The Influence of Communication About Safety Measures on Risk-Taking Behavior
Risk-taking behavior is an important contributing human factor to incidents and is notoriously difficult to influence. Anecdotal evidence suggests that people have a hard-wired optimal perceived risk level. People compensate for risk-reducing measures by behaving in a riskier fashion until the desired level of risk is reached again. This study looked at the effect of the number of shields of protection and uncertainty on the risk-taking behavior of the participants.
The main aim of safety research is to identify ways to prevent accidents and to ensure the safety of workers. Human error—or, in other words, unsafe behavior—has been found to be a major cause of accidents, and its elimination, therefore, is a prime goal for improving safety. The human factor is most effectively addressed by tackling the organizational system instead of focusing on incorrect actions by individuals. An effective strategy is to increase the level of protection or the number of safety barriers. The concept of the safety barrier features most prominently in the Swiss-cheese metaphor of accident causation. The Swiss-cheese model describes accidents as being caused by unchecked hazards that are allowed to cause losses. A series of barriers is placed between the hazard and that which may be harmed. The barriers keep the hazard under control and prevent it from causing harm. However, these barriers are always less than 100% adequate and contain weaknesses or holes. The barriers, therefore, often are compared to slices of Swiss cheese. Unlike real Swiss cheese, the holes in the barriers are dynamic and open and close at random. When these holes in the barriers are aligned, a path is created, leading to a potential accident.
Intuitively, one would assume that safety improves proportionally both to the protection measures taken and to the improvements in the design of such measures. The more protective equipment given to the workers, the safer they will be, either because of a reduced risk of accident or because such measures mitigate the effects of accidents. This approach assumes that human error can arise from unintended actions such as memory lapses and attention failures. However, it can also be attributed partly to intended actions such as risk taking. The question addressed in this paper is related to the extent to which people’s risk-taking behavior was influenced by their awareness of the numerous preventive interventions in place. The pivotal issue that arises is whether people adapted their risk-taking behavior as a result of their awareness of the number, and of the effectiveness, of the barriers in place.
Alteration of behavior is a recurring theme in the safety literature and is described in a variety of ways—“risk compensation,” “risk homeostasis,” or the Peltzman effect. When people feel safer, they tend to take greater risks. People do not want to reduce risk to an absolute minimum but, rather, to optimize it. People are willing to accept a certain level of risk if risky behavior (e.g., breaking a barrier in the Swiss-cheese model) comes with benefits.
There is virtually no behavior without a certain measure of risk attached to it. Therefore, the challenge is to optimize rather than to eliminate risk. This optimum, also known as the target level of risk, is the level that maximizes the overall benefit. Previous studies suggest that people constantly compare the amount of risk they perceive with their target level of risk and that they will adjust their behavior in order to eliminate any discrepancies between the two. This psychological mechanism constitutes a case of circular causality.
The mechanism is similar to a thermostat, where there are fluctuations in the room temperature but where such fluctuations are averaged over time; the temperature will remain stable unless set to a new target level. The risk homeostasis theory (RHT) transfers the homeostatic effect of a thermostat to risk behavior. RHT posits that, similar to a thermostat that has a target temperature, people have a target level of risk. People will change their behavior in order to maintain their target level of risk.
Previous research has given some indications that people compensate for safety measures such as barriers or shields by behaving in a riskier fashion. However, besides such anecdotal evidence, no systematic research has been carried out to consider the effect on behavior of informing people of the number of safety barriers in place for their protection. This paper sought to answer two questions:
- Do people indeed compensate for greater layers of, or more effective, protection by behaving in a riskier fashion?
- How do people behave when they are uncertain about the number of shields of protection?
A side-scrolling videogame was custom made for this experiment. It required the player to navigate a small spaceship through an asteroid field, with asteroids moving from right to left after materializing randomly on the y-axis. The spaceship was controlled with the arrow keys on the keyboard. The up and down keys moved the spaceship up and down, and the right and left keys increased and decreased the speed, respectively. Five different speed levels ranged from 1 (default) to 5 (maximum). The number of shields left was indicated as can be seen in Fig. 1. Whenever the ship crashed against an asteroid, the player lost a shield. This process continued until there were no shields left. A collision at that point would end the game. In circumstances where participants did not know the number of shields the ship had, a question mark was displayed in front of their spaceship. The game measured the time a player spent on each shield and the total amount of points a participant scored. Points were gained by staying alive. The faster a participant flew, the more points he or she gained per second. Such bonus points acted as an incentive for participants to increase speed in order to achieve higher scores.
To test if the layers of protection and the uncertainty about their number had an impact on the risk-taking behavior of participants, an experimental design was chosen. This design allowed for the determination of causal relationships. This experiment focused on the relation between the number of shields (maximum five) and the level of risk taking. There were 104 students participating in the experiment. Participants were randomized to participate in two out of six possible sets of conditions:
- Condition 1—Zero shields
- Condition 2—One shield
- Condition 3—Two shields
- Condition 4—Three shields
- Condition 5—Four shields
- Condition 6—Unknown number of shields (actually four shields)
The data were analyzed, and univariate analyses of variances (ANOVAs) were used to analyze the different trends to evaluate four hypotheses.
Hypothesis 1. The average mean of speed is different for varying conditions: The higher the number of barriers to which participants are exposed, the greater the degree of risk they take in flying.
A univariate ANOVA was calculated to see if the average speed was different for the varying conditions. Condition 6 was excluded. A significant difference was found, with a significant upward linear trend. Hypothesis 1 was confirmed.
Hypothesis 2. The average mean of time spent per shield is different for varying conditions: The fewer shields participants are exposed to, the more time they spend per shield.
A univariate ANOVA was calculated to see if the average speed was different for the varying conditions. Condition 6 was excluded. A significant difference was found, with a significant downward linear trend representing the data best. In Condition 5, players spent less time per shield than in Condition 1. Hypothesis 2 was confirmed.
Hypothesis 3. In conditions of uncertainty, participants fly more slowly than in conditions where they know how many barriers they are exposed to.
A repeated-measures ANOVA with a Greenhouse-Geisser correction determined that the average speed of play throughout Condition 6 differed significantly statistically between shields. A significant upward linear trend represented the data best. Hypothesis 3 was confirmed.
Hypothesis 4. In conditions of uncertainty, participants spend more time per shield than in equivalent conditions where they know how many barriers they are exposed to.
A repeated-measures ANOVA with a Greenhouse-Geisser correction determined that the average time played per shield throughout Condition 6 did not differ significantly statistically between shields. Hypothesis 4 was not confirmed.
When participants entered the game with five shields, they played in a significantly riskier fashion than when they entered with only a single shield. This is a highly relevant finding, considering that costly risk-assessment techniques may be of little value if improvements in safety systems are outweighed by the risks introduced by changes in operator behavior. However, removing safety features might not be a very ethical move. One suggestion could be to hide protection mechanisms from the system operators until needed. This goes toward creating a feeling of uncertainty or ambiguity among workers concerning their safety, which could be accompanied by a communication strategy emphasizing this uncertainty. Putting a greater number of layers of protection in place was not rendered ineffective completely by increased risk taking. Although the limitations of this study should be recognized, organizations might reconsider the practice of giving information to their employees on the number of safety measures that are taken. Preserving ignorance among employees concerning the enhanced protection in place creates a stronger safety buffer because it reduces risk-taking behavior and improves employees’ efforts to make sure that the presumed last layer holds.