Home page

Main topics
News Headlines

Frequently asked Questions
For Policy Makers

Research evidence
Misleading claims
Helmet laws

Search Engine

New Zealand
Other countries

Full index

Policy statement

Risk-Compensation Behavior in Children - Myth or Reality?

Pless IB, Magdalinos H, Hagel B.
Arch Pediatr Adolesc Med, 2006;160(6):610-614.

Original paper


Author's description of the study

In Canada, parents of children attending a children's hospital emergency department are asked to provide details of the event, which is recorded on a form used by the Canadian Hospitals Injury Reporting and Prevention Program (CHIRPP). CHIRPP records were used to identify 674 children aged 8-18 who were injured in a activity in which protective equipment (PE) can be used and who attended the Montreal Children's hospital from 1/12/01 to 30/11/02. 394 children were interviewed. Although, according to CHIRPP records 325 children wore PE, only 234 claimed to have done so in the interviews (55% used a helmet, 23% used knee pads, <1% used wrist guards, and <1% used other PE). There was no evidence of an association between indicators of risk-taking behavior and PE use after adjusting for age, sex, personality, and type of activity and no relationship between injury severity and PE use.

BHRF Commentary

Experimental evidence contradicts the conclusion from this study

The best way of investigating risk compensation is a controlled trial to evaluate the same people with and without protective equipment (PE). For example, children running an obstacle course in a gym were more reckless and finished the course quicker when wearing a helmet and wrist guards than without (Morrongiello, Walpole and Lasenby, 2007). The differences were quite substantial. 10-12 year old girls tripped, fell or bumped into things on average 2.85 times without PE, compared to 4.55 times with PE, an increase of 60%. Increases in risk taking averaged 49% for 7-9 year old girls and 48% for both 7-9 and 10-12 year old boys. Sensation-seeking children showed greater risk compensation. (Morrongiello, Walpole and Lasenby, 2007)

Another example of the unreliability of comparing self-selected groups

The inability of this study to detect risk compensation, despite evidence from other studies of a 50% increase, is another example of the unreliability of comparing self-selected groups. Such groups (e.g. women who chose to take hormone replacement therapy vs those who chose otherwise) often differ because of circumstances relating to the choice (e.g. high-income groups are more likely to use expensive treatments) as well as the effects of the chosen treatment or equipment. Although studies try to adjust for the former, research shows that great care is needed to avoid inaccurate, biased, or misleading results (Lawlor, Smith and Ebrahim, 2004).

The contradiction between the conclusions of this study and the experimental data showing clear evidence of risk compensation demonstrates that direct measurement of effects can provide more reliable results. The main author of this paper has argued that enforced helmet laws are beneficial because comparisons of self-selected groups suggest that helmets are effective (Pless, 2006).Yet even if the effect of helmets could be estimated reliably in this way, helmet laws have other important consequences such as reduced cycling, reduced safety in numbers and increased risk taking by cyclists forced to wear helmets. The only realistic way to determine the effect of all these factors is by observing the effect of legislation, as discussed in a review of helmet laws (Robinson, 2006).

It is to be hoped that the warning message concerning unreliability of comparisons of self-selected groups will now be heeded.

Power of study acknowledged to be weak

The authors state that the sample they have used is large enough to detect a 2-fold or greater difference in risk-compensatory behavior between the groups. However, much smaller increases in risk, such as the 50% increase reported from the direct observation trial, is likely to result in PE-users having many more injuries than non-users. Indeed, if bicycle helmet laws result in a modest 10% increase in risk (due to risk compensation or reduced 'safety in numbers'), the total number of injuries would increase (Robinson, 2007). Because of its inadequate design, the study would not be able to detect this.

The study was also designed only to detect conscious changes in risk taking, as it relied upon self-reporting by interviewees as to whether they were taking more chances than usual. However, much risk taking is sub-conscious, like correcting the movement of a bicycle helmet on the head while cycling. This can increase risk by interfering with bike control and distracting attention from traffic, but it isn't a conscious decision to ride more riskily on the part of the cyclist. If asked, the cyclist would probably not say that they were (consciously) taking more chances.

Inconsistencies, contradictions and evidence of major problems with confounding

This paper is confusing and contradictory. For example, Table 2 presents odds ratios (OR) for protective equipment (PE) use, whereas Table 3, with no comment or explanation, apparently presents OR for non-use. Casual readers would assume that odds ratios in Table 3 are the same as Table 2, and be misled or confused.

Also confusing is the lack of any statement or explanation about how the OR were calculated, even though this information is necessary for their interpretation. Some are easy to understand. For example, in Table 2, the reported OR of 2.43 (identical to the crude odds ratio) clearly shows that boys were more likely to use PE than girls. Reported OR for age (also identical to the crude odds ratios) show that children under 11 years are twice as likely to wear PE as older children.

But for activity, the crude odds ratio for bicycling (relative to hockey or skating) is 0.58, yet the reported OR is 4.80. In other words, fewer cyclists actually wore PE, but after adjusting for other effects, we are supposed to believe cyclists were almost 5 times more likely to wear PE. This 8-fold change in OR suggests a major problem with confounding that could lead to difficulties interpreting the data, especially if readers are left to guess what terms were fitted in the model(s) used to calculate OR.

Non PE-users 94% less likely to been injured previously – evidence of risk compensation?

Another contradiction is the authors' statement that "PE users were more likely to report having used a helmet previously and to have been injured previously in the same activity, although in neither case did the odds ratio exclude the null." Table 3 reports an OR of 0.06 (CI 0.02 to 0.18), implying that children not using PE were 94% less likely to have been injured in the same activity. The confidence interval in Table 3 clearly excludes the null, contradicting the text. As with the statement in the text that 234 children used PE, but reporting 235 in the tables, it suggests some lack of attention to detail and thoroughness in the research.

Older children may have participated in the same activity for a longer period of time, so perhaps they might be more likely to have been injured before. However, the reported odds ratio was presumably adjusted for age, sex, type of activity and perhaps other factors. So if Table 3 is correct (as opposed to the authors' comments), this phenomenon needs further investigation. On the basis of the information presented, risk compensation cannot be excluded as a possible explanation of why PE users were much more likely to have had a previous injury.

Protective Equipment didn't work!

Table 5 of the paper reports that 49 PE users and 269 non-users had injuries to an unprotected body part. But if only 49 of the 235 PE users had injuries to an unprotected body part, presumably the remaining 186 PE users had injuries to a protected body part. If this is correct, it suggests the protective equipment didn't work.

Study doesn’t actually measure risk compensation

The best way to measure risk compensation is to compare the behaviour of the same people with and without PE. If this is not possible, subjects should be randomly assigned to PE or non-PE categories. For example, a randomised study found that drivers shown the benefits of antilock-braking systems (ABS) drove 5-10% faster than control drivers (Grant and Smiley, 1993).

Another study showed that overtaking drivers are twice as likely to drive uncomfortably close to cyclists who wear helmets (Walker, 2007). Because of the design, the samples of motorists passing the helmeted and non-helmeted cyclist should be representative of the same population.

People who chose to use PE tend to be more cautious than those who do not. So asking children if they were taking chances, going fast or taking part in a dangerous activity might provide some information on their risk tolerance, but not the absolute level of risk. A cautious child might believe that it is dangerous to cycle downhill at 20 km/hr, but an experienced rider who normally travels at this speed might think otherwise. Similarly, asking children who normally wear PE whether they were going faster or taking more chances than usual tells us nothing about how they would behave in unusual circumstances such as if they did not have their PE. In other words, this study tells us nothing about risk compensation.

In contrast, when Canadian children were asked hypothetical questions, the majority reported changes toward riskier behaviour when using protective equipment (Mok et al, 2004). Children clearly understand the concept. Indeed few parents would doubt that an observational experiment at a skateboard park would find that children wearing helmets, gloves, elbow and knee pads would indulge in more risky manoeuvres than the same children with bare knees and elbows.

Conclusions not valid

Rather than the authors conclusion that "efforts must continue to persuade children to protect themselves by whatever means possible", individual efforts should be evaluated on merit. Risk compensation is only one of a number of possible side effects. For example, enforced bicycle helmet laws lead to reduced cycling (a healthy environmentally friendly activity) and potentially increased risk of injury per km cycled because of reduced safety in numbers (Robinson, 2005b).

To avoid these problems, all interventions should be evaluated. If clear benefits are evident in response to the intervention, and the benefits outweigh the costs, the measure is worthwhile. In contrast, if there are no clear benefits, attention should be focused on identifying and implementing more effective measures.


Grant and Smiley, 1993

Grant BA, Smiley A, 1993. Driver response to antilock brakes: a demonstration of behavioural adaptation. Canadian Multidisciplinary Road Safety Conference VIII; Saskatoon, Saskatchewan June 14-16, p211-220; 1993.

Lawlor, Smith and Ebrahim, 2004

Lawlor DA, Smith GD, Ebrahim S, 2004. The hormone replacement - coronary heart disease conundrum: is this the death of observational epidemiology?. Int Journal of epidemiology 2004;33:464-467.

Mok et al, 2004

Mok D, Gore G, Hagel B, Mok E, Magdalinos H, Pless IB, 2004. Risk compensation in children's activities: A pilot study. Paediatr Child Health 2004;9(5):327-330.

Morrongiello, Walpole and Lasenby, 2007

Morrongiello BA, Walpole B, Lasenby J, 2007. Understanding children's injury-risk behavior: Wearing safety gear can lead to increased risk taking. Accident Analysis & Prevention 2007 May;39(3):618-23.

Pless, 2006

Pless B, 2006. Are editors free from bias? The special case of Letters to the Editor. Injury Prevention 2006;12:353-355.

Robinson, 2005b

Robinson DL, 2005. Safety in numbers in Australia: more walkers and bicyclists, safer walking and bicycling. Health Promotion Journal of Australia 2005;16:47-51.

Robinson, 2006

Robinson DL, 2006. Do enforced bicycle helmet laws improve public health?. BMJ 2006;332:722-725.

Robinson, 2007

Robinson DL, 2007. Bicycle helmet legislation: Can we reach a consensus?. Accident Analysis & Prevention 2007;39(1):86-93.

Walker, 2007

Walker I, 2007. Drivers overtaking bicyclists: Objective data on the effects of riding position, helmet use, vehicle type and apparent gender. Accident Analysis & Prevention 2007 Mar;39(2):417-25.

See also