“IF you drink alcohol, THEN you will get cancer” –Investigating how reasoning
accuracy is affected by pictorially presented graphic alcohol warnings
Daniel Zahra1*, Rebecca L. Monk2, and Emma Corder3
1
Department of Psychology, Plymouth University
2
Department of Psychology, Edge Hill University
3
Department of Psychology, Oxford Brookes University
*Corresponding author. Dr Daniel Zahra, C506 Portland Square, Plymouth University,
Plymouth, Devon, PL4 8AA. e: daniel.zahra@plymouth.ac.uk
Running Head: Reasoning accuracy in pictorial warnings
Word Count: 3,486
Key Words: Reasoning, Pictorial Warnings, Visual Imagery, Inferences, Alcohol
Declaration of Interests
This research was funded by the Plymouth University Department of Psychology. The
authors have no conflicts of interest to declare.
1
Abstract
Aims: To examine variation in reasoning accuracy when evaluating emotive pictorial health
warnings and their outcomes.
Methods: In Study 1 individuals completed a version of the Wason selection task evaluating
warnings in which content type (Alcohol, Non-Alcohol) and emotional valence (Positive,
Negative) were manipulated using different images. Study2 presented people with If-Then
warnings (alcohol-related or non-alcohol-related) and asked them to evaluate whether certain
outcomes followed.
Results: Study 1 found that when warning messages were about alcohol, there was no
difference in reasoning accuracy between positive and negative content. However, fewer
correct reasoning responses followed exposure to negative general health messages
(t(152)=3.58, p<.001, d=.19). Study 2 suggested that when a warning involves considering
the potential consequences of drinking alcohol, accuracy is improved when the content is
negative. However, when considering outcomes that will result from abstaining from
consumption, accuracy was greatest when the content was positive. This was supported by an
inference by content interaction (F(3,171)=4.11, p< .01, 𝜂𝑝2 =.07).
Conclusions: Negative imagery should be used with caution in health warnings. In some
cases the use of pictorial images of negative outcomes may improve reasoning, however,
their use in alcohol-related messages does not appear to be consistently beneficial. Goals for
using pictorial warnings therefore need to be carefully considered when designing such
warnings.
2
Highlights

Negative imagery in alcohol warnings does not consistently improve the accuracy of
the inferences people draw from such warnings.

Outcomes of drinking alcohol are judged more accurately when warnings contain
negative imagery.

Outcomes of not drinking alcohol are judged more accurately when warnings contain
positive imagery.

Use of imagery in alcohol warnings should take into account the intended purpose –
whether that is to convey potential risks associated with alcohol consumption, or
promote potential benefits of abstaining.
3
1. Introduction
The reasoning processes and cognitions elicited by exposure to health warnings remain a
complex area requiring further examination (Ruiter, Abraham, & Kok, 2001). However, if
heath messages do not elicit accurate reasoning processes, or if they create avoidant reactions,
their behaviour modification goals may be negated. There is therefore a need to examine
these reasoning processes in greater detail. Most health warnings can be reduced to the form
of conditional statements. That is, “If you do p, then q will happen”. The Wason selection
task presents individuals with rules or warnings and a number of situations on cards.
Participants are then asked to indicate which cards must be turned over to determine whether
the rule is true or false (Wason, 1968). Similarly, other conditional reasoning tasks provide
people with information about the state of the world and then ask what can be inferred based
on this information (if-then logic). The manner in which such conditional statements are
approached is not consistent nevertheless. In conditional reasoning tasks, participants appear
to accept more of the logically valid inferences when the content is related to real-world
situations than when the content is based on abstract concepts such as p’s and q’s (Blanchette
& Richards, 2010; Manktelow, 2000). It is also interesting to note that people tend to be more
logical when reasoning about neutral than positive or negative content (Blanchette, 2006;
Blanchette & Richards, 2004), except when the content relates directly to their experiences
(Blanchette, Richards, Melnyk, & Lavda, 2007). Personal salience and emotive content
therefore appear to be important dictates of reasoning processes. Such findings have clear
relevance to understanding warnings on products such as tobacco and alcohol, given that how
an individual thinks about the information presented is likely to guide their behaviour.
However, to the authors’ knowledge, no work has examined how these effects might alter
with the inclusion of emotive visual imagery. This should be of particular interest within
alcohol research as those expected to draw conclusions from alcohol warnings typically have
4
some experience with alcohol, or are making decisions about their own current experience.
The fact that the emotional content may influence people’s reasoning processes is also highly
pertinent given the ongoing debate in relation to the use of emotive imagery in alcohol
warnings.
Warnings inform the public of potential hazards and direct people away from unsafe acts or
objectionable behaviours (Stewart & Martin, 1994), and tobacco and alcohol warnings are
now found on packaging worldwide (European Commission, 2012; Ford, Moodie, &
Hastings, 2012; Hammond, 2011). Furthermore, Thailand’s government has proposed that
similar images should be introduced on alcohol packaging (O'Brien, 2013), and the UK’s
Faculty of Public Health (2012) has called for a factual “no-nonsense” approach to
combining warning text and images on alcohol packaging. Accordingly, it has been noted
that the ‘fear appeal’ of health messages is important (Witte, 1998) and there is a strong
theoretical basis for such assertions (Leventhal, 1970; Witte & Allen, 2000). Large scale
reviews and meta-analyses have concluded that effectively designed fear appeals and
pictorial health warnings eliciting strong emotional reactions can effectively reduce smoking
(Hammond, 2011; Witte & Allen, 2000). However, O’Brien (2013) questions the
generalisation from research on smoking to alcohol, especially when evidence in favour of
the utility of graphic alcohol warnings is scarce (Thamarangsi & Puangsuwan, 2010). The
basis for assuming fear in health messages improves efficacy has been criticised (Hastings &
MacFadyen, 2002; Sweet, Willis, Ashida, & Westman, 2003) and neuroimaging research
(Kessels, 2014) indicates that fear arousing messages can create a defensive reaction (Janis,
1967; Witte & Allen, 2000), diverting attention away from the warning. Research on the
effectiveness of such warnings on alcohol packaging is also inconsistent (Wilkinson & Room,
2009).
5
An unclear understating of the cognitive processes which occur in response to graphic
health warnings may be a contributor to this inconsistency. It is therefore necessary to
increase our understating of the cognitive and reasoning processes elicited by pictorial and
emotive alcohol warnings (Agostinelli & Grube, 2002). If these warnings are not inducing the
desired reasoning, then a change of approach may be necessary in order that resources be
better allocated to address alcohol consumption. The aim of this research was therefore to
adapt existing paradigms such as the Wason selection task (using real-world alcohol-related
content) and a conditional reasoning task which replaces verbal terms in real-world
conditional warnings with emotive imagery. Utilising both methodologies allows an
investigation of how people respond to warnings, and an examination of what people infer
given a particular warning and situation. The results are framed in terms of logical accuracy,
and in practical terms relate to how accurate people are in evaluating and drawing
conclusions from such warning labels.
2. Pre-testing of Visual Imagery
In order to construct the necessary visual conditionals, 92 simple, high-resolution, and freely
available images were collected by the authors. Images were compiled into an online format
for presentation to an independent group of participants (N=26; 8 male, 18 female; Mean age
28 years, SD=10.67). Each image was presented separately and participants rated valence and
arousal of each picture using the Self-Assessment Manikin (Figure 1), which has been used to
assess similar visual stimuli (Lang, Bradley, & Cuthbert, 2005)
[Insert Figure 1]
6
Images were then formed into pairs which illustrated warnings that were positive or negative,
alcohol-related or not. Based on pre-test ratings, positive pairs were rated as more positive
(M=5.60, SD=0.65) than negative pairs (M=4.59, SD=0.39; F(1,16)=17.15, p=.001, 𝜂𝑝2 =.52),
whilst valence did not vary as a function of content type (Alcohol, M=4.97, SD=0.80; Nonalcohol, M=5.21, SD=0.68), nor did emotion and content interact (F(1,16)=.12, p=.738,
𝜂𝑝2 =.07). Arousal ratings did not differ as a function of emotion (Positive, M=4.44, SD=0.21;
Negative, M=4.59, SD=0.14) or content type (Alcohol, M=4.54, SD=0.14; Non-alcohol,
M=4.49, SD=0.22), nor was there any interaction (F(1,16)=.54, p=.474, 𝜂𝑝2 =.03).
3. Study 1
As outlined above, the Wason selection task requires individuals to test different situations in
order to evaluate the truth of a rule or warning. Applied to alcohol, Griggs and Cox (1982)
report a study where people were told that the law states “If a person is drinking alcohol, then
they must be 21 years old”. They are then asked to imagine they are a police officer entering
a bar where there is a person drinking beer, a person drinking cola, an older gentleman, and a
young child. Their task is to indicate whose age and drink must be determined in order to tell
whether the law is being broken in this scenario. In this case, the correct action is to ascertain
the age of the person drinking beer and determine what the younger person is drinking (by
turning over cards with verbal labels on each side relating to age and drink). If the beer
drinker is under 21 years, or the younger person is drinking alcohol, the law has been broken.
The other cases are irrelevant. The age of the cola drinker and whether the older gentleman is
drinking alcohol or not have no bearing on the law1. The present study adapts this task in
order to investigate how peoples’ evaluations of alcohol-related warnings are affected by the
1
Structurally, this scenario can be represented as “If p (alcohol) then q (21yo)”. The people in the bar represent
states of the world corresponding to p (beer), not-p (cola), q (older man) and not-q (child). The ‘correct’
response, which allows a complete, logically valid evaluation of the law, is to choose p and not-q.
7
inclusion of emotive visual imagery. By manipulating the valence and presentation of the
warning and available options we can establish the effects of visual imagery (positive or
negative) and warning content (alcoholic or non-alcohol beverages). Of primary interest are
positive and negative alcohol-related warnings as any differences here will speak most
directly to the relative effectiveness of negatively valenced graphic warnings. The inclusion
of non-alcohol related warnings provides a comparison group, whilst also informing us more
generally about the impact of emotive imagery in warnings. In the studies reported here,
alcohol-related warnings focussed on the consumption of a range of alcoholic beverages,
including beers, wines, and spirits, whereas non-alcohol related warnings focussed on the
consumption of fruit juices and water.
3.1 Method
Participants
The 153 people (14 men, 139 women; Age, M=21years SD=5years) who participated were
postgraduate volunteers, or psychology undergraduates participating for course credit.
Scrutiny of the undergraduate psychology programmes from which the students were
sampled showed they are unlikely to have been familiar with the Wason selection task prior
to the current study.
Warning Materials
A variant of the Wason selection task was used in which participants were shown a series of
‘cards’, and a warning. For example, “Each card has a liver on one side and the number of
units of alcohol drank in a week on the other. Which cards must be turned over to test the
warning ‘If you drink more than five units of alcohol in a week, you will develop liver
damage’?” The cards presented depict affirming and negating cases of each part of the
8
warning; in this example, >5 units, no alcohol, a healthy liver, and a damaged liver (Figure 2).
Participants could then select 1-4 cards before continuing to the next trial.
[Insert Figure 2]
Procedure, Design and Analytic Strategy
Participants completed 40 online trials, consisting of ten positive warnings and ten negative
warnings. Each ten included 5 alcohol-related and 5 non-alcohol-related. For each trial,
participants were presented with the warning and the card options. They were then allowed to
select the card or cards they thought needed to be evaluated to test the warning before moving
on to the next trial. Each of the warnings also appeared with the terms reversed. For example,
“If you drink alcohol then you will become depressed” also appeared as “If you become
depressed then you will drink alcohol”. This controlled for the converse probability of the
conditional statement which has been shown to affect the inferences people draw (Cummins,
Lubart, Alksnis, & Rist, 1991). Data were subject to a 2 Content Type (Alcohol-related, NonAlcohol-related) by 2 Emotional Valence (Positive, Negative) repeated-measures ANOVA
which focussed on logically correct responses (the proportion of participants’ responses
where p and not-q were selected). Participants were also asked to complete the AUDIT (26),
a 10-item questionnaire designed to screen for individuals suffering from or at risk of
developing alcohol use disorders. The relationship between AUDIT scores and response
profiles was evaluated using correlational analyses prior to consideration of their inclusion in
the above ANOVA.
3.2 Results
9
Table 1 displays descriptive statistics for p-and-not-q choices across trial types, and the
ANOVA revealed a significant main effect of emotional valence (F(1,152)=4.63, p=.03,
𝜂𝑝2 =.03) but not content type (F(1,152)=.002, p=.97, 𝜂𝑝2 < .001).
[Insert Table 1]
Logical accuracy was significantly lower in response to negatively valenced content when
compared to positive content, but there was no difference between alcohol-related and nonalcohol-related content. These findings were qualified by a significant 2-way interaction
between valence and content (F(1,152)=8.58, p< .01, 𝜂𝑝2 = .05), as shown in Figure 3..
[Insert Figure 3]
Post-hoc repeated measures t-tests reveal that when the warning content is alcohol-related,
emotional valence has no significant impact on how often people choose the logically correct
p and not-q combination. However, when the content is not related to alcohol, positively
valenced items lead to significantly more choices of the logically correct combination than
negatively valenced items (t(152)=3.58, p<.001, d=.19). Scores on the AUDIT showed no
relationship to patterns of responding, or to rates of any response in particular.
3.3 Discussion
These findings suggest that the pictorial presentation of negative health outcomes in health
warnings that are not related to alcohol may have unintended and adverse consequences,
potentially decreasing the accuracy with which people process these warnings, and thus
limiting their potential to positively affect behaviour. However, when warning messages were
10
about alcohol, no difference in reasoning accuracy was found between positive and negative
warnings. These findings suggest that pictorial images of negative outcomes do not improve
reasoning about alcohol-related warnings. It may be of interest, however, to note that for both
types of warning in Study 1, accuracy rates are higher than the ~10% typically found for
abstract (lacking real-world content) versions of the selection task (Manktelow, 2000; Wason,
1968).
4. Study 2
Study 1 combined trial valence and content, as well as controlling for image valence and
arousal. However, in real-world warnings valence and content are typically found in certain
combinations. Alcohol warnings are primarily negative (Glock & Krolak-Schwerdt, 2013), as
with smoking warnings (Glock et al., 2013). Indeed, it is for this reason that many have
warned of a potential boomerang effect following exposure to alcohol warnings whereby
readers act contrary to the warning (Ringold, 2002). However, smoking and drinking also
have potentially positive effects (Glock & Krolak-Schwerdt, 2013; Glock et al., 2013). It has
also been noted that alcohol warnings targeting positive expectancies (utilising expectancyrelated messages) may be more effective than negatively worded health warnings. In order to
provide a more ecologically valid comparison of alcohol warnings, study two utilised a
collection of real-world warnings from published work (e.g. Brandish & Sheron, 2010;
Cherpitel et al., 2003; Van Duyn & Pivonka, 2000) which fall into two categories; those
which are negative and alcohol-related, and those which are positive and not alcohol-related.
The terms in these warnings were then replaced with images, and people were asked whether
particular outcomes followed, allowing an investigation of the inferences people accept from
the different warning types.
11
4.1 Method
Participants
Participants were 58 (8 men, 50 women; Age, M=22years SD=6years) psychology
undergraduates and postgraduate students participating voluntarily or for course credit (Age,
M=22 years, SD=5.61).
Warning Materials
A collection of real-world warnings were selected which were either negative and about
alcohol, or positive and not about alcohol. Words in these warnings were then replaced with
images, creating ‘visual conditionals’ - “If p then q” statements in which the terms p and q
are substituted for images. In conditional statements, the “If p” component is called the
antecedent, and the “Then q” component the consequent. Additional information presented in
the form of “Given p” is called a premise, and the component being evaluated (“Does it
follow that q?”) is the inference. Conditional statements, premises, and inferences to be
evaluated can be combined to create four problem types. These are commonly referred to in
the reasoning literature as Modus Ponens (MP), Modus Tollens (MT), denial of the
antecedent (DA), and affirmation of the consequent (AC) 2 . Table 2 details these four
structures, the combinations of premises and inferences which constitute them, and the
logically ‘correct’ responses in evaluation of each inference.
[Insert Table 2]
2
To use the example of Griggs and Cox (1982) again, the logical validity of MP and MT can be seen by
considering the statement “If a person is drinking alcohol, they are aged over 21 years”. Assuming it’s truth,
which participants are typically instructed to do, by MP, if a person is drinking alcohol, then they will be 21
years old or over. By MT, if the person is not 21 years or over, then they will not be drinking alcohol. However,
the rule says nothing at all about how old a person who is not drinking alcohol might be (DA), or what drink
people over 21 years old will have (AC).
12
Recording how often people respond correctly to each problem type indexes their accuracy in
evaluating inferences drawn from images only. This has implications for the simplification of
warnings to image-only formats such as those that are commonly found in a range of settings,
from road-signs to chemical labelling conventions. Images were chosen from the pool created
for this research and negated instances of the image were represented by placing a large red
cross through the image (Figure 4). This generated a selection of ten statements, five of
which were related to alcohol, and five of which were related to other beverages. Because
each statement could be accompanied by a premise image which affirmed p, denied p,
affirmed q, or denied q, each of 5 negative-alcohol and 5 positive-non-alcohol visual
conditionals appeared in each of the four problem types in Table 2.
[Insert Figure 4]
Procedure, Design, and Analytic Strategy
Participants completed 40 online trials. Content types were blocked together and
counterbalanced, and trials within each block were randomised. Accuracy across trial types
was analysed using a 2 Content Type (Negative-Alcohol, Positive-Non-Alcohol) by 4
Problem Type (MP, MT, DA, AC) repeated measures ANOVA. This facilitated an
investigation of the effect of problem type on people’s accuracy in drawing inferences.
Although this does not allow consideration of the interaction between content and valence as
in Study 1, the aim of Study 2 was to examine the accuracy of participants’ reasoning in
response to types of warning messages typically found in relation to alcohol (negative alcohol
warnings), relative to messages of a more positive nature and which were not related to
alcohol (positive non-alcohol warnings). Reasoning accuracy was defined as the proportion
of formally logically correct responses to each type of problem. For MP, this was the
13
proportion of “Yes it follows” responses, for MT, “No” responses were correct For DA and
AC inferences, this was the proportion of “Maybe” responses. AUDIT scores were also
included, as in Study 1.
4.2 Results
Table 3 displays descriptive statistics for the accuracy rates obtained across each trial type.
The ANOVA revealed a significant main effect of problem type (F(2.02,115.24)=170.28,
p< .001, 𝜂𝑝2 =.75), suggesting that participants found MP and MT trials significantly easier
than DA and AC problem types (all LSD post-hoc pairwise comparisons showed significant
differences at p<.05 except between DA and AC inferences).
[Insert Table 3]
Furthermore, although there was no significant difference in response accuracy between
positive-non-alcohol and negative-alcohol trials (F(1,57)=.76, p=.39, 𝜂𝑝2 =.01), content
showed a significant interaction with problem type (F(3,171)=4.11, p<.01, 𝜂𝑝2 =.07), which
suggests that negatively valenced alcohol warnings lead to better reasoning on MP and MT
problems, but positively valenced warnings lead to better reasoning on DA and AC problems
(Figure 5). MP shows the largest (and only statistically significant) difference between
Positive-Non-Alcohol and Negative-Alcohol content (t(57)=-3.47, p=.001, d=-.47), followed
by DA, MT, and finally least difference is found on AC problems between the two content
types. Scores on the AUDIT showed no relationship to patterns of responding, or to rates of
any response in particular.
[Insert Figure 5]
14
4.3 Discussion
Overall, negative-alcohol content appeared to lead to better reasoning on MP and MT but
worse on DA and AC reasoning tasks. More specifically, the current findings suggest that
negative alcohol-related messages lead to significantly more accurate reasoning in situations
most likely to be pertinent to decisions to drink – “What will happen if I drink?” (MP).
Interestingly, the second largest accuracy difference was for the DA inference –“What will
happen if I don’t drink?” In this case, positive non-alcohol related images elicited higher
accuracy than negative alcohol-related images (although this did not reach statistical
significance). The largest differences in reasoning accuracy therefore appear to result from
health messages which relate to reasoning situations most relevant to a drinker – what will
happen if they drink versus if they do not drink.
5. Overall Discussion
Much of the policy around alcohol warnings has been based on a potential overgeneralisation
of the smoking literature (O'Brien, 2013), and there has been little research which has
examined the conditional reasoning processes in relation to exposure to alcohol warnings,
and how these may be affected by warning content. The current work therefore aimed to
address this knowledge-gap - investigating the impact of content type and emotional valence
in warnings. Specifically this entailed comparing positive and negative imagery across
alcohol-related and non-alcohol related warnings. This was done by developing paradigms
that assess peoples’ evaluation of the warnings themselves (Study 1) and investigating their
evaluation of the implications of the warnings (Study 2). Study 1 found no difference in
accuracy of responses between positive and negative warnings about alcohol. Yet, relative to
positive content, negative content reduced reasoning accuracy on other health related
15
warnings. This would seem to suggest that people are no better at evaluating negative than
positive warnings about alcohol and so there is little to be gained in this domain from making
the content overly negative.
There is a distinction here between evaluating emotional content, and the impact of
any emotion generated by the content. Such a distinction underpins work related to anxiety
avoidance which highlights a psychological immune system or rebound effect. This therefore
suggests a potentially fruitful and informative area of future work. However, the current
results appear contrary to what might be expected following the claims used to justify
recently developed graphic warning labels (Hammond, 2011) and supports the assertion that
graphic images are not necessarily beneficial (Thamarangsi & Puangsuwan, 2010).
Nevertheless, the comparable reasoning accuracy presently revealed across positive and
negative warning conditions suggest that there was equally no evidence that negative warning
images had a more detrimental effect than their positive counterparts. Study 2 was therefore
intended to provide a more ecologically valid comparison of negative alcohol-related
messages in contrast with positive non-alcohol-related messages (those typically intended to
promote healthy lifestyles). Here, results suggest that when warnings about alcohol are
negative, people are more accurate at evaluating what outcomes will follow if they drink
(MP). In this respect, the results of Study 2 seem to suggest that when trying to provide clear,
understandable warnings about the consequences of drinking, accuracy can be improved by
making the content negative. However, in the current study this is only the case when
considering what the outcome will be when you drink alcohol. Such warnings therefore
appear of most relevance to individuals already engaging in, or intending to engage in,
alcohol consumption. If the goal of a public health initiative is to prevent initial alcohol
consumption, people are more accurate (though not significantly so) at determining outcomes
about not drinking alcohol (DA) when the content is positive. Concurrently, there was an
16
apparent improvement in the evaluation of non-alcohol warnings when the content is positive
(Study 1). This might be used to improve understanding of the health benefits of activities
other than drinking alcohol.
It should be noted that the current research findings may not represent the cognitive
processes which follow from exposure to alcohol warnings in alcohol-related environments
such as pubs or bars. It has been noted that interaction with alcohol warnings in different
contexts needs further investigation (Thomson, Vandenberg, & Fitzgerald, 2012) and
research has shown that alcohol-related beliefs change across situational and social contexts
(Monk & Heim, 2013a, 2013b). This means that in vivo examinations of alcohol-related
cognitions may not replicate laboratory-based assessments (Monk & Heim, 2013a). Informed
by recent research examining the effectiveness of smoking warnings in real-life contexts
(Glock et al., 2013), future research may therefore be benefited by implementing the current
paradigms in real-life drinking contexts. Nevertheless, the findings from this research offer
important insights for the ongoing debate in relation to the most effective use of pictorial
health warnings on alcohol packaging. It can be regarded as an indication of the need to
carefully align the warning content with the goals of the initiative, and not simply aiming to
startle individuals into compliance.
17
References
Agostinelli, G., & Grube, J. W. (2002). Alcohol Counter-Advertising and the Media: A
Review of Recent Research. Alcohol Research and Health, 26, 12-21.
Blanchette, I. (2006). The effect of emotion on interpretation and logic in a conditional
reasoning task. Memory and Cognition, 34(5), 1112-1125. doi: 10.3758/BF03193257
Blanchette, I., & Richards, A. (2004). Reasoning about emotional and neutral materials.
Psychological Science, 15(11), 745-752. doi: 10.1111/j.0956-7976.2004.00751.x
Blanchette, I., & Richards, A. (2010). The influence of affect on higher level cognition: A
review of research on interpretation, judgement, decision making, and reasoning.
Cognition and Emotion, 24(4), 561-595. doi: 10.1080/02699930903132496
Blanchette, I., Richards, A., Melnyk, L., & Lavda, A. (2007). Reasoning about emotional
contents following shocking terrorist attacks: a tale of three cities. Journal of
Experimental Psychology: Applied, 13(1), 47-56. doi: 10.1037/1076-898X.13.1.47
Brandish, E., & Sheron, N. (2010). Drinking patterns and the risk of serious liver disease.
Expert Review of Gastroenterology and Hepatology, 4(3), 249-252. doi:
10.1586/egh.10.27
Cherpitel, C. J., Borgers, G., Giesbrecht, N., Hungerford, D., Peden, M., Poznyak, V., . . .
Stockwell, T. (2003). Alcohol and injuries: Emergency department studies in an
international perspective. Geneva: WHO.
Cummins, D. D., Lubart, T., Alksnis, O., & Rist, R. (1991). Conditional reasoning and
causation. Memory and Cognition, 19(3), 274-282.
European Commission. (2012). Tobacco packaging heath warning labels. Brussels:
http://ec.europa.eu/public_opinion/archi ves/quali/ql_5818_tobacco_en.pdf. alcohol
[Accessed October, 2013].
Faculty of Public Health UK. (2012). Health warnings needed on alcohol. Available
from:http://www.fph.org.uk/health_ warnings_needed_on_alcohol [Accessed October,
2013].
Ford, A., Moodie, C., & Hastings, G. (2012). The role of packaging for consumer products:
understanding the move towards 'plain' tobacco packaging. Addiction Research and
Theory, 20(4), 339-347. doi: 10.3109/16066359.2011.632700
Glock, S., & Krolak-Schwerdt, S. (2013). Changing outcome expectancies, drinking
intentions, and implicit attitudes toward alcohol: a comparison of positive expectancyrelated and health-related alcohol warning labels. Applied Psychology: Health and
Well- Being, 5, 332-347. doi: 10.1111/aphw.12013
Glock, S., Ritter, S. M., Engels, R., Dijksterhuis, A., van Baaren, R. B., & Müller, B. C. N.
(2013). “Smoking kills” vs.“Smoking makes restless”: Effectiveness of different
warning labels on smoking behaviour. Science, 2, 181-187. doi:
10.11648/j.pbs.20130205.12
Griggs, R. A., & Cox, J. R. (1982). The elusive thematic-materials effect in Wason's selection
task. British Journal of Psychology, 73(3), 407-420.
Hammond, D. (2011). Health warning messages on tobacco products: a review. Tobacco
Control, 20, 327-337. doi: 10.1136/tc.2010.037630
Hastings, G., & MacFadyen, L. (2002). The limitations of fear messages. Tobacco Control,
11, 73-75. doi: 10.1136/tc.11.1.73
Janis, I. L. (1967). Effects of fear arousal on attitude change: Recent developments in theory
and experimental research. In L. Berkowitz (Ed.), Advances in Experimental Social
Psychology (Vol. 1). New York: Academic Press.
Kessels, L. T. E. (2014). Neuroscientific evidence for defensive avoidance of fear appeals.
International Journal of Psychology, 49(2), 80-88. doi: 10.1002/ijop.12036
18
Lang, P. J., Bradley, M. M., & Cuthbert, B. N. (2005). International affective picture system
(IAPS): Instruction manual and affective ratings: Technical Report A-6, The Centre
for Research in Psychophysiology, University of Florida.
Leventhal, H. (1970). Findings and theory in the study of fear communications. Advances in
Experimental Social Psychology, 5, 119-186. doi: 10.1016/S0065-2601(08)60091-X
Manktelow, K. (2000). Reasoning and thinking. Hove: Psychology Press.
Monk, R. L., & Heim, D. (2013a). Environmental context effects on alcohol-related outcome
expectancies, efficacy and norms: A field study. Psychology of Addictive Behaviors,
27, 814-818. doi: doi: 10.1037/a0033948.
Monk, R. L., & Heim, D. (2013b). Panoramic projection; affording a wider view on
contextual influences on alcohol-related cognitions. Experimental and Clinical
Psychopharmacology, 21, 1-7. doi: 10.1037/a0030772
O'Brien, P. (2013). Australia's double standard on Thailand's alcohol warning labels. Drug
and Alcohol Review, 32, 5-10. doi: 10.1111/j.1465-3362.2012.00485.x
Ringold, D. (2002). Boomerang effects in response to public health interventions: some
unintended consequences in the alcoholic beverage market. Journal of Consumer
Policy, 25(1), 27-63. doi: 10.1023/A:1014588126336
Ruiter, R. A. C., Abraham, C., & Kok, G. (2001). Scary warnings and rational precautions: A
review of the psychology of fear appeals. Psychology and Health, 16(6), 613-630.
Stewart, D. W., & Martin, I. M. (1994). Intended and unintended consequences of warning
messages: a review and synthesis of empirical research. Journal of Public Policy and
Marketing, 13(1), 1-19.
Sweet, K. M., Willis, S. K., Ashida, S., & Westman, J. A. (2003). Use of fear-appeal
techniques in the design of tailored cancer risk communication messages: implications
for healthcare providers. Journal of Clinical Oncology, 21(17), 3375-3376. doi:
10.1200/JCO.2003.99.049
Thamarangsi, T., & Puangsuwan, A. (2010). Why Thailand should have the pictorial warning
label on alcoholic beverage packages? Thailand: Centre for Alcohol Studies and
International, Health Policy Program, Availible from: http://ihppthaigov.net/pu
blication/attachresearch/257/chapter251.pdf. [Accessed Oct 2013].
Thomson, L. M., Vandenberg, B., & Fitzgerald, J. L. (2012). An exploratory study of
drinkers views of health information and warning labels on alchohol containers. Drug
and Alcohol Review, 31(2), 240-247. doi: 10.1111/j.1465-3362.2011.00343.x
Van Duyn, M. A., & Pivonka, E. (2000). Overview of the health benefits of fruit and
vegetable consumption for the dietetics professional: selected literature. Journal of the
American Dietetic Association, 100(12), 1511-1512. doi: 10.1016/S00028223(00)00420-X
Wason, P. C. (1968). Reasoning about a rule. The Quarterly Journal of Experimental
Psychology, 20(3), 273-281. doi: 10.1080/14640746808400161
Wilkinson, C., & Room, R. (2009). Warnings on alcohol containers and advertisements:
International experience and evidence on effects. Drug and Alcohol Review, 28, 426435. doi: 10.1111/j.1465-3362.2009.00055.x.
Witte, K. (1998). A theoretically based evaluation of HIV/AIDS prevention campaigns along
the trans-Africa highway in Kenya. Journal of Health Communication, 3, 345-363.
doi: 10.1080/108107398127157
Witte, K., & Allen, M. (2000). A meta-analysis of fear appeals: implications for effective
public health campaigns. Health Education and Behavior, 27, 591-615. doi:
10.1177/109019810002700506
19
Figure 1
Self-Assessment Manikin used by participants to rate each image for valence and
arousal
20
Figure 2
Example format of the Wason selection task presented in Study 1.
21
Figure 3
Percentage of trials responded to with selection of p-and-not-q (logically correct) by
content type and emotional valence (Error bras: 95%CI).
22
Figure 4
Example of a visual conditional from Study 2
23
Figure 5
Interaction between problem type and content type on reasoning accuracy (Error bras:
95%CI).
24
Table 1
Percentage of trials responded to with p-and-not-q selections (accuracy) by content type
and emotional valence.
Content
Valence
Positive
Negative
Logically correct (p and not-q)
Alcohol
Non-alcohol
26.54 (31.96)
29.67 (31.26)
26.86 (32.72)
23.66 (32.66)
25
Total
28.11 (31.61)
25.26 (32.69)
Table 2
Image pairs presented accompanying “If p, then q” for each possible problem type
(inference), including the logically correct response for each type of problem.
Problem Type
Presentation Pair
MP
MT
DA
AC
Given p
Given not-q
Given not-p
Given q
Does q follow?
Does p follow?
Does q follow?
Does p follow?
26
Correct
Response
Y
N
M
M
Table 3
Participants’ reasoning accuracy (%) by problem type and warning content (N=58 for
all cells).
Problem
Type
MP
MT
DA
AC
Total
Negative-Alcohol
M
SD
85.09
23.29
75.78
29.20
16.21
24.70
17.59
24.30
48.66
25.37
Positive-Non-Alcohol
M
SD
74.83
20.71
71.38
30.00
22.76
18.90
19.66
24.99
47.16
23.65
27
M
79.96
73.58
19.48
18.62
47.91
Total
SD
22.00
29.60
21.80
24.65
24.51