Article Text


The association between booster seat use and risk of death among motor vehicle occupants aged 4–8: a matched cohort study
  1. T M Rice1,
  2. C L Anderson2,
  3. A S Lee1
  1. 1
    Traffic Safety Center, Department of Environmental Health Sciences, University of California Berkeley, Berkeley, California, USA
  2. 2
    Center for Trauma and Injury Prevention Research, Department of Emergency Medicine, University of California Irvine, Orange, California, USA
  1. Correspondence to Dr T M Rice, Traffic Safety Center, Department of Environmental Health Sciences, University of California Berkeley, 2614 Dwight Way #7374, Berkeley, CA 94720-7374, USA; tomrice{at}


Objective: To estimate the effectiveness of booster seats and of seatbelts in reducing the risk of child death during traffic collisions and to examine possible effect modification by various collision and vehicle characteristics.

Methods: A matched cohort study was conducted using data from the Fatality Analysis Reporting System. Death risk ratios were estimated with conditional Poisson regression, bootstrapped coefficient standard errors, and multiply imputed missing values using chained equations.

Results: Estimated death risk ratios for booster seats used with seatbelts were 0.33 (95% CI 0.28 to 0.40) for children age 4–5 years and 0.45 (0.31 to 0.63) for children aged 6–8 years (Wald test of homogeneity p<0.005). The estimated risk ratios for seatbelt used alone were similar for the two age groups, 0.37 (0.32 to 0.43) and 0.39 (0.34 to 0.44) for ages 4–5 and 6–8, respectively (Wald p = 0.61). Estimated booster seat effectiveness was significantly greater for inbound seating positions (Wald p = 0.05) and during rollovers collisions (Wald p = 0.01). Significant variability in risk ratio estimates was not observed across levels of calendar year, vehicle model year, vehicle type, or land use.

Conclusions: Seatbelts, used with or without booster seats, are highly effective in preventing death among motor vehicle occupants aged 4–8 years. Booster seats do not appear to improve the performance of seatbelts with respect to preventing death (risk ratio 0.92, 95% CI 0.79 to 1.08, comparing seatbelts with boosters to seatbelts alone), but because several studies have found that booster seats reduce non-fatal injury severity, clinicians and injury prevention specialists should continue to recommend the use of boosters to parents of young children.

Statistics from

The use of booster seats for children who have outgrown traditional forward-facing safety seats has increased greatly over the last decade. In 1995, only 6% of children weighing 40–60 pounds were restrained in either a child safety seat or booster.1 This weight range corresponds to children aged 4 to 10 years.2 Glassbrenner and Ye3 estimated booster seat usage in 2007 to be 46% among children aged 4 or 5 years, and 36% among those aged 6 or 7 years using data from the National Survey of Use of Booster Seats (NSUBS). The Partners for Child Passenger Safety research programme reported that booster seat use in 15 states increased substantially from 1999 to 2002, from 10% to 22% for children of age 4, from 3% to 25% for age 5, from 1% to 25% for age 6, and from 1% to 5% for age 7 or 8.4

As of December 2008, 43 states and District of Columbia mandate the use of booster seats (increased from 33 in 2005), and legislation is pending in three states.5 In addition, several states have recently strengthened existing laws. The laws have varying age requirements, requiring booster seats for children younger than 6, 7, or 9 years of age. Twenty of the 43 jurisdictions have exemptions for children who exceed a specified weight, generally 60 or 80 pounds; 22 have exemptions for children who exceed a specified height, generally 4′ 9″. In addition, several states have recently strengthened existing laws.

The need for boosters seats was determined largely by studies of seatbelt-related injury, ie, “seatbelt syndrome”, and by ergonomic examinations of seatbelt fit among young children. Seatbelt syndrome is characterised by abdominal and spinal injuries in restrained motor vehicle occupants6 7 8 9 10 as a result of pressure on the abdomen and hyperflexion of the spine.11

The objectives of this study were to estimate the effectiveness of booster seats used with seatbelts in reducing the risk of child death during traffic collisions, to examine possible effect modification by various collision and vehicle characteristics, and to compare the effects of booster seats with seatbelts to seatbelts alone.


We conducted a matched cohort study using data from the Fatality Analysis Reporting System (FARS) for years 1996–2006. FARS is a national surveillance system for fatal motor vehicle collisions operated by the National Highway Traffic Safety Administration.

Matched cohort studies, in which investigators match exposed persons to unexposed persons prior to outcome determination, have been used by epidemiologists to improve study efficiency. They are typically used when the factors of interest are strongly associated with potentially confounding covariates.12 Matched cohort designs have been used in traffic safety analyses using FARS data because all occupants within collision-involved vehicles share values for most vehicle- and collision-level characteristics. The occupants have in effect been matched, not by an investigator but by circumstance. Because only matched sets with one or more fatalities are required by the matched analysis methods, information from surveillance systems like FARS can be used.

Matched cohort methods have been used to study the effects of seatbelts,13 14 child restraints,15 air bags,16 17 18 pickup truck cargo area travel,19 occupant seating position,20 21 22 and motorcycle helmets.23 24 25

An initial sample of passenger cars, minivans, pick-ups, and SUVs with model years 1990–2006 was assembled. The following selection criteria were applied to the vehicles: 2 or more occupants were travelling in the first two rows of seating, 1 or more occupants were aged 4–8 years, and 1 or more occupants died, resulting in a sample of 6006 vehicles. Data from 499 vehicles from Alabama, Indiana, Iowa, Maryland (for 1996), Massachusetts (for 1996–2001), Nebraska, Rhode Island (for 1996–1999), and Virginia were excluded. An earlier study found that these states had incomplete occupant inclusion or incomplete occupant data collection.15 In addition, four vehicles were excluded because a child age 8 or younger was recorded as having been in the driver seat of the vehicle, resulting in data records for 19 185 persons travelling in 5503 vehicles.

We dichotomised the injury severity variable into fatal and all non-fatal outcomes. The codes for belt use (shoulder, lap, lap/shoulder, improper) were collapsed. Proper and improper safety seat use were also collapsed because the determination of proper use is likely to be influenced by the reporting police officer’s knowledge of the occupant injury severity.26 We assumed that children aged 4–8 years coded as using a child safety seat were using a booster seat together with a seatbelt. Some children, particularly 4-year-olds, may have been using forward-facing safety seats.3 4

We used conditional Poisson regression to estimate death risk ratios (RRs) and obtained standard errors with a bootstrap method because conditional maximum likelihood variance estimates are incorrect.27 The model contained terms for child seatbelt use with booster, child seatbelt use without booster, adult seatbelt use, exposure to air bag deployment, seating position, sex, age, and quadratic age spline terms (nodes at 10, 20, and 40 years), and interaction terms between age category and each of the child restraint codes. Effect modification was estimated with additional interaction terms and assessed with Wald tests of homogeneity.12 All programming and statistical procedures were performed with the Stata MP 10.1 software package.28

Of the 19 185 occupants, 2210 (12%) had one or more missing values: 1486 (8%) were missing information on restraint use, 450 (2%) on air bag deployment, 184 (1%) on seating position, and 10 (<1%) on sex; 80 (<1%) occupants had missing values on two or more of these variables. We imputed values29 for these four variables using multiple imputation with chained equations (MICE).30 Each equation used to predict values contained some combination of the above variables and also calendar year, model year, land use, vehicle type, rollover status, direction of impact, fatality, and age. When only the seating row was known, the specific position was imputed.


The 5503 identified vehicles carried 6851 children aged 4–8 years. Of these, 2193 (32%) suffered fatal injury (table 1). More than three-quarters (5221, 76%) were seated in the rear seat at the time of collision, 380 (6%) were known to have experienced an air bag deployment, and 986 (14%) were partially or fully ejected.

Table 1

Occupant characteristics, children aged 4–8 years

The use of seatbelts and booster seats varies greatly between ages 4 and 8 years (table 2). The proportion of children using a booster seat declined with age; 26% of children aged 4 to 13%, 6%, 2%, and 1% for those aged 5, 6, 7, and 8, respectively. Seatbelt use (without a booster seat), on the other hand, increased steadily from 34% of children aged 4 to 56% of those aged 7. Seatbelt use was more prevalent than booster/safety seat use at all ages.

Table 2

Restraint use category by age, children aged 4–8 years (Fatality Analysis Reporting System restraint codes)

Estimated risk ratios for booster seat use were 0.33 for children aged 4–5 years and 0.45 for children aged 6–8 years (table 3; p<0.005). The estimated risk ratios for seatbelt use were similar for the two age groups, 0.37 and 0.39 for ages 4–5 and 6–8, respectively (p = 0.61).

Table 3

Restraint use death risk ratios (RRs)*

Booster seat risk ratio estimates varied significantly by seating position (Wald p = 0.05) and rollover status (p = 0.01) (table 4). The risk ratio for front-seated children was similar to that for rear-seated children in the middle (RR 0.22 and 0.24, respectively). Risk ratio for rear-seated children in the outboard positions was notably higher, 0.43 and 0.35 for left and right positions, respectively. The estimated risk ratio was significantly less during rollover collisions than during non-rollover collisions (p = 0.01). Significant variability in the risk ratio estimates was not observed across levels of calendar year (p = 0.64), vehicle model year (p = 0.33), vehicle type (p = 0.26), or land use (p = 0.49).

Table 4

Potential modifiers of booster seat effectiveness


This analysis indicates that seatbelts, used with booster seats, are highly effective in preventing death among young motor vehicle occupants. Among 4- to 8-year-old children, those who were travelling unrestrained were 2.8 times more likely to die than those restrained in seatbelts with boosters, given involvement in a severe traffic collision. Effectiveness was slightly higher for children aged 4–5 than those aged 6–8 years. We also estimated the effectiveness of seatbelts (without booster seats) in reducing death risk. The estimated effects of seatbelt use alone were very similar to those of seatbelts with boosters. Unrestrained children were 2.6 times more likely to suffer fatal injury than belted children. The estimated death risk ratio comparing seatbelts with boosters with seatbelts alone was 0.92 (0.79–1.08).

This finding of comparable effectiveness is consistent with at least one other study. Miller et al31 examined the need for booster seats by comparing injury outcomes among belted children aged 4–7 years to those of belted children aged 8–13 years. They hypothesised that if seatbelts failed to reduce injuries due to poor fit among young children, the average injury severity among children aged 4–7 years would be observably higher than among older children, for whom seatbelts fit better. Overall, they found no evidence in support of their hypothesis. The younger children did not have a significantly higher odds of death (OR 0.87, p = 0.56) than the older children. Their odds of severe or fatal injury appears to be lower than that of the older children (OR 0.52, p = 0.01). The analysis was methodologically rigorous but was limited by sparse data.

Other studies have found booster seats to be better than seatbelts alone in protecting children from non-fatal injury during collisions. Durbin et al32 used insurance claim data supplemented with parent interviews to compare restraint effectiveness among children aged 4–7 years. They estimated an injury odds ratio of 0.41 for seatbelts with boosters compared with seatbelts alone, after accounting for seating position, crash severity, vehicle type, and child age. They found that effectiveness did not appear to vary by seating position or child age. The reliance on parent-reported restraint use may have introduced some bias in the study findings. Parents of unrestrained children may misreport restraint use for a variety of reasons, including feelings of guilt or denial, fears of societal disapproval, or legal liability for not having restrained their children. The authors found that only extreme levels of non-differential misreporting by parents would have meaningfully influenced their results, but they did not assess possible bias from differential misreporting.

Edgerton et al33 compared the performance of shield booster seats with traditional safety seats among children 20–40 lb in weight, whom NHTSA recommends be restrained in a forward-facing, 5-point harness safety seat. Children of this weight are generally between 2 and 5 years of age.2 Children using shield boosters were found to have higher Abbreviated Injury Scores (AIS), to have an Injury Severity Score (ISS) >15, and were more likely to be admitted to intensive care. The study was limited by sparse data (30 children in safety seats and 16 children in shield boosters) and the use of crude Mantel–Haenzsel odds ratios.

Only one study has examined the association between booster seat use and fatal outcomes. Elliot et al34 analysed data on children aged 2–6 years in FARS and the Crashworthiness Data System. After adjusting for vehicle model year and type, seating position, and driver characteristics, they reported an overall child restraint death risk ratio, comparing children in child restraints to those using belts, of 0.79 (0.59–1.05) and a risk ratio of 0.72 (0.54–0.97) after excluding children coded as having been improperly restrained. Several factors limit the usefulness of these findings in assessing booster seat effectiveness. First, the restriction to children coded as properly restrained may have led to the overestimation of restraint effectiveness. Second, the study included children aged 2 and 3 years who, when using a child restraint system, are likely to be in forward-facing safety seats and not in booster seats. The authors did not estimate effectiveness separately for the older children likely to be using boosters. Given that 75% of the children using child restraints were aged 2 or 3 years, it is likely that the study’s estimates reflected the effectiveness of child safety seats. Third, crash severity cannot be fully controlled for in CDS data and thus residual confounding may have been present.

One reason that booster seats, which improve seatbelt fit, may not improve collision survivability is that the effect of improving seatbelt fit is to lower the probability of injuries to the abdomen and lumbar spine, which can be severe but are much less often fatal than injuries to the head and thorax. Starting in 1956,6 numerous studies have shown the association of paediatric seatbelt use with often-severe injuries, including abdominal contusions, pelvic fractures, lumbar spine dislocations and fractures, and intra-abdominal organ injuries.6 9 35 36 Properly fitting seatbelts may be no better than poorly fitting seatbelts in terms of preventing severe or fatal injuries suffered during collisions with major cabin intrusion or during rollover collisions. It may be that occupant ejection, for example during rollover collisions, is effectively prevented by seatbelts regardless of fit.

The methods used in the current analysis have several advantages over other methods. First, the matched nature of the data, together with the conditional regression model, provides excellent control of potential confounding by measured and unmeasured factors, including but not limited to collision severity, collision type, vehicle type, and the availability and quality of emergency medical services. The matched cohort study design also allowed us to directly estimate death risk ratios instead of approximating them with odds ratios.12 27 37 In addition, the data used in this analysis came exclusively from fatal collision investigations and are likely to be of higher quality than data reported for non-fatal injury collisions.

In this analysis we used matched sets of occupants that were, in some instances, comprised of only one child and one or more adults. Elliott et al asserted that the practice of using these mixed sets of occupant data may produce bias in the restraint estimates.38 To examine this possible bias, we fitted our model to matched sets of two or more children (a subset of our study data). Using one of the imputed datasets, we estimated risk ratios for booster seats with seatbelts and for seatbelts used alone for all children aged 4–8 years as 0.35 and 0.38, respectively (based on 5503 sets). We then restricted the dataset to the 1377 child-only matched sets and estimated risk ratios of 0.42 and 0.42, respectively. While much precision is lost by using the restricted data, the magnitude of change in point estimates does not support the concern of bias.

Multiple imputation allowed us to utilise the available information for occupants who were missing information on other factors. The risk ratios estimates were not meaningfully different from those obtained in a complete-data analysis. The risk ratio for seatbelts with boosters for children aged 4–8 years was 0.35 (0.30–0.41) using the imputed data (5503 matched sets) and 0.35 (0.28–0.43) using only data for occupants with no missing values (3937 matched sets).

Our study has several limitations. First, the possibility of misclassification of restraint coding among children in this age group exists. Because booster seats function with an existing seatbelt system, police officers may be inconsistent in their coding of booster seat use. Officers might code booster seat use as seatbelt use, which is not entirely incorrect. So, while child occupants in FARS with a safety seat code are likely to have been using a safety seat or booster seat, those with a seatbelt code are likely to be a mixture of children using seatbelts alone and seatbelts with boosters. This phenomenon may have caused our seatbelt and booster seat estimates to be closer together than is true. Research is needed to better understand how police officers code for the various types of child restraint.

Second, FARS does not differentiate between child safety seats and booster seats. Some of the children in our study, primarily 4-year-olds, may have been restrained in safety seats rather than booster seats.

Third, because FARS does not differentiate between booster seat types, we were prevented from comparing the performance of the different types.

Fourth, we collapsed proper and improper codes for both seatbelt use and booster seat use. Police officers’ use of improper use codes may be influenced by their knowledge of injury outcomes. This differential coding of properness would likely cause estimates of improper use to be biased towards or past the null (appearing less effective than is true) and estimates of proper use to be biased away from the null (appearing more effective than is true). The collapsed restraint codes may provide an “as used” estimate of restraint effectiveness.

This analysis provides some evidence that booster seats do not improve the performance of seatbelts with respect to preventing death. But because several studies have found that booster seats reduce non-fatal injury severity among collision-involved child occupants, clinicians and injury prevention specialists should continue to recommend the use of boosters to parents of young children. The use of lap belts without shoulder belts should be strongly discouraged because of its association with often-severe injuries to the abdomen and lumbar spine.36

Research is needed to clarify the possible influence of restraint misclassification by parents and by reporting police officers on studies of restraint effectiveness.

What is already known on the subject

  • Booster seat use reduces injury severity among young children who have outgrown forward-facing safety seats.

  • Booster seat usage has steadily increased over the past decade.

What this study adds

  • Seatbelts, with or without booster seats, greatly reduce the risk of death among vehicle occupants aged 4–8 years during severe traffic collisions.

  • Seatbelts used with booster seats appear to be no better at preventing death than seatbelts alone.


TR designed the study, obtained funding, analysed the data, and drafted the manuscript. CA contributed to study design, analysed the data, and revised the manuscript. AL reviewed literature and revised the manuscript.


View Abstract


  • Funding Funding for this research was provided by the AAA Foundation for Traffic Safety (Grant # AAAFTS-51096) and the California Office of Traffic Safety (Grant # AL0801).

  • Competing interests None.

  • Human participant protection This study was certified as exempt from institutional review by the University of California Berkeley Office for the Protection of Human Subjects.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.