Professional Documents
Culture Documents
ALPHONSE CHAPANIS
To cite this article: ALPHONSE CHAPANIS (1994) Hazards associated with three
signal words and four colours on warning signs, Ergonomics, 37:2, 265-275, DOI:
10.1080/00140139408963644
Article views: 89
Download by: [York University Libraries] Date: 24 May 2016, At: 08:39
ERGONOMICS, 1994, VOL. 37, NO. 2, 265-275
This study assessed perceptions of hazard levels associated with three signal
words, CAUTION, WARNING, and DANGER, combined with four background
colours, white, yellow, orange, and red. In general, DANGER was perceived as
Downloaded by [York University Libraries] at 08:39 24 May 2016
1. Introduction
Warning signs usually have two principal pans: An upper panel containing a signal
word that identifies the level of the hazard, and a lower panel containing messages
that identify the nature of the hazard, the consequences that could result from
ignoring the hazard, and instructions on how to avoid the hazard. Although attempts
have been made to develop standardized terminology to indicate the level of hazard
present in a particular product, different organizations recommend different signal
words (see Lehto and Miller (1986: 153-154) for a detailed comparison of seven
different sets of recommendations). Among the most popular signal words
recommended are: DANGER, to indicate the highest level of hazard; WARNING, to
indicate an intermediate hazard; and CAUTION, to indicate the lowest level of
hazard. Most standards also recommend that the field surrounding the signal word be
coloured, generally red, orange, and yellow for the highest, intermediate, and lowest
levels of hazard, respectively.
Since warning signs figure prominently in many product liability law suits, it is
important to know how people interpret the signal words and colours that appear on
those signs. Bresnahan and Bryk (1 975) were apparently the first to study the levels of
hazard associated with certain elements of accident-prevention signs. They
compared four signal words, DANGER, CAUTION, THINK, and NOTICE, with
and without colour, but colours were confounded with signal words. Red was used
only on the DANGER, yellow on the CAUTION, green on the THINK, and blue on
the NOTICE sign. In addition, groups of signs were presented in a fixed order with
increasing amounts of cues on them. For example, all the signs with no colour were
tested before those with colour. For purposes of this study about the only thing one
can conclude .validly from the data is that their subjects (workers from two light
industries) thought DANGER indicated more hazard than CAUTION. Because of
0014-0139/94 $10.00 0 1994 Taylor & Francis Lid.
266 A. Chapanis
their order of presentation one cannot validly conclude anything about the use of
colour.
Dunlap et al. (1986) had participants rate seven signal words (Attention, Beware,
Caution, Danger, Notice, Think, and Warning) and six colour names (Blue, Green,
Orange, Red, White and Yellow) 'on the basis of the degree of PERSONAL hazard
you believe is implied or associated with each item'. Colour names, not actual
colours, were rated individually, that is, they were not com'bined with signal words.
Participants were 1169 IBM employees from plants in the USA, Canada, and
Belgium. The mean ratings for the signal words from highest to lowest perceived
hazard were: Danger, Warning, Caution, Beware, Attention, Notice, and Think.
Danger stood out by itself, and the mean ratings for Warning, Caution, and Beware
were very close to each other. The mean ratings for colour names from highest to
lowest perceived hazard were: Red, Orange, Yellow, Blue, Green, and White. Some
ratings differed significantly according to the native language of participants. For
Downloaded by [York University Libraries] at 08:39 24 May 2016
example, persons whose native language was French, German, or Scandinavian rated
Caution as indicating a greater degree of hazard than Warning.
Ursic (1 984) had students study display boards containing information on three
hypothetical brands of bug killers and hair dryers. One of the factors studied was the
safety warning. On some displays the warning said DANGER, on others CAUTION,
and on still others the signal word was omitted. Although Ursic shows no data on this
point, he says that the strength of the signal word had no impact on perceptions of the
brand or memory of the safety message.
Negative findings have also been reported by Leonard et al. (1986) who had
students (1) rate the perceived riskiness of signs describing 12 hazardous
situations; and (2) say whether they would obey or disregard the sign in particular
situations. All signs were prepared with either the signal word DANGER,
WARNING, or CAUTION, and some signs were printed either in red or in black.
Their results show no differences among the signal words or between the two
colours. In a later study Leonard el al. (1989) found that subjects could assign the
words DEADLY, DANGER, WARNING, CAUTION, and ATTENTION to
match fairly closely the riskiness of a set of hazard scenarios that had been
determined independently.
Wogalter et al. (1987) report the results of two laboratory experiments, three
rating experiments, and three field demonstration studies. Signal words were varied
in one of the laboratory experiments and the three rating experiments. In the first
laboratory experiment the type of signal word (WARNING vs NOTE) did not affect
compliance with a warning message. In the three rating studies, deleting the signal
word (DANGER, WARNING, or CAUTION-for different hypothetical situations)
decreased mean ratings of the effectiveness of warnings, but deletion of the signal
word had a smaller effect on ratings than deletion of the consequence, instruction, or
hazard statement.
Three studies (Leonard et al. 1988, Silver and Wogalter 1989, and Wogalter and
Silver 1990) have scaled a number of signal words, including CAUTION,
WARNING, and DANGER. All agree in finding that DANGER is perceived as
indicating more hazard than WARNING or CAUTION and that there appears to be
little difference between the latter two words.
With the exception of the investigations by Bresnahan and Bryk (1 975) and that
by Dunlap et al. (1986), all the studies cited have used college students as subjects
and none has systematically studied actual colours in combination with signal words.
Hazards and warning signs 267
2. Purpose
This study contributes additional data on perceptions of hazard levels associated
with three signal words, DANGER, WARNING, and CAUTION, by subjects
selected from the general population. The contribution of colour was also studied by
testing each signal word with four background colours, white, yellow, orange, and
red.
3. Method
3.1. Subjects
Eleven female and seven male subjects were solicited by advertising in a
neighbourhood newsletter. The only restriction placed on their participation was that
they have normal colour vision, as determined by the A 0 H-R-R
Pseudoisochromatic Plates (Hardy et al. 1957). The male subjects had a mean age of
36.0 years (range = 23-51), and an average of 15.3 years of education
Downloaded by [York University Libraries] at 08:39 24 May 2016
(range = 13- 18). The female subjects had a mean age of 35-5 years (range = 2 1-62)
and an average of 13-6 years of education (range = 12- 16). All subjects were tested
individually and paid for their participation.
4.2. Results
Fifteen subjects picked the words DANGER, WARNING, and CAUTION, in that
A. Chapanis
ACAUTION
Do Not U s e
This Equipment
Without Proper
Instructions
Downloaded by [York University Libraries] at 08:39 24 May 2016
AWARNING ADANGER
Do Not Use Do Not Use
This Equipment This ~qui'pment
Without Proper Without Proper
lnstruct,ions Instructions
order; three picked the order DANGER, CAUTION, and WARNING. In other
words, all subjects agreed that DANGER was the strongest word, and although most
subjects thought WARNING was stronger than CAUTION, there was no unanimity
on this point.
5. Part 2: Level of hazard associated with signal word and colour combinations
5.1. Procedure
At the conclusion of Part 1, all 12 cards were laid out in front of the subject (with a
Hazards and warning signs 269
different random order for each subject), and the experimenter read the following
instructions:
OK, fine. Now here are 12 signs. You'll notice that I have added some signs with
colours. Look at them carefully, and pick out the one that is the least forceful, the
least emphatic, the weakest, and put it here [pointing to the left end of the table].
Fine. Now pick out the sign that is the most forceful, the most emphatic, the
strongest and put it here [pointing to the right end of the table].
When the subject had made the two selections, the experimenter said,
Now put all the others in a row between these two according to where you think
they fall between the two at the ends. Take your time. You can change your mind
as often as you like.
Downloaded by [York University Libraries] at 08:39 24 May 2016
Table 1. Mean rankings of the hazards associated with combinations of signal words and
colours.
Table 2. Analysis of variance of the mean rankings of the hazard associated with
combinations of signal words and colours.
5.2. Results
The results of this part of the study are summarized in table 1. The data for the signal
words agree with the findings of Part 1. On the average CAUTION was ranked least
forceful, WARNING intermediate, and DANGER strongest. An analysis of variance
of the data (table 2) shows that the differences among the signal words and among the
colours are highly significant. The colours, in order from weakest to strongest are:
white, yellow, orange, and red.
The interaction of words with colo'urs was significant and is due to the slightly
lower average ratings for yellow-CAUTION, and slightly higher average ratings for
yellow-WARNING, yellow-DANGER, and red-CAUTION than one would expect
from the marginal values.
Post hoc analyses show that the difference between CAUTION and WARNING is
not statistically significant by the Scheffi test, but is significant at approximately the
5% level by the Newman-Keuls test. Differences between DANGER and each of the
Downloaded by [York University Libraries] at 08:39 24 May 2016
other two words is highly significant ( p < 0.001) by both tests. The difference
between white and each of the other colours is highly significant ( p < 0.001) by both
the Scheffk and Newman-Keuls test; the difference between red and orange is
significant (0.025 >p > 0.01) by the Scheffi test and highly significant ( p < 0.001) by
the Newman-Keuls test. The difference between yellow and orange is not statistically
significant by either test.
Since the raw data for the analysis in table 2 are ranks assigned by individual
subjects to the 12 signs, there is no variability among subjects, resulting in a
somewhat unconventional analysis. The table shows the literal meaning of each of
the interaction terms with their conventional equivalents in parentheses.
The data were also analysed by Friedman two-way analyses of variance by ranks
for the word and colour data separately. The net result was that the two principal sets
of differences (among words and colours) are highly significant, no matter how one
analyses them. Since the analysis of variance has been shown to be robust even when
assumptions underlying it have been violated, I show only the results of that analysis
in table 2.
6.2. Results
The results of this part of the study are shown in table 3. There is almost complete
Hazards and warning signs
agreement that DANGER should be associated with red. Results for the WARNING
and CAUTION signs indicate little agreement.
7.2. Results
The results in table 4 are copied exactly from the experimenter's protocols, which, in
a few instances are cryptic. The general impression one gets from these responses is
that people clearly associate the word DANGER with the highest hazard level. There
is, however, some ambiguity about CAUTION and WARNING. Subject 2, for
example, did not regard WARNING as much different from CAUTION, and Subject
1 I regarded WARNING as less serious than CAUTION. Subject 16 commented on
the possibility of damaging equipment and getting into difficulty with authorities,
but not about personal injury, a response that might, however, have been
conditioned by the message on these cards.
'May damage the equipment' 'Damage to equipment and perhaps self 'Don't d o anything without reading
instructions; personal danger'
'Seek advice, instructions; probably 'Just about the same as caution' 'Stop me from using it if tempted; same
Downloaded by [York University Libraries] at 08:39 24 May 2016
Figure 2. Rating scales with the locations (vertical lines)of the median ratings for the Caution
(C), Warning (W), and Danger (D) signal words. The thin horizontal lines show ranges of
ratings by all subjects; the heavier horizontal lines the ranges of ratings by the middle ten
subjects.
8.2. Resulls
For scoring, both scales were marked off into 14 equally spaced intervals and
numbered from 0 (on the left) to 14 (on the right). Each subject's rating was recorded
to the nearest whole integer. The two sets of ratings were not entirely independent.
Coefficients of correlation between them for the CAUTION, WARNING, and
+ +
DANGER signal words, respectively, are 0-64, 0.53, and +0.43.
Figure 2 shows the median ratings for each signal word, the range of ratings for all
subjects, and the range of ratings made by the middle ten subjects. Analyses of
variance showed that the differences among the mean values on both scales were
highly significant. (F(2,34) = 4 1.49 for the seriousness scale and F(2,34) = 53.41 for
the probability of injury scale; p < 0.001 in both instances). Post hoc analyses showed
that on both scales the difference between CAUTION and WARNING was not
significant by the Scheffk test, but was significant at approximately the 5% level with
the Newman-Keuls test. Differences between the ratings for DANGER and either of
the other two words were highly significant ( p < 0.001) by both statistical tests.
The data in figure 2 agree with what was found earlier about the degree of hazard
associated with the three signal words. The median ratings for CAUTION and
WARNING are well below the midpoints of both scales, and they are much closer to
each other than to the median rating for DANGER. In addition, the ranges of ratings
for CAUTION and WARNING overlap almost completely. Four subjects rated
WARNING lower than CAUTION on both scales. This cannot be explained by the
order in which the words were presented to the subjects, because for two of the four
subjects the word CAUTION preceded the word WARNING. One other subject gave
the same rating to CAUTION and WARNING on both scales, and another rated
WARNING lower than CAUTION on the seriousness scale, but not on the
probability scale.
274 A. Chapanis
.9. Discussion
This study has at least three important limitations. First is the small number of
subjects tested. Second, the five parts of the study were done serially and findings in
the later parts were undoubtedly affected by what subjects had already done in earlier
parts. Third, this study measured only how people say they interpret hazard levels
associated with signal words and colours in a laboratory situation. It does not address
the critical issue of whether people would behave differently when confronted with
warning signs so labelled.
Nonetheless, the findings in the five parts of this study are consistent with each
other and with the findings of several othe: studies that were conducted in different
ways and with different procedures. Despite the small number of subjects tested,
average results in this study are highly significant from a purely statistical standpoint.
However, the only really strong finding is that of all the combinations tested
DANGER on a red background is associated with the greatest amount of hazard.
Downloaded by [York University Libraries] at 08:39 24 May 2016
standards for warning signs. In view of the ambiguity that seems to be attached to
signal words and colours, it would also be useful to have more research on alternative
modes of warning, such as symbolic or iconic coding.
Acknowledgement
The data for this study were collected in 1983 in my Communications Research
Laboratory with private financing. I am indebted to Kristine A. Haig for scheduling
and testing the subjects in this study.
References
BRESNW, T. F. and BRYK,J. 1975, The hazard association values of accident-prevention
signs, Professional Safity, 20 (I), 17-25.
DUNLAP, G . L., GRANDA,R. E. and KUSTAS,M. S. 1986, Observer Perceptions oj'lmplied Hazard:
Safety Signal Words and CoIor Words (Technical Report TR 00.3428, IBM,
Poughkeepsie, NY).
HARDY, L. H., RAND, G. and R~TLER, M. C. 1957, A 0 H-R-R Pseudoisochromatic Plates, 2nd
edn (American Optical Company, Stockbridge, MA).
LEHTO,M. R. and MI=, J. M. 1986, Warnings. Volume I: Fundamentals, Design and
Evaluation Methodofogies (Fuller Technical Publications, Ann Arbor, MI).
LEONARD, S. D., HILL, G. W., IV,and LRMS, E. W. 1989, Risk perception and use of warnings,
Proceedings of the Human Factors Society 33rdAnnwl Meeting (Human Factors Society,
Santa Monica, CA), 550-554.
LEONARD, S. D., KARNES,E. W. and SCHNEIDER, T. 1988, Scale values for warning symbols and
words, in F. Aghazadeh (ed.) Trends in Ergonomics/Human Facrors V (Elsevier,
Amsterdam), 669-674.
LEONARD, S. D., MATIHEWS,D. and KAFNES,E. W. 1986, How does the population interpret
warning signals? Proceedings of the Human Factors Society 30th Annual Meeting
(Human Factors Society, Santa Monica, CA), 116-120.
SILVER,N. C. and WOOALTER, M. S. 1989, Broadening the range of signal words, Proceedings o j
the Human Factors Society 33rd Annual Meeting (Human Factors Society, Santa
Monica, CA), 555-559.
URSIC,M. 1984, The impact of safety warnings on perception and memory, Human Factors, 26,
677-682.
WOGALTER, M. S., GODFREY, S. S., FONTENELLE, G. A., DESAULMERS, D. R., ROTHERSTW, P. R.
and LUGHERY, K. R. 1987, Effectiveness of warnings, Human Factors, 29, 599-612.
WOGAL~FR, M. S. and SILVER,N. C. 1990, Arousal strength of signal words, Forensic Reports, 3,
407-420.