PII: S0003——6870(97)00047-1
Applied Ergonomics Vol. 29, No. 4, pp. 247 — 254, 1998 ( 1998 Elsevier Science Ltd All rights reserved. Printed in Great Britain 0003—6870/98 $19.00#0.00
The effectiveness of warning signs in hazardous work places: cognitive and social determinants Austin Adams,* Stephen Bochner and Lenka Bilik School of Psychology, University of New South Wales, Sydney NSW 2052, Australia (Received 16 September 1996; in revised form 20 May 1997)
Recommendations have been made that good warning signs should have a number of components: an alerting word such as danger, then statements of the hazard, of its seriousness, of the probable consequences and of how to avoid the hazard. Responses from 40 blue-collar workers and 44 students were measured to five industrial warning signs to determine the extent to which these components determine estimated sign effectiveness and behavioural intentions of compliance. Each sign was presented in five versions; the original (which in each case omitted one or more of the components), a ‘full’ in version in which missing components were generated and inserted, and versions omitting, in turn, the hazard, consequences, and instructions statements. Previous findings concerning the importance of the signal word were supported. When signs were seen singly there was no tendency for versions with components missing to be rated as poorer. Only when all versions were seen together was the ‘full’ version ranked as being better, a result which is taken as reflecting the demand characteristics of the method. The results argue against strict adherence to a formula requiring specified components in a sign. Data also supported the third person effect, indicating that respondents considered others to be more vulnerable to the hazard and less likely to comply with the sign than they themselves. The finding that social factors are of considerable importance in sign compliance is discussed. ( 1998 Elsevier Science Ltd. All rights reserved. Keywords: warnings design, sign design, safety climate
Introduction
the degree of its severity, such as whether it is a danger of injury, of poisoning or of death. This requirement appears to be self-evident, yet there is little direct evidence concerning its importance. The third requirement is that there should be a clear statement of the adverse consequences if the warning is not obeyed. Leonard et al (1986) found that including a consequences statement in a warning increased risk perception and the likelihood of compliance. It has also been found that the more explicit the consequence information provided, the greater the perceived hazardousness of the situation and the greater the subsequent recall of the warning information (Laughery and Stanush, 1989; Sherer and Rogers, 1984). Finally, it is assumed that good signs should include specific instructions for action. That is, they should state what should or should not be done to avoid the adverse consequences. Thus, the empirical literature has identified several cognitive components that affect the ability of a sign to fulfil its function. One implication of this research is that the more of these features are present in a sign the more likely it is that the warning will be obeyed, and there is some direct evidence for this contention. Wogalter et al (1985) found that signs which contained all of the four characteristics (signal word, hazard
There is experimental evidence to support the contention that, for warnings to be perceived as effective, they should contain a number of fundamental message components (ANSI Z535.4, 1991; Peters, 1984; Wogalter et al, 1985). They should attract attention, they should then alert persons to the specific hazard and its degree of seriousness, to the probable consequence of the hazard, and to the way in which the hazard can be avoided. The first requirement, that of gaining attention, is usually fulfilled through the use of a signal word such as danger or caution. For instance, Wogalter et al (1992) found that the presence of such a signal word increased the perceived hazard of the warnings they studied. Silver and Wogalter (1989) looked at the ability of different signal words to convey differing levels of hazard, finding that words such as danger, poison, and death were the most potent, but they failed to show any difference between warning and caution. The second requirement is that the warning should contain information about the nature of the hazard and
*Author to whom correspondence should be addressed 247
248
Designing warning signs: A. Adams et al.
statement, consequences and instructions) were rated as being more effective than signs where any one of these four elements was absent, although perceived effectiveness was little affected when some redundancies among these statements were removed. They suggest that in some real-world cases such removal may make a sign briefer and more effective. Despite the quite large body of research which has produced reasonable consensus in identifying the design principles for good warning signs, common experience, as well as some of the literature, indicates that even well-designed signs are not totally effective. In three different studies, DeJoy (1989) found rates of compliance with a warning ranging from 17 to 37%, but in each of these studies a much higher percentage of the respondents stated that they noticed (88—91%) or read (46—77%) the warnings. As Lehto and Miller (1986) and Edworthy and Adams (1996) have pointed out, the ultimate criterion of a good warning sign is its effectiveness, defined as the extent to which the warning leads to compliance compared to the amount of compliance that would have occurred if the sign had been absent. That even an apparently well-designed sign does not raise the compliance to 100% points to the involvement of additional factors, such as individual and social determinants. Some of these additional factors have been studied. The cost of compliance (an economic factor) and the involvement of others (a social factor) are two of these. The cost of compliance refers to all the negative economic factors associated with compliance, such as the financial cost of protective clothing, the time cost of carrying out what might be seen as non-essential but safety-related activities and even the lost opportunity cost of not doing something pleasurable but possibly unsafe. The involvement of others refers to the social factor of the effect on compliance of seeing others comply. Wogalter et al (1988) found that both had significant effects on compliance. Compliance increased as the cost of compliance decreased and as the involvement of others increased. Both of these factors, however, are complex and their effect on warning compliance deserves further study. The social factor mentioned can be seen as an example of behavioural change in response to a communication. Theoretical approaches identify and distinguish between the three components that feature in most persuasive communications, namely the message, the source of the message and the target of the message. Effective influence occurs when the characteristics of each of these three components are optimized with respect to their compliance-inducing effects (for instance, a high status communicator such as a physician prescribing a life-saving medical regime to a sick person), and when the three characteristics are part of a matched system (for instance, politicians giving campaign speeches to members of their own parties). Bochner (1989, 1994, 1996) has summarized this complex interaction with the phrase: ‘It depends on who says what to whom.’ In terms of this model, the cognitive warning sign literature tends to concentrate on the message. Laughery and Brelsford (1991) have discussed some of the relevant interactions but the present suggestion is that studies of warning signs might well benefit from being more explicitly studied within the context of social influence research.
Two theoretical models prominent in social influence seem to be particularly relevant to the problem of designing effective warning signs. These are the Health Belief Model (Becker, 1974), and the Theory of Reasoned Action (Ajzen and Fishbein, 1980), both of which have been used extensively in designing health education campaigns (Baum et al, 1984; Fishbein, 1982), where the issue of compliance also looms large. According to these models, the intention to comply with warning signs will depend on the workers’ believing that the hazard is likely to happen to them, that it will have serious consequences for them, that the specific action recommended by the sign will prevent or reduce the likelihood of injury, and that taking the recommended preventive action will not have other adverse consequences that would cancel out any benefit from complying. Within these models the utility of the various principles of sign design listed earlier is a function of the extent to which they draw attention to such considerations. However, placing them within a coherent theoretical system should provide more explicit guidelines on how to translate the principles into action, in particular in composing the message contents of the signs. In addition, the role of normative influences also has to be taken into account. These include beliefs about whether significant other persons would approve or disapprove of performing the behaviour. In industrial settings this usually means the peer group and the company culture, the latter referring to the presence or absence of clear safety rules and whether these are strictly and uniformly enforced. In summary, the response to a sign will depend not only on whether it is seen and understood, but also on a cost/benefit analysis concerning the consequences of compliance, the value placed on those consequences and the prevailing norms. In addition to the general determinants of compliance reviewed above, there are also specific processes that act as barriers. Those likely to be operating in the warnings signs domain include psychological reactance, the Third Person Effect, and the False Consensus Effect. These three processes can explain why increasing the shrillness of the message, usually by making it more fear arousing, may not be effective. Psychological reactance (Brehm, 1966; Brehm and Brehm, 1981) is based on the assumption that people do not like to have their autonomy curtailed, and will oppose any attempt to influence them which could imply diminished control over their own behaviour. However, although there is considerable empirical evidence to support this account, it does not satisfactorily deal with the question of why people should resist information that is potentially of great benefit to them. More recently, two overlapping theoretical models have been developed which can explain why persons with unsafe work habits may resist attempts to modify their behaviour. The two constructs are the False Consensus Effect and the ¹hird Person Effect. The False Consensus Effect (Mullen et al, 1985) refers to individuals’ assuming that their values and practices are more widely shared than is empirically the case, thereby greatly overestimating the number of people who support their views (Mullen and Hu, 1988). Persons who engage in dangerous practice are particularly prone to the False Consensus Bias, because it serves to reassure them that they are not alone, and that many other people behave as they do (Ross et al, 1977; van der Plight, 1984).
Designing warning signs: A. Adams et al.
The second construct is the Third Person Effect, which was originally introduced to describe the tendency to assume that persuasive messages, particularly in the mass media, exert a stronger influence on others (third persons). For instance, people will assert that others are far more vulnerable to advertising, or to a political campaign speech than they are (Davison, 1983). Technically, individuals tend to overestimate the effect of persuasive communications on others, and underestimate their own susceptibility to influence. The Third Person Effect has received considerable empirical support (Perloff, 1993) and can account for the phenomenon of psychological reactance referred to earlier. Individuals may have a general, negative stereotype of the gullibility of others (Rucinski and Salmon, 1990), and will seek to enhance their own self-esteem by distinguishing themselves as someone who is unique, and does not readily submit to influence (Brewer, 1991; Duck et al, 1995). The Third Person effect may also account for a sense of invulnerability to harm by individuals who live dangerously, because it tends to shift the nature of the problem on to other people: ‘This message has nothing to do with me, it is for them.’ They will justify this by attributing greater experience, or skill, or luck to themselves relative to the average ‘other’ person. In the present study, only the Third Person Effect was manipulated, because to a large extent the other two principles are subsumed by it. The hypothesis being tested is that workers will underestimate their own risk relative to the vulnerability of other employees.
Overview of the study The hypotheses derived from previous research, and implied by the three theoretical models reviewed above, were tested in both field and laboratory settings. Subjects were presented with five different versions of several warning signs. The experimental manipulation consisted of systematically varying the presence or absence of some of the message characteristics reviewed above. Three categories of dependent variables were measured. Subjects rated the signs for (a) their perceived effectiveness; (b) the extent to which they would comply with the messages the signs conveyed; and (c) the extent to which they believed that other people would comply.
249
students, conclusions that could then be applied retrospectively to some of the existing literature. Stimulus materials The stimulus materials were a set of five warning signs. They were selected from a larger collection of signs observed and photographed from visits to a number of industrial sites in the Sydney area. The final set was chosen to represent the diversity of signs referring to hazards in a heavy industry environment. They ranged from a sign with only a few words to one with long and complex instructions regarding clothing requirements. As is often the case with industrial warning signs, one or more of the desirable elements discussed earlier were absent from each of the signs. All of the signs lacked a precise statement of the consequences, and each of the signs failed to include at least one other element. When an element was missing an appropriate element was generated. The originals of the signs, redrawn for the present experiment, are shown in Figure 1, and details of the additional elements required to produce a ‘full’ version are given in ¹able 1. Five different versions of each sign were prepared as the stimuli. The first was a reproduction of the ‘original’ version. The second was a ‘full’ version with all the required elements present, which meant adding, where they were missing, an appropriate signal word, hazard statement, consequences statement, and instructions on how to avoid the stated hazard(s). The next three versions were variants of the full version, one omitting the hazard statement, another omitting the consequences statement and the final one omitting the instructions. To create the versions which were used in the study, a mock-up of each original sign was prepared graphically, based on photographs of the originals, using black text on a yellow background with a thin black line border. Additional elements were added or removed, where necessary, to produce the other versions. Care was taken to ensure that the layout of each of these constructed versions conformed to the style of the original as closely as possible. Each sign was laminated for protection. Finally, an A4-sized photograph of a large entrance
Method Rationale Most of the previous research into the effect of the various message components of warning signs has been conducted in the laboratory, usually with students and usually using specially constructed signs. As the ecological validity of this literature may be limited, the decision was made to base the present study on actual warning signs in current use in heavy industry. To assess the utility of using students in such studies, two groups of subjects participated, namely a group of actual workers, who were tested in naturally occurring field settings, and a group of students. This served two purposes: It increased the ecological validity and hence the generalizability of the results; and by comparing the results of the two groups it enabled inferences to be drawn about the utility of conducting such research in the laboratory with
Figure 1 The original version of the signs used in this study
Designing warning signs: A. Adams et al.
250
Table 1 The additions required to make up a ‘full’ version of each sign Sign Clothing
Forklift
Molten steel
Lead
Transfer car
Additional signal word DANGER Additional hazard FROM HOT AND/OR statement SHARP OBJECTS
— —
— —
DANGER LEAD IN THE AIR
— MOVING TRANSFER CARS
Additional consequences statement
YOU COULD EASILY BE INJURED OR KILLED
YOU COULD EASILY BE INJURED OR KILLED
YOU COULD GET LEAD POISONING
YOU COULD EASILY BE INJURED OR KILLED
LOOK UP! LOOK AROUND!
LOOK UP! LOOK AROUND!
—
—
YOU CAN EASILY INJURE POORLY PROTECTED SKIN
Additional instructions — statement
doorway to an industrial workshop was produced. A rectangle on the photograph indicated where the sign would be located, to one side of the doorway. The procedure, detailed below, consisted of showing the respondents the photograph of the door and pointing out the position of the sign. The relevant sign was then placed on the table next to the photograph, and respondents rated the signs on a brief questionnaire. Measures Subjects rated each sign on several dimensions. The measures consisted of estimates of the effectiveness of the signs, and behavioural intentions of compliance. The questionnaire was deliberately kept brief. It consisted of 12 items of which three were open ended and six were rating scales. In two items subjects had to rank order a set of signs. The questionnaire items will be described in conjunction with the presentation of the results. Design Each respondent was presented with a different version of each of the five signs. The sign, version and the order of presentation were randomly selected, with the restriction that, overall, each version of each sign was rated an equal number of times. In the final two questions, which involved a rank-ordering task, each respondent saw all five versions of two randomly selected signs. (Time permitted only two — the two that each subject saw were balanced across the whole study.) Subjects Two groups of subjects participated in the study. Group 1 included 40 blue-collar workers employed by four different heavy-industry companies — 37 males and 3 females, ranging in age from five participants in the 15—24 age group, to 10 in the 35—44 age group. None of the signs shown in this study were used in the workplaces of any of these subjects. Group 2 included 44 first year psychology students — 10 males and 34 females, ranging in age from 38 participants in the 15—24 age group to three in the 35—44 age group. Thus, the two samples differed in their age and gender composition, group 1 (the workers) were older and contained a larger proportion of males than group 2 (the students).
Procedure The industrial worker participants were tested individually at their place of work, in a quiet room away from the production line. They were shown the photograph of the entrance to a factory, as described earlier. They were then shown the randomly selected sign that had been allocated to that particular experimental condition and asked to imagine that the sign was beside the doorway shown in the photograph. With the photograph and sign in view, the experimenter then administered the questionnaire verbally in the form of a structured interview, recording the responses in the relevant spaces on the schedule. This method was used to forestall any difficulties that might have arisen due to poor English or literacy with some of the respondents. The University student participants were tested individually in a University laboratory. The same procedure was used, except that the students completed the questionnaire themselves. However, the experimenter was present to answer any questions.
Results and discussion Each question (or set of questions) was analysed separately, in terms of the hypothesis that had generated that particular item. The presentation of the results reflects this pattern, in turn summarizing the contents of each item, the prediction being tested, and the outcome. The worker/student distinction One of the aims of this study was to determine whether there were important differences between the worker and student populations in their reaction to the questions regarding warning signs. For each of the analyses, a preliminary investigation of the worker/student variable was conducted. In only two cases was there a significant worker/student effect or interaction. One was an interaction between that variable and the rankings given to signs in one of the final questions (Lead, Obeyed by self, Students"4.72, Workers"3.38, Bonferroni-corrected p"0.05). The other, discussed below, was due to the slightly differing procedures used with workers and students. Overall, it can be stated that, for the present series of questions at least, there were no important differences between the two groups. Results are therefore presented collapsed over this variable.
Designing warning signs: A. Adams et al.
Signal word As noted earlier, a major design rule for warning signs is that they should contain a clear hazard-alerting signal word to draw attention to the sign’s safety-promoting function. But are these words seen as being the prominent keyword in the signs in which they appear, as the rule would imply, or are they simply seen as part of the sign’s background? The first question investigated this using an open-ended format: ‘In the sign, is there one keyword that alerts you to the danger?’ Respondents could answer Yes or No, and if yes, they were asked to say what the word was. The distinctiveness of the various signal words was measured by the frequency with which they were chosen as the keyword when they were present. The three signal words that were shown all achieved high distinctiveness ratings: caution was chosen in 83% of the trials in which it occurred, beware in 82%, and danger in 77%. However, for this final figure, only, there was a significant difference between workers and students. The signal word danger was chosen 90% of the time by students but only 66% of the time by workers. This was almost certainly due to students seeing the written word ‘danger’ in the question, whereas workers were asked the question verbally, which would have drawn less attention to the visual qualities of that word. Overall, when the words caution, beware and danger appeared in a sign their presence was noticed and they were generally, although by no means universally, regarded as constituting the key alerting element in the sign. It should be noted that these alerting words were in all cases printed in rather larger type than the rest of the sign, so it may simply be that the largest word gets the attention. However, each sign did include other words more specific to the relevant danger, such as hot, sharp, forklifts, cranes, molten steel, poisoning, injured, killed, but only killed was selected more than four times—on 7.2% of the 153 occasions on which it was seen. The signs in which there was no large and clear signal word were the originals of the Clothing and Lead signs. For both of these signs the word danger, even though it did not occur in the sign, was chosen as the alerting keyword a surprising number of times: 71% for the Clothing sign and 77% for the Lead sign. Those respondents who had produced a signal word in response to the previous item were then asked, ‘How effective is this word in drawing attention to the hazard?’, indicating their responses on a five-point scale ranging from Not at all effective [1], to Extremely effective [5]. Means were calculated for the three most frequent words, and for the sum of all of the other words, as follows: Danger "4.04; Beware "3.64; Caution "3.51; Others combined "3.30. A Tukey’s Post hoc test showed that danger was perceived to be significantly more effective than caution (p(0.01), and that danger was also perceived to be significantly more effective than the others combined (p(0.001).
The role of consequences and instructions statements As noted earlier, the literature regarding the construction of warning signs suggests that it is important that each sign should contain a statement of the hazard, a statement of the consequences of ignoring the sign and some instructions as to how to avoid those consequences. As it was necessary to keep the questionnaire short, questions
251
were restricted to the second and third of these components, as follows. Consequences. The relevant questionnaire item had three parts. The first drew attention to the ‘‘consequences’’ characteristics of the sign by asking, ‘What, if anything, does the sign say could happen to you if you ignore its message?’ Of the 420 presentations, a consequences statement was actually present in the sign on 253 occasions, and one was reported as being present on 196 of those 253 occasions. It was reported as present on 100% of the occasions it was present for the Forklift, and from 66 to 77% of the occasions it was present for the other signs. The higher figure for the forklift truck was possibly because it was the only generally recognizable hazard among the five signs used, so respondents did not have to think about the sign in order to generate a plausible consequence. Those who reported a consequence were then asked, ‘If there is something that the sign says could happen to you, to what extent do you believe that these things could actually happen to you?’ The word ‘extent’ was deliberately used to suggest an overall subjective combination of probability and severity. They were also asked, ‘To what extent do you believe that these things could actually happen to other people?’ The scale for each of these questions was from Unlikely to happen [1] to Very likely to happen [5]. The data are shown in ¹able 2. An examination of the means suggests that the stated consequence was, if anything, seen as less likely to happen to both the respondent and to others when part of the full version than when part of the other versions. However, there were no significant differences. These results do not support the hypothesis that the full version would be superior. Results pertaining to the ‘‘Third Person Effect’’ as it relates to the consequences will be presented in a separate section. Instructions. This item consisted of a three-part question which began by drawing the respondent’s attention to the instructions characteristics of the sign by asking, ‘What, if anything, does the sign say you should do to avoid injury?’ Of the 303 instances in which an instruction was present, one was stated as being present on 289 occasions. Across the five signs the reported frequency ranged from 95 to 98%. Those who reported the presence of an instruction were then asked, ‘To what extent do you believe that these recommendations will work in helping someone to avoid injury?’ The scale was from Definitely will not work [1] to Definitely will work [5]. The results appear in ¹able 3. An examination of the means suggests that the instructions were seen as potentially less efficacious when part of
Table 2 Means (and standard deviations) for the question, ‘If there is something that the sign says could happen to you, to what extent do you believe that these things could actually happen to you?’ and, ‘To what extent do you believe that these things could actually happen to other people?’ scaled from Unlikely to happen [1] to Very likely to happen [5] Version
Happen to ‘you’
happen to ‘other people’
Full No Hazard No Instructions Total
3.19 3.35 3.23 3.27
3.39 3.41 3.39 3.39
(1.14) (1.17) (1.25) (1.19)
(1.10) (1.18) (1.09) (1.12)
Designing warning signs: A. Adams et al.
252
Table 3 Means (and standard deviations) for the question, ‘To what extent do you believe that these recommendations will work in helping someone to avoid injury?’ scaled from Definitely will not work [1] to Definitely will work [5] Version
Mean (SD)
Original Full No hazard No consequences Total
4.08 3.45 3.87 3.69 3.74
(0.98) (1.00) (0.87) (0.98) (0.97)
Table 4 Means (and standard deviations) for the question, ‘How likely is it that you (or other people) would obey the sign if you (or they) saw it in a potentially hazardous situation?’ The scale was from Unlikely to obey it [1] to Very likely to obey it [5] Version
Happen to ‘you’
Happen to ‘other people’
Original Full No hazard No consequences No instructions Total
4.38 4.07 4.27 4.35 4.14 4.24
4.10 3.62 3.93 3.98 3.81 3.88
(0.73) (1.05) (0.83) (0.74) (0.87) (0.86)
(0.87) (1.03) (0.93) (0.91) (1.01) (0.96)
the full version. This was borne out by a significant contrast between the full version and the mean of the other two versions containing instructions [t(283)"2.49, p"0.01]. These results, too, do not support the hypothesis that the full version will be seen as superior to a version with a major component missing. Reported compliance Responses to the questions, ‘How likely is it that you (or other people) would obey the sign if you (or they) saw it in a potentially hazardous situation?’ were analysed as a function of sign version. The means are shown in ¹able 4. An examination of the means suggests that for both questions the full version is considered, overall, to be less likely to be obeyed than the other versions. This was borne out for both the you question [t(414)"2.1, p(0.05] and for the other people question [t(414)"2.9, p(0.01). These results are contrary to the prediction. Since there is no prior prediction as to which of the other versions might be better, an overall one-way ANOVA on the last three versions for each question was appropriate. Neither was significant. Results pertaining to the ‘Third Person Effect’ as it relates to reported compliance will be presented in a separate section. The other data relevant to the question of which version elicited the greatest amount of intended compliance comes from the final question, in which respondents were given all five versions of two signs. This was the first time that they had seen more than one version of a single sign. They were instructed to ‘Take each set of signs and sort
them into five piles according to how likely it is that each sign would be obeyed by you. That is, order the signs from the most likely to be obeyed by you, to the least likely to be obeyed by you.’ A second question asked subjects to sort with respect to how likely it is that each sign would be obeyed by other people. The five piles were anchored at Most likely to be obeyed [1] and Least likely to be obeyed [5]. Separate repeated-measures analyses of variance were conducted for the you and the other people sorts. Each analysis also included a between-subjects comparison for the Student/Worker factor. The mean rankings in each condition are presented in ¹able 5. In the analysis, two predictions were tested. One was that the full version would be ranked higher than all the other versions, each of which omits at least one component. A repeated-measures ANOVA investigated this with the appropriate contrast, which was significant for you [F(1,167)"24.9, p(0.001], and also for other people [F(1,167)"30.7, p(0.001]. The other question asked whether there would be any differences among the three versions which omit only one component. A further repeated-measures ANOVA involved an overall test with only those three versions included. The overall tests were not significant for each question separately, but we note that the means in each question are in the same direction, and when the groups are combined there was a significant overall effect (Roy’s F(2,334)"4.2, p(0.02). Post hoc tests on repeated measures are controversial, but a planned test comparing the No Hazard version with the other two was significant [F(1,335)"8.0, p(0.01] and there was no difference between the No Consequences and the No Instructions versions. Thus, our analysis produced two opposing answers to the question of whether omitting components of the signs would make any difference to the stated likelihood that the signs will, in one way or another, be efficacious. We should note that the present signs are all very much of the self-evident type. That is, as Ayres et al (1989) have pointed out, many warnings are of the type where much of the safety knowledge is included in the pre-existing repertoire of the reader. Even students without any industrial experience could figure out, from any of the versions of the present signs, what it is that they would be required to do or not to do. They may not have known what a ‘transfer car’ was, but when shown a sign outside a workshop door saying ‘beware of transfer car’ they would have been able to infer that the appropriate behaviour would be to watch out for something moving that is large and presumably dangerous. Thus, we suggest that when only one version of each sign was seen the response to our questions was based on the simplicity and clarity of the sign. It is only when the complete range of signs is available that the respondent stops and thinks and reacts in a way which implies that the fuller the information supplied the greater the compliance. Our present conclusion regarding this matter is that simplicity and clarity are what is required of a sign, and
Table 5 Mean (and standard deviations) of the rankings for the five sign version in each of the two questions Question
Full version
Original version
No hazard
No consequences
No instructions
Obeyed by self Obeyed by others Total
2.46 (1.39) 2.41 (1.38) 2.44 (1.38)
3.82 (1.50) 3.96 (1.34) 3.89 (1.42)
2.76 (1.30) 2.64 (1.31) 2.70 (1.31)
3.01 (1.09) 2.98 (1.18) 3.00 (1.13)
2.95 (1.40) 3.00 (1.38) 2.97 (1.39)
Designing warning signs: A. Adams et al.
that pre-existing knowledge within the user population should be taken into account when designing warnings. Strict adherence to a prescription requiring certain components in a sign will not necessarily make the sign better. The practical implication is that the aim in designing a warning should be to make it as simple as possible, consistent with comprehension by the group of workers for whom it is being displayed. Of course, if there is doubt over comprehension then that comprehension should be tested rather than assumed. Details of such testing can be found in Adams (in press). Third person effect Two of the questions already discussed in another context were designed to assess the third person effect. Respondents were asked, ‘If there is something that the sign says could happen to you, to what extent do you believe that these things could actually happen to you?’ This was followed by the question, ‘To what extent do you believe that these things could actually happen to other people?’ The answers to both questions were on a five-point scale ranging from Unlikely to happen [1] to Very likely to happen [5]. For the purposes of this analysis, responses to all of the signs across all of the versions were collapsed, and the following overall means were calculated: Happen to you"3.27; Happen to other people"3.39, which were significantly different [t(197)"3.06, p"0.002]. This finding is consistent with the prediction from the Third Person model that other persons will be regarded as more vulnerable to mishap. Respondents were also asked, ‘How likely is it that you would obey the sign if you saw it in a potentially hazardous situation?’ This was followed by the question, ‘How likely is it that other people would obey it?’ Answers to both questions were given on a five-point scale ranging from Unlikely to obey it [1] to Very likely to obey it [5]. The analysis, identical to the one conducted for the previous item, found the following means: you would obey"4.24, other people would obey"3.88, which were significantly different [t(418)"11.98, p(0.001]. Thus, the respondents felt that they would be more likely to comply with the signs than other people, presumably because they regard themselves as more law abiding and responsible. These findings are consistent with the results of the previous item in which respondents described other persons as more vulnerable, and they relate to the question of devising company policy regarding warning signs. As the literature discussed earlier has shown, and as our everyday observations tell us, even the best signs are not obeyed by 100% of those seeing them. Designing warnings according to good cognitive principles may improve them, but when faced with a strong and basic tendency to imagine that we ourselves are invulnerable, and that it is only others that are at risk, management must seriously consider whether the mere posting of signs is sufficient to ensure that appropriate behaviour will take place. As management is realizing, social aspects associated with the workplace, such as the concept of safety culture (McDonald and Ryan, 1992; Pidgeon, 1991) are important determinants of compliance with safety regimes. We can compare the cognitive and the social variables manipulated in the present study by examining, across the different sign versions, the results of the two questions which compared you vs other people, as shown in ¹ables 2
253
and 4. How do the differences between the ‘full’ version and the versions with a part missing compare to the differences between the you and the other people versions of the questions? For both these pairs of questions there were significant differences for the repeat factor (you vs other people) but not for the version factor [for the first question F(1,225)"16.0, p'0.001 in comparison to F(3,225)"0.48, p"0.69, and for the second question, F(1,331)"122.2, p(0.001 in comparison to F(3,331)" 2.19, p"0.09]. An examination of the means provides the simple observation that differences between the versions are relatively small in relation to the differences between the you and the other people questions. In interpreting these relativities there are relevant questions of effect sizes in relation to the range through which a variable was manipulated, but these results certainly suggest that the social variable was of considerable importance.
Conclusions Several caveats must be added to the conclusions from this work. One is due to the inclusion of some students as subjects. However, as there appeared to be only minor differences between the student and worker populations, and as the important differences in the study as a whole were within-subjects, it is unlikely that uniquely studentrelated factors have contributed to the findings. The other possible limitation is that the signs were presented to the respondents in such a way that may have led to more attention being paid to them, or in some way processed more thoroughly, than would be the case in real life. McGrath (1994) has shown that, in a situation where almost no one noticed or responded to warning labels when a behavioural measure was used, the same subjects later stated in a questionnaire that they thought such devices would be effective. Thus, the process of drawing attention to the warnings may have increased the present ratings, but there is no reason why this would have changed the ‘you/others’ interaction. In conclusion, previous findings have shown that, in normal use, that is without attention being unnaturally drawn to them, warnings are often not seen and very often not responded to. The present twofold findings are that, within the limitations of the present signs and of the versions of them investigated, when attention was drawn to warnings the missing ‘components’ within them did not result in their being rated as poorer, and that respondents tended to see warnings as directed more to others than to themselves. Taken together, these results suggest that compliance with warnings will be enhanced by developing social and motivational strategies to increase the likelihood that warnings will be considered relevant and salient to the individual, rather than by attempting to fine tune the warnings themselves. The image of the manager walking under a travelling crane near a sign saying ‘DANGER, keep away from travelling crane’ will probably be more significant in relation to sign compliance than any amount of effort devoted to the details of the sign’s design.
References Adams, A. S. (1998) The role and methodology of usability testing in information design. In »isual information for Everyday ºse: Design and Research. H. Zwaga, T. Boersema and H. Hounhout (eds). Taylor and Francis, London, UK (in press)
254
Designing warning signs: A. Adams et al.
Ajzen, I. and Fishbein, M. (1980) ºnderstanding Attitudes and Predicting Social Behavior. Prentice-Hall, Englewood Cliffs, NJ, USA ANSI Z535.4 (1991) Product Safety Signs and ¸abels. American National Standards Institute, New York, USA Ayres, T. J., Gross, M. M., Wood, C. T., Horst, D. P., Beyer, R. R. and Robinson, J. N. (1989) What is a warning and when will it work? Proceedings of the 33rd Annual Meeting of the Human Factors Society, Human Factors and Ergonomics Society, Santa Monica, CA, USA, pp 426—430 Baum, A., Taylor, S. E. and Singer, J. E. (eds) (1984) Handbook of Psychology and Health, Vol 4, Social Psychological Aspects of Health. Lawrence Erlbaum, Hillsdale, NJ, USA Becker, M. H. (ed.) (1974) The health belief model and personal health behavior Health Education Monographs 2 Bochner, S. (1989) The effectiveness of political campaigning: a field test of reference group theory Australian J Psychol 41, 61—68 Bochner, S. (1994) The effectiveness of same-sex versus opposite-sex role models in advertisements to reduce alcohol consumption in teenagers Addictive Behaviors 19, 69—82 Bochner, S. (1996) Pre-election perceptions of politicians and their promises as a function of the reference group match between speaker and listener Asian J Commun 6, 50—71 Brehm, J. W. (1966) A ¹heory of Psychological Reactance. Academic Press, New York, USA Brehm, S. S., and Brehm, J. W. (1981) Psychological Reactance: A ¹heory of Freedom and Control. Academic Press, NY, USA Brewer, M. B. (1991) The social self: On being the same and different at the same time Personality Social Psychol Bull 17, 475—482 Davison, W. P. (1983) The third-person effect in communication Public Opinion Quart 47, 1—15 DeJoy, D. M. (1989) Consumer product warnings: review and analysis of effectiveness research Proceedings of the 33rd Annual Meeting of the Human Factors Society. Human Factors and Ergonomics Society, Santa Monica, CA, USA, pp 936—939 Duck, J. M., Hogg, M. A., and Terry, D. J. (1995) Me, us and them: Political identification and the third-person effect in the 1993 Australian Federal Election European J Social Psychol 25, 195—215 Edworthy, J and Adams, A. (1996) ¼arning Design: A Research prospective. Taylor and Francis, London, UK Fishbein, M. (1982) Social psychological analysis of smoking behavior in Eiser, J. R. (ed) Social Psychology and Behavioral Medicine. Wiley, Chichester, UK Laughery, K. R. and Brelsford, J. W. (1991) Receiver characteristics in safety communications Proceedings of the 35th Annual Meeting of the Human Factors Society. Human Factors and Ergonomics Society, Santa Monica, CA, USA, pp 1068—1072 Laughery, K. R. and Stanush, J. A. (1989) Effects of warning explicitness on product perceptions Proceedings of the 33rd Annual Meeting of the Human Factors Society. Human Factors and Ergonomics Society, Santa Monica, CA, USA, pp 431—435
Lehto, M. R. and Miller, J. M. (1986) ¼arnings: »ol I: Fundamentals, Design and Evaluation Methodologies. Fuller Technical Publications, Ann Arbor, MI, USA Leonard, S. D., Matthews, D. and Karnes, E. W. (1986) How does the population interpret warning signals? Proceedings of the 30th Annual Meeting of the Human Factors Society. Human Factors and Ergonomics Society, Santa Monica, CA, USA, pp 116—120 McDonald, N. and Ryan, F. (1992) Constraints on the development of safety culture: a preliminary analysis Irish J Psychol 13, 273—281 McGrath, J. M. (1994) Perceptions and responses to warning labels: Implications for products liability and litigation J Products ¹oxics ¸iability 16, 19—32 Mullen, B., Atkins, J. L., Champion, D. S., Edwards, C., Hardy, D., Story, J. E., and Vanderklok, M. (1985) The false consensus effect: a meta-analysis of 115 hypothesis tests J Exp Social Psychol 21, 262—283 Mullen, B. and Hu, L. (1988) Social projection as a function of cognitive mechanisms: two meta-analytic integrations Brit J Social Psychol 27, 333—356 Peters, G. A. (1984) A challenge to the safety profession Professional Safety 29, 46—50 Perloff, R. (1993) Third-person effect research 1983—1992: a review and synthesis Int J Public Opinion Res 5, 167—184 Pidgeon, N. F. (1991) Safety culture and risk management in organizations J Cross Cultural Psychol 22, 129—140 Ross, L., Greene, D. and House, P. (1977) The false consensus effect: an egocentric bias in social perception and attribution processes J Exp Social Psychol 13, 279—301 Rucinski, D. and Salmon, C. T. (1990) The other as the vulnerable voter: An experiment of the third-person effect in the 1988 campaign Int J Public Opinion Res 2, 345—368 Sherer, M. and Rogers, R. W. (1984) The role of vivid information in fear appeals and attitude change J Res Personality 18, 321—334 Silver, N. C. and Wogalter, M. S. (1989) Broadening the range of signal words Proceedings 33rd Annual Meeting of the Human Factors Society. Human Factors and Ergonomics Society, Santa Monica, CA, USA, pp 555—559 van der Plight, J. (1984) Attributions, false consensus and valence: two field studies J Personality Social Psychol 46, 57—68 Wogalter, M.S., Desaulniers, D.R. and Godfrey, S. S. (1985) Perceived effectiveness of environmental warnings Proceedings of the 29th Annual Meeting of the Human Factors Society. Human Factors and Ergonomics Society, Santa Monica, CA, USA, pp 664—668 Wogalter, M. S., Jarrard, S. W. and Simpson, S. N. (1992) Effects of warning signal words on consumer product hazard perceptions Proceedings of the 36th Annual Meeting of the Human Factors Society. Human Factors and Ergonomics Society, Santa Monica, CA, USA, pp 935—939