johns Hopkins University School of Hygiene and Public Health
All rights reserved

Vol. 153, No. 11
Printed in U.S.A.

All-Terrain Vehicle Risk Factors Rodgers and Adler

Risk Factors for All-Terrain Vehicle Injuries: A National Case-Control Study

Gregory B. Rodgers1 and Prowpit Adler2
A case-control study design was used to determine and quantify all-terrain vehicle (ATV) risk factors. The
analysis was based on the results of two national probability surveys conducted in 1997: a survey of injured ATV
drivers treated in hospital emergency departments and a survey of the general population of ATV users. Cases
were drawn from the injury survey; controls (ATV drivers who had not been injured) were drawn from the user
survey. Risk factors were quantified by means of a binary logistic regression analysis. After adjustment for
covariates, injury risks were systematically related to a number of driver characteristics (age, gender, driving
experience), driver use patterns (monthly driving times, recreational vs. nonrecreational use), and vehicle
characteristics (number of wheels, engine size). The results of the analysis suggest that future safety efforts
should focus on reducing child injuries, getting new drivers to participate in hands-on training programs, and
encouraging consumers to dispose of the three-wheel ATVs still in use. Am J Epidemiol 2001;153:1112–18.
accidents; case-control studies; logistic models; off-road motor vehicles; risk factors; wounds and injuries

All-terrain vehicles (ATVs) are three- and four-wheel
motorized vehicles intended for use on various types of
unpaved terrain. They have large low-pressure tires, seats
designed to be straddled, handlebars for steering, and motorcycle-type engines. Engine sizes range from 50 to 500 cm3
of displacement, and vehicle weights range from about 100
to 600 pounds (1 pound = 0.454 kg).
Concern about the safety of ATVs grew during the 1980s
as the number of ATV-related injuries and deaths rose. Annual
estimates of ATV-related injuries treated in US hospital emergency departments increased from an estimated 32,000 in
1983 to 106,000 in 1985, an increase of about 230 percent in
just 2 years (1). There were also an estimated 295 deaths in
1985, the first year in which national estimates were available
(2). Given an estimated 1.9 million ATVs in use in 1985, there
were about 5,580 injuries treated in emergency departments
and 15.5 deaths for every 100,000 ATVs in use.
In 1985, the Consumer Product Safety Commission initiated a regulatory proceeding to evaluate and address ATV
hazards and ultimately settled with the industry (3–5). The
terms of the settlement were specified in the 1988 consent
decrees between the Department of Justice and ATV distributors. The consent decrees, which expired in April 1998,
included agreements by the ATV distributors to 1) stop selling new three-wheel ATVs; 2) put into effect more stringent

driver age requirements; 3) implement a nationwide training
program approved by the Consumer Product Safety
Commission and to provide free hands-on training to all
new buyers and their immediate families; and 4) develop a
voluntary standard to make ATVs safer.
The consent decrees appear to have played an important
role in reducing rates of injury and death (6). By 1997, the
rate of injuries treated in emergency departments (injury
rate) had fallen to about 1,490 injuries per 100,000 ATVs in
use, a decline of over 70 percent from the 1985 level.
Similarly, the death rate declined to about 8.3 deaths per
100,000 ATVs in use, a decrease of almost 50 percent from
the 1985 level (1, 2).
Despite the reduction in the overall injury and death
rates, however, the number of injuries and deaths remains
high. During 1997, for example, there were an estimated
54,000 injuries treated in emergency departments and 300
deaths involving ATVs; more than 35 percent of these
injuries and deaths have involved children aged less than 16
years (1, 2).
Because of the continuing large numbers of ATV-related
injuries and deaths, and because the consent decrees were
set to expire in April 1998, the Commission sponsored a
1997 study of ATV risk and hazard patterns to determine
what, if any, further regulatory actions might be warranted.
Central to that effort was a case-control study based on the
results of two national probability surveys. This article presents the results of the case-control study.

Received for publication January 4, 2000, and accepted for publication October 20, 2000.
Abbreviations: ATV, all-terrain vehicle; CI, confidence interval; OR,
odds ratio.
1
Directorate for Economic Analysis, US Consumer Product
Safety Commission, Washington, DC.
2
Directorate for Epidemiology, US Consumer Product Safety
Commission, Washington, DC.
Reprint requests to Dr. Gregory Rodgers, US Consumer Product
Safety Commission, Washington, DC 20207 (e-mail: grodgers@cpsc.
gov).

MATERIALS AND METHODS
Data collection
Injury and user surveys. Cases consisted of a sample of
injured ATV drivers treated in hospital emergency departments and reported through the US Consumer Product Safety
1112

All-Terrain Vehicle Risk Factors

Commission’s National Electronic Injury Surveillance System
from May 1, 1997, to August 31, 1997.
The National Electronic Injury Surveillance System is a
stratified national probability sample of US hospitals; at the
time of the ATV survey, it consisted of 101 of the approximately 5,400 US hospitals that had at least six beds and provided 24-hour emergency service (7, 8). The sample, which
was updated in January 1997 (9), is stratified by hospital
size, based on the annual number of emergency department
visits. Additionally, to ensure a wide geographic coverage,
hospitals within each stratum were ordered by state and zip
codes and selected systematically.
At the time of the survey, National Electronic Injury
Surveillance System hospitals identified all injuries involving
ATVs, with the exception of injuries that occurred when ATVs
were being used for commercial or occupational tasks other
than for farming or ranching. Coders at the hospitals collected
information on a number of variables, including date of treatment, age and sex of victim, injury diagnosis, body part
injured, disposition of the case (i.e., treated and released or
admitted), and as many as two products that were involved in
the injury. A brief narrative describing the injury was also
included. The Consumer Product Safety Commission collected the information each evening electronically.
All reported ATV injuries during the study period were
assigned for follow-up in-depth telephone interviews. The
in-depth interviews were conducted as soon after the injury
as possible, usually within 1 month of the injury.
Controls were drawn from a national telephone survey of
ATV users conducted for the Consumer Product Safety
Commission by Abt Associates, Inc., from September 15,
1997, to November 18, 1997 (10). Eligible respondents
included drivers who had used an ATV owned by the driver’s household at least once during the preceding 12-month
time period. For households with multiple drivers, the driver
who had the most recent birthday was selected to be interviewed. The survey was designed to provide a national
probability sample of about 500 respondents.
The user survey used a single stage list-assisted randomdigit-dialing sample design. The sample was selected using
the GENESYS Sampling System, a system which uses the
American Telephone and Telegraph (AT&T) master tape of
telephone exchanges as the basis for constructing sampling
frames (11, 12). The sample was stratified by census region.
The screenings needed to complete approximately 500 interviews were distributed among the strata using Neyman
allocation to minimize the variance of the estimated overall
eligibility rate (13). Up to eight attempts were made to
obtain an answered call for each sampled telephone number.
Questionnaire development and interviewing procedures.

Questionnaires for both surveys were developed by Consumer
Product Safety Commission staff. The injury survey obtained
detailed information about the characteristics and use patterns
of injured drivers, the ATVs being used, and the injury scenarios. The user survey collected parallel information on the
characteristics and use patterns of the driver population.
Respondents targeted for the risk analysis were the
injured drivers from the injury survey and the selected drivers from the user survey. In both surveys, however, parents
Am J Epidemiol Vol. 153, No. 11, 2001

1113

or guardians were asked to respond on behalf of children
under age 16. If respondents were not available, call-backs
were scheduled.
Survey weighting procedures.
Because the National
Electronic Injury Surveillance System is a national probability sample of hospitals, each of the injured drivers was
assigned a sample weight based on the inverse of the known
probability of selection of the hospitals in each stratum.
Weights for the completed interviews were adjusted for unit
nonresponse.
The user survey observations were also weighted to provide statistically valid national estimates of ATV drivers
from ATV-owning households. Each of the sampled households received a weight that was adjusted for the number of
telephones in the household and unit nonresponse. Because
only one driver per household was interviewed, the household weight was multiplied by the number of household drivers, yielding a driver population weight that reflected the
total number of ATV drivers in the United States (14).
Data adjustments. To avoid the possibility of bias in the
comparison of cases and controls, several adjustments to the
databases were required prior to analysis. First, because the
user survey was limited to drivers from ATV-owning households, the cases were limited to injured drivers who owned,
or whose household owned, the ATV involved.
Second, because the sample of controls was intended to
represent ATV drivers who had not been injured, user observations were excluded from the analysis if the driver had
suffered an ATV injury that required emergency department
medical treatment during the overall May to November
study period. Finally, user survey respondents who used
their ATVs entirely for commercial or occupational tasks
other than farming or ranching were excluded from the controls because the National Electronic Injury Surveillance
System does not routinely collect data on those types of
occupational injuries.
Because of these adjustments, inferences are limited to drivers from ATV-owning households, except for those who use
their ATV(s) entirely in nonfarm occupational applications.
Statistical analysis

SUDAAN software (Research Triangle Institute,
Research Triangle Park, North Carolina) was used in the statistical analysis because of the complex design of both surveys. In the National Electronic Injury Surveillance System
survey, the strata were hospital sizes, the primary sampling
units were the hospitals, and the secondary sampling units
were the injured drivers. Injured drivers were therefore
modeled in SUDAAN software as cluster samples of injured
drivers from the hospitals. In the exposure survey, the strata
were geographic regions, the primary sampling units were
households, and the secondary sampling units were uninjured persons in the household. Uninjured drivers were
therefore modeled as cluster samples of uninjured drivers
from the household.
Thus, both samples were modeled in SUDAAN software
as cluster samples within distinct strata and primary sampling units; the SUDAAN design parameter was “without

1114

Rodgers and Adler

replacement,” a design which was appropriate for each sample. Because both surveys were mutually exclusive in their
population of interest, they were appended (i.e., stacked)
into a single data set representing all ATV users. Variance
estimation was based on Taylor linearization methods (15).
A univariate comparison of the cases and controls was
conducted as a preliminary evaluation of possible risk factors. Crude odds ratios, with 95 percent confidence intervals, were calculated.
The principal method of analysis was a binary logistic
regression analysis of the case-control data. The explanatory
variables included various driver characteristics and use patterns, as well as characteristics of the ATVs driven. Age was
included as a polytomous categorical variable, with five discrete age categories. The driver’s gender and the ATV’s
number of wheels (three or four) were included as dichotomous variables. The remaining variables were included as
continuous variables. Various scales (i.e., functional forms)
of the continuous variables and possible interactions
between the risk factors were explored.
In addition to estimating adjusted odds ratios, we estimated annual injury probabilities to illustrate the relation
between selected risk variables and risk. Injury probabilities
were estimated using an adjustment procedure described by
Maddala (16) and by Hosmer and Lemeshow (17) that combined the results of the regression analysis with additional
independent information on the number of injured and uninjured ATV drivers to account for the relative sampling proportions within the cases and controls (1, 18).
RESULTS

sis were not affected by the inclusion or exclusion of these
observations.
The denominator degrees of freedom, which equals the
number of primary sampling units less the number of strata,
was 451 (15, 19). This reasonably large value suggests that
the analysis should provide enough power, even after adjusting for a number of explanatory variables, to detect and to
test for meaningful differences in risk factors.
Descriptive evaluation

The final database used in the analysis included 133 cases
and 460 controls. Table 1 shows the characteristics of driver
injuries. Almost 16 percent resulted in hospital admission,
compared with only about 4 percent of all injuries reported
through the National Electronic Injury Surveillance System.
Just over 11 percent of the ATV injuries involved the head.
The characteristics of the cases and controls, along with
crude odds ratios for the weighted data, are presented in
table 2. With the exception of the gender and wheels variables, information on the variables shown in the table was
collected as continuous data. The variable categories shown
in table 2 were chosen only for expository purposes. The
univariate results indicate that the cases are more likely to
involve young drivers and children, males, drivers who use

TABLE 1. Characteristics of injuries, US Consumer Product
Safety Commission Nationwide All-Terrain Vehicle Injury
Survey, 1997

Survey response

A total of 487 injured riders were identified through
National Electronic Injury Surveillance System hospitals
during the study period. All were assigned for follow-up
telephone interviews. A total of 353 interviews with injured
victims (or their representatives) were completed, yielding
an overall response rate of 72.5 percent.
After data adjustments, cases were limited to 133 of the
353 completed case interviews. Of the 220 cases dropped
from the analysis, 34 were determined to be out-of-scope
(e.g., the ATV was not being operated at the time of injury),
92 involved passengers, and 79 involved nonowners.
Finally, 15 cases were dropped because the ownership of the
vehicle could not be determined.
The user survey completed 464 interviews with drivers
(or their parents). The screening response rate, the proportion of numbers successfully screened to determine household eligibility, was 80.4 percent. The interview response
rate, the proportion of eligible respondents who completed
interviews, was 82.7 percent. The overall response rate, estimated as the product of the screening and interview
response rates, was 66.5 percent.
Three user survey observations were excluded from the
analysis because the driver had suffered an injury that
required emergency department medical treatment, and one
was excluded because the driver used the ATV entirely for
commercial or occupational tasks. The results of the analy-

Characteristic

Type of injury (primary)
Concussion
Internal injury
Contusions/abrasions
Fracture/dislocation
Strain/sprain
Laceration
Other
Total

1.9
2.6
25.2
30.7
15.1
22.9
1.6
100.0

Body part injured (primary)
Head
Neck
Face
Upper trunk
Lower trunk
Arm/shoulder
Leg/foot
Multiple body parts
Total

11.1
1.6
6.2
5.6
7.4
39.8
25.0
3.3
100.0

Hospital disposition
Treated and released
Admitted
Total

%*

84.2
15.8
100.0

* Weighted on the basis of 133 driver injuries treated in hospital
emergency departments.

Am J Epidemiol Vol. 153, No. 11, 2001

All-Terrain Vehicle Risk Factors

1115

TABLE 2. Selected characteristics of cases and controls, US Consumer Product Safety Commission
Nationwide All-Terrain Vehicle Injury and User Surveys, 1997
Variable

Cases
No.

Age (years)
≤15
16–25
26–35
36–45
≥46

133

Gender
Male
Female

133

Wheels
Three wheels
Four wheels

133

Engine size (cm3)
<100
101–200
201–300
≥301

122

Experience (years)
≤1
1.1–4.0
4.1–8.0
8.1–12.0
≥12.1

130

Nonrecreational use
(hours out of 10)
0–2
3–7
8–10
Hours per month
0–10
11–25
26–50
≥51

Controls
%

No.

%

Crude
OR*

14.3
29.0
20.5
18.5
17.6

4.9
2.2
1.6
1.2
1.0

1.8,
0.8,
0.6,
0.4,

65.7
34.3

3.3
1.0

1.5, 7.4

15.5
84.5

2.4
1.0

1.3, 4.5

5.5
19.5
48.5
26.5

1.3
1.6
1.3
1.0

0.4, 4.4
0.6, 4.0
0.7, 2.6

8.5
22.8
15.8
24.0
28.9

3.7
1.6
1.5
1.3
1.0

1.5,
0.8,
0.6,
0.5,

53.8
28.0
18.2

3.2
1.1
1.0

1.5, 7.1
0.4, 3.1

42.6
27.7
15.4
14.3

0.7
0.5
1.3
1.0

0.3, 1.7
0.2, 1.0
0.4, 4.3

95% CI*

456
33.6
31.2
16.1
10.5
8.5

13.1
6.3
4.7
3.5

456
86.4
13.6
456
30.5
69.5
438
5.8
23.9
49.4
20.9
454
20.8
24.5
15.3
20.1
19.3

130

9.0
3.5
3.5
3.1

457
77.8
14.1
8.1

107

443
39.3
17.0
25.3
18.3

* OR, odds ratio; CI, confidence interval.

three-wheel (as opposed to four-wheel) ATVs, and drivers
who use their ATVs in recreational (as opposed to nonrecreational) activities a greater proportion of the time.
Logistic regression analysis

The results of the logistic regression analysis are presented in table 3. For the most part, they mirror the univariate findings. Additionally, because emergency department
injury risks are small on an absolute scale (i.e., generally
under 1 percent per year), the adjusted odds ratios approximate relative risks (20).
The estimated risk was highest for children under age 16
and generally declined with age. Relative to drivers “over
age 45,” the odds ratio for children was 12.0 (95 percent
Am J Epidemiol Vol. 153, No. 11, 2001

confidence interval (CI): 4.6, 31.3). To compare estimated
risks for children under age 16 with all older drivers, the
table 3 model was reestimated with age as a dichotomous
variable; in this model the odds ratio for children, relative to
drivers age 16 or older, was 3.9 (95 percent CI: 1.7, 8.7).
The estimated risk was also higher for males (odds ratio
(OR) = 3.0; 95 percent CI: 1.6, 5.5) and for riders who drove
three-wheel ATVs (OR = 3.1; 95 percent CI: 1.5, 6.4).
With the exception of the nonrecreational use variable, all
of the continuous variables were scaled as natural logarithms, transformations that substantially improved the fit of
the model. Based on the table 3 results, the estimated risk
declined with driving experience, declined with the proportion of time ATVs were used in nonrecreational applications,
and rose with engine size. Given the logarithmic scale of the

1116

Rodgers and Adler
TABLE 3. Logistic regression results, US Consumer Product Safety Commission Nationwide AllTerrain Vehicle Injury and User Surveys, 1997
Variable*

Intercept

Coefficient (SE†)

Adjusted OR†,‡

95% CI†

–8.892 (1.966)

Age group (years)
≤15
16–25
26–35
36–45
≥46

2.484
1.365
1.640
1.082
0 (0)

(0.489)
(0.417)
(0.405)
(0.388)

12.00
3.92
5.16
2.95
1.0

Gender
Male
Female

1.087 (0.319)
0 (0)

2.97
1.0

1.58, 5.55

Wheels
Three wheels
Four wheels

1.130 (0.369)
0 (0)

3.10
1.0

1.50, 6.39

ln† (engine)§

0.912 (0.293)

1.39

1.13, 1.71

ln (experience)¶

–0.404 (0.139)

0.96

0.93, 0.99

Nonrecreational hours#

–0.240 (0.039)

0.79

0.73, 0.85

ln (hours)**
[ln (hours)]2

–2.572 (0.356)
0.434 (0.061)

0.90
1.18

0.88, 0.93
1.08, 1.15

4.59,
1.73,
2.33,
1.38,

31.34
8.89
11.42
6.32

* The analysis is based on 96 cases and 421 controls with complete data.
† SE, standard error; OR, odds ratio; CI, confidence interval; ln, natural logarithm.
‡ The odds ratios of the continuous variables specified as natural logarithms depend upon the level of the variable at which the odds ratios are calculated. In these cases, the odds ratio is calculated at the mean unit value of
the variable.
§ The natural logarithm of the engine size in cm3 of displacement. The odds ratio is calculated per 100 cm3.
¶ The natural logarithm of the months of driving experience (i.e., months since learned to ride). The odds ratio
is calculated per 12 months of riding experience.
# The number of hours (of every 10) that the driver uses the all-terrain vehicle for nonrecreational purposes.
The odds ratio is calculated per hour of nonrecreational use.
** The natural logarithm of the number of hours the driver rides an all-terrain vehicle in an average month of
use. The odds ratio is calculated per hour of monthly riding.

driver experience and engine size variables, it can be shown
that a 1 percent increase in driving experience results in an
estimated risk reduction of about 0.4 percent and that a 1
percent increase in engine size increases the estimated risk
by 0.9 percent (21).
Additionally, the estimated risk exhibited a quadratic relation with hours of use in an average month. The estimated
risk was high for those who rode infrequently, declined over
a brief range of driving times, and then rose as driving times
increased. This relation seems to suggest that, while estimated risk generally rises with driving times, it also tends to
be relatively high for infrequent drivers.
A number of interaction variables were also evaluated,
including interactions between experience and number of
wheels, between experience and hours of monthly driving,
and between the age and gender of the driver. None of these
interactions significantly improved the fit or predictive ability of the model.
Figure 1 illustrates the types of estimated risk relations
that can be generated from the regression model when

additional information on the sampling proportions for the
cases and controls is available. Figure 1 shows the estimated risk relations between engine size and the number of
wheels on ATVs for a typical driver, who is assumed to be
a male aged 30 years with 8 years of driving experience
and who rides for about 25 hours per month and uses the
ATV for nonrecreational applications about 10 percent of
the time.
DISCUSSION

The results of this study suggest that ATV injury risks are
systematically related to a number of driver and vehicle
characteristics. Moreover, the relations are substantial as
well as statistically significant and help to explain, at least in
part, the observed reduction in the ATV-related injury rate
during the 1985–1997 time period.
The results of the 1997 user survey (18), when compared with those of a similar survey conducted in 1986
(22), suggest a decrease in the proportion of riders under
Am J Epidemiol Vol. 153, No. 11, 2001

All-Terrain Vehicle Risk Factors

FIGURE 1. Risk of injury by all-terrain vehicle engine size (for both
three- and four-wheel all-terrain vehicles), US Consumer Product
Safety Commission Nationwide All-Terrain Vehicle Injury and User
Surveys, 1997. ATV, all-terrain vehicle.

age 16 (14.3 percent in 1997 vs. 27 percent in 1985), a
decrease in the proportion of riders with 1 year or less of
experience (8.5 percent vs. 20 percent), a decrease in the
proportion of three-wheel ATVs in use (21.5 percent vs. 76
percent), and an increase in the proportion of drivers who
use ATVs in nonrecreational activities at least some of the
time (73.7 percent vs. 52 percent). Based on the findings of
the risk model, each of these population changes would
have been expected to reduce the rate of injury. Only
increases in engine size, which increased from an average
of under 200 cm3 of displacement in 1985 to about 250 cm3
in 1997, would have mitigated against the decreasing
injury rate.
Possible limitations

The results of this analysis need to be interpreted with
some caution. Because the surveys required information on
past behavior, responses may be subject to some recall bias.
Additionally, because children’s ATV use was reported by
parents or guardians, there exists the possibility of reporting
bias.
However, there is no reason to believe these potential
biases had any systematic impact on the statistical results.
Most of the variables, such as age, gender, driver experience, engine size, and number of wheels, were factual in
nature. Moreover, while reported nonrecreational use and
average monthly driving times were undoubtedly approximations, the statistical significance of the remaining relations was not altered when one or both of these variables
were excluded from the analysis.
Additionally, the risk findings are plausible. Young children may not have the strength, cognitive abilities, and
motor skills to operate large adult-sized ATVs safely (23).
With respect to gender, males may be more risk-taking than
Am J Epidemiol Vol. 153, No. 11, 2001

1117

females, as evidenced by automobile and other injury rates
(24, 25).
Higher risks for inexperienced drivers may be explained
by the unique dynamic properties of ATVs and the high
level of skill needed for safe ATV operation (26). The reasons for lower risks in nonrecreational applications are less
clear, but they may be related to differences in nonrecreational driving patterns or terrains.
The finding of higher risks on three-wheel ATVs is consistent with engineering findings that three-wheel ATVs are
less stable than are their four-wheel counterparts (26–28).
The positive relation between engine size and risk is probably related to the speed and acceleration characteristics of
ATVs with larger engines.
Although the findings are plausible, it should also be
noted that not all of the potentially important risk factors
were evaluated in this analysis. It would have been useful,
for example, to explore the relation between the use of alcohol and drugs and the ATV injury risk. Although some information about the use of these products was collected in both
surveys, this information could not be evaluated directly in
the risk analysis. The user survey asked all drivers about
their frequency of alcohol or drug use when riding; in contrast, the injury survey asked only adult riders about alcohol
or alcohol use prior to the injury.
The effect of helmet use on risk was not evaluated in this
study. In part, this was for theoretical reasons. A driver’s
decision to use a helmet is arguably interdependent with
risk. Thus, while helmet use is expected to affect risk, the
risk a rider faces may also affect the likelihood of helmet
use; for example, some drivers who ride in more risky
environments may be more likely to use helmets.
Assuming this to be the case, the decision to use a helmet
is not truly a “predetermined” variable eligible for use in a
single equation risk model (29). Additionally, helmets are
expected to prevent only head injuries. Because head
injuries account for only about 11 percent of injuries, most
of the outcomes studied would not have been addressed by
helmet use.
It should also be reiterated that inferences from the analysis are technically limited to drivers from ATV-owning
households. However, there is no reason to believe that the
general risk relations are different for nonowners.
Conclusions

The results of this analysis suggest that future safety
efforts should focus on reducing injuries and deaths involving children, getting beginning drivers to participate in the
existing training programs, and encouraging consumers to
dispose of the three-wheel ATVs still in use.

ACKNOWLEDGMENTS

The authors thank Thomas J. Schroeder for assistance in
adapting the data sets for analysis and Dr. Michael Greene
for helpful comments in the preparation of the article.

1118

Rodgers and Adler

REFERENCES
1. Kyle SB, Adler PW. Report on the 1997 ATV Injury Survey.
Washington, DC: US Consumer Product Safety Commission,
1998.
2. David J. Annual report of ATV deaths and injuries.
Washington, DC: US Consumer Product Safety Commission,
2000.
3. US Consumer Product Safety Commission. All-terrain vehicles: advance notice of proposed rulemaking; request for comments and data. Federal Register 1985 May 31;50:23139–44.
4. Report of the all-terrain vehicle task force to the CPSC.
Washington, DC: US Consumer Product Safety Commission,
1986.
5. US District Court for the District of Columbia, United States of
America vs. American Honda Motor Co, Inc, et al.
Memorandum and order. Civil action no. 87-3525, April 28,
1988.
6. Moore MJ, Magat WA. The injury risk consequences of the allterrain vehicle consent decrees. Int Rev Law Economics 1997;
17:379–93.
7. McDonald AK. The National Electronic Injury Surveillance
System: a tool for researchers. Washington, DC: US Consumer
Product Safety Commission, 1994.
8. Kessler K, Reiff L, Schroeder T. The NEISS sample (design
and implementation). Washington, DC: US Consumer Product
Safety Commission, 1997.
9. Marker D, Lo A. Update of the NEISS sampling frame and
sample. Rockville, MD: Westat, Inc, 1996. (Contract no.
CPSC-C-95-1165).
10. Stoner D, Srinath KP. All-Terrain Vehicle Exposure Survey.
Cambridge, MA: Abt Associates, Inc, 1998. (Contract no.
CPSC-C-94-1115, no. 4).
11. Kulp DW. Dynamics of “list-assisted” random dialing (RRD)
frame coverage. Fort Washington, PA: Marketing Systems
Group, 1995.
12. GENESYS sampling system methodology. Fort Washington,
PA: Marketing Systems Group, 1997.
13. Cochran WG. Sampling techniques. New York, NY: John
Wiley & Sons, Inc, 1977.
14. Kish L. Survey sampling. New York, NY: John Wiley & Sons,

Inc, 1965.
15. Shah BV, Barnwell BG, Bieler GS. SUDAAN user’s manual.
Release 7.5. Research Triangle Park, NC: Research Triangle
Institute, 1997.
16. Maddala GS. Limited dependent and qualitative variables in
econometrics. New York, NY: Cambridge University Press,
1983.
17. Hosmer DW, Lemeshow S. Applied logistic regression. New
York, NY: John Wiley & Sons, Inc, 1989.
18. Rodgers GB. The characteristics and use patterns of all-terrain
vehicle drivers in the United States. Accid Anal Prev 1999;31:
409–19.
19. Korn EL, Graubard BI. Analysis of health surveys. New York,
NY: John Wiley & Sons, Inc, 1999.
20. Kleinbaum DG, Kupper LL, Morgenstern H. Epidemiologic
research: principles and quantitative methods. New York, NY:
Van Nostrand Reinhold Company, 1982.
21. Chiang AC. Fundamental methods of mathematical economics. New York, NY: McGraw-Hill, 1984.
22. Rodgers GB. Results of the ATV Exposure Survey.
Washington, DC: US Consumer Product Safety Commission,
1986.
23. Benel DCR, Mayor AS. Developmental characteristics and use
of all-terrain vehicles. Alexandria, VA: Essex Corporation,
1986.
24. Evans L. Traffic safety and the drivers. New York, NY: Van
Nostrand Reinhold Company, 1991.
25. Accident facts, 1998 edition. Itasca, IL: National Safety
Council, 1998.
26. Deppa RW. Report on the engineering evaluation of all-terrain
vehicles. Washington, DC: US Consumer Product Safety
Commission, 1986.
27. Deppa RW, Hauser JA. ATV lateral stability and the limits of
control. Presented at the Society of Automotive Engineers
Government/Industry Meetings, Washington, DC, May 2–4,
1989. (SAE no. 891108).
28. Allen RW, Szostak HT, Rosenthal TJ, et al. Man machine systems analysis of ATV stability and control problems.
Hawthorne, CA: Systems Technology, Inc, 1988. (Technical
report no. 1253.1, contract CPSC-C-87-1286).
29. Pindyck RS, Rubinfeld DL. Econometric models and economic forecasts. New York, NY: McGraw-Hill, 1991.

Am J Epidemiol Vol. 153, No. 11, 2001