Exploring the effectiveness of message framing and time context on
alcohol warning labels
Author: Xueyao Yin
Student number: 10842152 Thesis type: Master’s Thesis
School: Graduate School of Communication Science Program: Research Master’s
Supervisor: Prof. Dr. S.J.H.M. (Bas) van den Putte Completion date: 27th June, 2017
Abstract
There was limited research on alcohol warning messages regarding message phrasing,
showing inconsistent results. In order to explore message effectiveness in the context of
alcohol container warning labels, the current study compared the effects of messages that
adopt different framing styles and different time contexts. A 2 (message framing: gain vs.
loss) × 2 (time context: short vs. long term) between-subjects experiment was conducted
among 166 young adults worldwide. The main dependent variables in this study were
attitude towards reducing alcohol drinking and intention of reducing alcohol drinking.
The result showed different findings compared to previous research. On contrary to what
has been expected, loss-framed messages worked more effectively to increase the
intention of reducing alcohol drinking. Time context had no significant effect on attitude
or intention. The interaction between time context and message framing and the
interaction between time context and future orientation did not show any significant
effect either. In addition, attitude did not mediate the effect of alcohol warning messages
on intention. Additional analysis with group comparison suggested that the path
Exploring the effectiveness of message framing and time context on
alcohol warning labels
Heavy alcohol consumption is positively correlated with many health problems
among young adults, such as physical injury, risky sexual behaviors, mental health issues,
long-term brain functioning damage, and more. (Hermens et al., 2013; World Health
Organization [WHO], 2014). These problems can have life-changing effects to young
adults, which can further lead to detrimental social and cultural standing and position
(e.g., reduced economic productivity; Courtney & Polich, 2009). Despite the negative
consequences, excessive drinking is most popular among young adults, compared to
other age groups (Johnston, O'malley, Bachman, Schulenberg, & Miech, 2015). Thus it is
urgent to reduce alcohol consumption within this group.
Countries worldwide have implemented various strategies to reduce alcohol
consumption. In previous years, research findings revealed that taxation, restriction of
availability and advertisements are the most cost-effective options to reduce
alcohol-caused death and disabilities on population level (WHO, 2014). When targeting college
students specifically, brief motivational interventions have been widely adopted to reduce
heavy drinking, however, studies showed that this strategy only had small effects on
alcohol consumption related outcomes (Huh et al., 2015). Therefore, it is necessary to
explore new approaches to target this problem. Similar to alcohol consumption, tobacco
usage is also a prevalent health threatening behavior. Regarding smoking, warning labels
(e.g., text message stating “Smoking Kills”) on tobacco packages have been used as an
effective health promotion message in many regions (Noar et al., 2016). Alcohol warning
warning messages can reduce alcohol consumption as what tobacco warning labels did to
cigarette usage. The current study aims to examine whether using warning labels on
alcohol containers can be an effective approach to reduce alcohol consumption among
young adults.
In 1989, the United States became the first country that requires mandatory
warning labels on alcohol containers (Greenfield, Graves, & Kaskutas, 1999). By 2012,
at least 31 countries required health and safety warning labels on alcohol containers,
including France, Belarus, South Africa, and Brazil (WHO, 2014) Even though
Europeans on average consume more alcohol than any other region (WHO, 2017), most
of European countries (e.g., The Netherlands and Germany) do not require warning labels
on alcohol containers. Previous research found mixed results of the effects of alcohol
warning labels on drinking behaviors. Many of the studies (Wilkinson & Room, 2009)
indeed found increased awareness of the negative effect of alcohol, however, alcohol
warning labels has little effect on behavior or even resulted in boomerang effects (Bell,
Zizzo, & Racine, 2015). So far, alcohol warning label did not raise enough attention in
the academic world and this has not been much improved since first implementation.
Thus, the current study compared different message types to test how to increase the
effectiveness of alcohol warning labels, which might further serve as evidence to
persuade more countries to require mandatory alcohol warning labels.
In order to search for more effective alcohol warning labels, two comparisons
were made in the current study. The first comparison was between different message
frames. According to framing theory (Rothman & Salovey, 1997), message receivers
Gain-framed messages emphasize the benefits of behaving in the recommended manner; while
loss-framed messages emphasize the disadvantages of not behaving in the recommended
manner. Meta-analysis on previous studies (Gallagher & Updegraff, 2012; O'Keefe &
Jensen, 2009) found that gain-framed messages, compared to loss-framed messages, had
bigger effects on encouraging prevention behaviors, but the effect size depends on the
behavior type. Until now, the effectiveness of gain- and loss- framed messages have not
been tested on alcohol warning labels. The current study made up this gap.
Apart from message framing, the current study also explored the effect of time
context. Previous studies showed that people discount future outcomes, meaning that
individuals have the tendency to consider a reward as more attractive if it comes sooner
and the tendency to consider punishment is more attractive if it comes later (Green &
Myerson, 2004). When applied to alcohol intervention, this process of temporal
discounting can influence the effect of a health message, depending on whether the
message states long-term consequences (discounted future outcomes) or short-term
consequences. Previous research did not show consistent findings on the effect of time
context on alcohol drinking behavior or related problems (Bernstein, Wood, & Erickson,
2016; Churchill, Pavey, Jessop, & Sparks, 2016; Gerend & Cullen, 2008). For example,
Gerend and Cullen (2008) found a main effect of time context, such that short-term
messages worked more effectively compared to long-term messages, while another study
(Bernstein, Wood, & Erickson, 2016) found a marginally significant effect of time
context, in which long-term messages leaded to more desired outcomes.
As mentioned above, the effect of message framing depends on specific behavior,
message has more effectiveness than loss-framed message in changing recipients’ alcohol
related outcome variables. The current study provided empirical evidence in testing if
framing theory can be applied to alcohol warning labels by comparing labels that contain
gain- and loss-framed messages. Regarding time context, there was limited research on
alcohol consumption and the results were mixed. Thus by comparing alcohol warning
labels that contain short- or long-term messages, the current study added new insights
into the discussion and helped testing the application of time context variation in alcohol
warning label settings. The two comparisons made in this study provided great value for
future alcohol warning message design because it made up the research gap of message
framing in alcohol warning labels, and provided new information to the discussion of the
effect of time context.
Theoretical Framework Message framing
Persuasive messages often present the consequences of a certain behavior in
order to motivate individuals to adopt desired behaviors or abandon undesired ones. The
same consequence can be framed in different ways by focusing on the advantages of
compliance (i.e., gain-framed messages), or by focusing on the disadvantages of
non-compliance (i.e., loss-framed messages). For example, the same underlying content can be stated as “Pursue education for a better job opportunity” or as “Dropping-out of school
makes you perform worse in the job market”. There were discussions about which type of
message is more persuasive. In health message domain, framing theory (Rothman &
Salovey, 1997) proposed that the effectiveness of message framing depends on the type
of intended behavior: for health-affirming behaviors (e.g., quit smoking to avoid lung
for illness-detecting behaviors (e.g. going to the hospital for breast cancer check),
loss-framed messages are more persuasive. When connected to the current research topic,
alcohol consumption falls into the category of health-affirming behavior because
individuals can increase their chances of staying healthy by reducing the consumption of
alcohol, Meta-analyses on framing (Gallagher & Updegraff, 2012; O'Keefe & Jensen,
2009) found that overall message framing showed significant effects on behavior change,
but they also suggested that the effectiveness of message framing depends on the specific
behavior.
O'Keefe and Jensen (2009) investigated the effect of message framing on disease
prevention behavior (e.g., alcohol related behaviors) by analyzing the general effect on
attitude, post-communication agreement, behavioral intention, and behavior, all together.
They found that gain-framed messages in general had a small advantage (r = .03) over
loss-framed messages. Focusing on specific behavioral categories, the only large and
significant effect of message framing was on dental hygiene behaviors (r = .15). Thus,
the result was misleading in a way that the advantage of gain-framed messages was
mainly in one specific behavior. Another meta-analysis (Gallagher & Updegraff, 2012) of
message framing found that message framing did not have significant effects on attitude
or intention. When examining specific behaviors, the gain-framed message advantage
was found in encouraging three prevention behaviors (i.e., physical activity, skin cancer,
and smoking; r = .01). This study did not include alcohol behavior into its analysis
neither.
Even though in general gain-framed messages showed higher effectiveness
framing depends on the specific behavior type. Thus it is necessary to conduct a study
that focuses on alcohol drinking behavior specifically, which is not included in previous
meta-analyses. The lack of alcohol related studies in previous meta-analysis is probably
because there was only limited research investigating message framing under the scope of
alcohol consumption. Kingsbury, Gibbons and Gerrard (2015) investigated the
effectiveness of gain- versus loss- framed message of health consequences on alcohol
consumption (e.g., maintaining a healthy weight) and found that gain-framed health
consequences were associated with lower heavy drinking intentions versus loss-framed
messages. Contrarily to this study and the previous meta-analyses, Yu, Ahern,
Connolly-Ahern, and Shen (2010) found that loss-framed messages were more persuasive
compared to gain-framed messages in changing college students’ intention to know more,
perception of severity, as well as their perceived fear about fetal alcohol spectrum
disorder. However, the dependent variables in this study were not necessarily linked to
behavior change. From this perspective, it is improper to fully apply the result of this
study into the current research.
There is limited evidence regarding the effect of message framing on alcohol
consumption. Previous meta-analysis showed that in general gain-framed messages work
better than loss-framed messages, even though the effect size vary per behavior type.
There was one study (Kingsbury, Gibbons, & Gerrard, 2015) about alcohol consumption
and it was in line with framing theory. Along with the previous findings, the following
Hypothesis 1: Gain-framed alcohol warning labels will be more effective in changing alcohol related outcomes such as attitude (H1.1) and intention (H1.2), compared to loss-framed alcohol warning labels.
Time context
Time context represents the occurrence time of the consequences that imbedded
in the alcohol warning labels. Different time context might influence the effectiveness of
alcohol warning labels because of temporal discounting, which is a type of psychological
distance proposed in construal-level theory (CLT; Trope & Liberman, 2010). Temporal
discounting formed the tendency that make individuals overvalue a sooner outcome
compared to a later outcome (Green, Myerson, Lichtman, Rosen, & Fry, 1996;Trope &
Liberman, 2010). CLT proposed that people build up an abstract mental construal of
distal objects to plan for the distant future. Future events (e.g., behavior outcome) with a
longer abstract distance from individuals cause less concrete thoughts and are less valued.
For example, if a reward of the same economic value is provided immediately or delayed,
the delayed reward has less present subjective value compared to the immediate reward.
In a classic example, participants have preference for an immediate gain of $100,
compared to a gain of $120 one month later (Green & Myerson, 2004). When connected
to health intervention messages, temporal discounting may influence the effectiveness of
the messages, depending on the content presenting long-term outcomes, or short-term
outcomes.
In practice, a commonly used warning label on alcohol containers includes a
general consequence, regardless of time context: “Excessive consumption of alcohol is harmful to your health” (WHO, 2014). Even though it is not often explicitly stated,
alcohol consumption among young adults can cause both short-term and long-term
consequences (Hermens et al., 2013; WHO, 2014). When applying the idea of temporal
discounting on alcohol interventions, messages stating long-term consequences will have
less effects on recipients compared to short-term consequences because the long-term
future outcome is discounted and less valued. Thus, changing the time context of the
message will influence related outcome variables. For example, “Drinking increases the possibility of getting cancer” might be treated less valued than “Drinking increases the
possibility of a car accident” and has less effect on recipients. Up to this time, no research
has investigated the effect of temporal discounting on alcohol warning labels. Thus the
second hypothesis has been made based on construal-level theory:
Hypothesis 2.1: Alcohol warning labels that include short-term consequences will be more effective in changing alcohol related outcomes, such as attitude (H2.1.1), and intention (H2.1.2), compared to alcohol warning labels that includes long-term
consequences.
Even though there was no research conducted in the context of alcohol warning
labels, the previous research of temporal discounting on other health messages might
enlighten the current research. One study showed that the effect of temporal discounting
might be influenced by the personalities of individuals, such as future orientation (Kees,
2010). Future orientation concretes the inclination of individuals to place more emphasis
on future events versus immediate events. Kees (2010) found an interaction effect of subjects’ future orientation and temporal discounting: for participants who are low in
regards to future orientation (i.e. placing more emphasis on immediate events), a
in changing participants’ concern and risk perception about healthy food choice. The
difference between long- or short-term consequence messages was not observed among
high future orientation (i.e., placing more emphasis on future events) participants. Low
future-oriented individuals may have a steeper discounting rate for a long-term
consequence because they do not take future event into account as much. For high
future-oriented individuals, the discounted rate might be much milder, which makes a short-term
and a long-term consequence remain at a similar level of importance.
Hypothesis 2.2: The effects of time context on alcohol related outcome variables are moderated by the future orientation of participants, such that for high future
orientated individuals, the effect of long-term consequence messages on outcome variables, such as attitude (H2.2.1) and intention (H2.2.2), will be stronger. Message framing and time context
Previous research on time context showed that the rate of the discounting of
choices depends on whether the outcome is positive or negative: the value of gains
discounts faster compared to the value of losses. The asymmetry of discounting rate has
been observed in hypothetical monetary and health choices (Gerend & Cullen, 2008). For
example, Hardisty and Weber (2009) found that when the participants were asked to
evaluate monetary options, the absolute value of equivalent future outcomes differ
between an immediate gain of $250 and an immediate loss of $250. An immediate gain
of $250 was as attractive as a gain of $337.50 after one year; while an immediate loss of
$250 was as much as a loss of $265 after one year. Further analysis showed that the
discounting rate of gain or loss was significantly different in monetary choices. The same
temporal discounting rate differs for gain and for loss within different subjects. They also
found that the discounting rate of gain in one domain correlated with discounting rates of
gain in other domains (i.e., monetary choices and environmental choices), indicating a
universal discounting rate across subjects. Thus it is possible to apply the previous
research of temporal discounting in other subjects to the current research interests.
There is limited research that was conducted to explore the interaction effect of
message framing and time context on alcohol related outcomes and they yielded
inconsistent results. Gerend and Cullen (2008) tested the effect of message framing (gain
versus loss) and time context on alcohol behavior among college students. The result
showed that gain-framed messages were more effective than loss-framed messages, but
only when participants received short-term consequence message. Inconsistent findings
were found later by Bernstein, Wood, and Erickson (2016). In a field experiment among
college students, no interaction effect between message framing and time context was
found. Churchill, Pavey, Jessop, and Sparks (2016) also tested the interaction effect on
drinking behaviors and did not find a significant two-way interaction between gain- and
loss-framed messages with time context. The inconsistent findings call for more
exploration in this subject. Meanwhile, all of these studies suffer from one problem that
they used the same consequences in two time context conditions but explicitly indicated
if the consequence is in long- or in short-term. For example, driving accidents were used
as a consequence in both short- and long-term conditions. Even though the manipulation
check showed that participants can correctly indicate the messages focused on the long-
or short-term consequences, this might occur because the time context was explicitly
which leaves a big question about whether they indeed manipulated participants to
believe the commonly encountered short-term consequence (e.g., driving accident) can be
a long-term drinking consequence. This defect makes the results less trustworthy.
Depending on the effectiveness of warning message and believability of the message,
participants will react differently to the experimental exposure. When believability is not
taken into consideration, the result is rather random instead of valid. For example, in a
long-term condition, driving accident showed great effectiveness but this advantage
actually comes from driving accident itself (as a short-term consequence), instead of the
manipulation message that defined driving accident as a long-term consequence. This
kind of scenario can explain the insignificant effect of time context found by Bernstein,
Wood, and Erickson (2016). Apart from these three studies, there was no research testing
the interaction effect on alcohol warning labels, which was investigated in the current
study.
Due to the lack of valid research into message framing and temporal discounting
on alcohol related outcomes, the following hypothesis has been formulated based on
research that showed asymmetric discounting rate regarding monetary and health
consequences:
Hypothesis 3: Message framing (gain- vs loss-framed message) interacts with time context (long-term or short-term consequences) in influencing alcohol warning label effectiveness on attitude (H3.1) and intention (H3.2). Gain-framed messages that include short-term consequences will lead to more desired changes.
The effect pathways: attitude as a mediator
When designing health promotion messages, it is necessary to explore the effect
mechanism to identify the key factors in influencing individual’s health behavior. Based
on the theory of planned behavior (TPB; Ajzen, 1991), an individual’s attitude can lead to
intention change, which further leads to behavior change. This model was widely tested
in health behavior settings (Albarracin, Johnson, Fishbein, & Muellerleile, 2001;
Armitage & Conner, 2001). For example, Godin and Kok (1996) conducted a
meta-analysis to test the prediction power of the theory of planned behavior on health related
behaviors, including smoking, screening behavior, eating, etc. The results showed that the
model explained an average of 41 percent of variance of the health behaviors (R2 = .41)
and attitude indeed predicted behavior through intention change.
However, the findings of Gallagher and Updegraff (2012) challenged the
commonly accepted mediators in TPB model. Inconsistent to TPB, Gallagher and
Updegraff (2012) found that message framing do not influence attitude and intention,
instead it has a direct effect on behavior. They proposed that gain-framed messages may
convey information that changes outcome expectations, which might further change
behavior more directly than attitude and intention. However according to TPB, if
message framing indeed influences the outcome expectations, it will eventually lead to
attitude change, intention change and then behavior change. Apart from the previous
insignificant findings, another study (Kingsbury, Gibbons, & Gerrar, 2015) found
message framing indeed has effect on drinking intention, indicating a direct effect of
message framing on intention. The inconsistency of theory and research findings leaves
important to test possible models that include message effect on attitude, as well as on
intention to test how does message influence attitude and intention, and to test the
mediation role of attitude.
Both intention and attitude are two elements in the TPB model, in which attitude
predicts intention together with perceived behavioral control and injunctive norms. In
order to test the causality between attitude and intention, all the other predictors of
intention were included in the conceptual model (Figure 1) to control their effects on
intention. Apart from perceived behavioral control and injunctive norm, descriptive norm
was taken into account because it was shown to be a valid predictor of intention (Rivis &
Sheeran, 2003). In a meta-analysis (Rivis & Sheeran, 2003), the researchers found that
descriptive norm increases five percent more of the explained variance of intention, after
including attitude, perceived behavioral control and injunctive norm. Descriptive norm
refers to the attitudes and behaviors of significant others, which provides information that
individuals might use in making decisions (Rivis & Sheeran, 2003). It is different from
injunctive norm (Hagger & Chatzisarantis, 2005), which refers to the approval of
important others about the conduction of a certain behavior. Thus after including
descriptive norm in the analysis, perceived behavioral control, descriptive norm and
injunctive norm were all taken as control variables.
Because the previous findings were mixed, the following research question was
made to be answered in the current study.
Research question: Does attitude mediate the effect of message framing, time context, future orientation and their interaction terms on intention?
The conceptual model and hypothesis are presented in Figure 1: main effects of
message framing are indicated by H1.1 and H1.2; main effects of time context are
indicated by H2.1.1 and H2.1.2; interaction effects of message framing and time context
are indicated by H3.1 and H3.2; while interaction effects of time context and future
orientation are indicated by H2.2.1 and H2.2.2. Attitude is located in the model as a
mediator between message type and intention.
Figure 1. Conceptual model and hypothesis
Methods Participants and Design
A total of 194 participants were recruited from personal network, in which 7 were
excluded from the data analysis due to age restriction (i.e., only young adults between the
age of eighteen to forty were kept in the data analysis); and further 21 participants were
excluded because they indicated zero alcohol consumption. In total, the data of 166
participants (M age = 25.70 years, SD = 3.22; 60.24% female) were included in the
analysis. Most of the participants are highly educated (e.g., 91.2% of participants are Message Framing gain/loss Time Context short/long-term Attitude Intention Future Orientation H2.2.1 H1.1 H3.1 H2.1.1 H1.2 H2.1.2 H2.2.2 H3.2
currently following or have completed a Bachelor’s Degree or a higher degree).
Regarding nationality, there were 47 different nationalities held by the participants,
among which the major nationalities were Dutch (19.3%) and Chinese (8.4%).
A 2 (time context: long-term versus short-term messages) x 2 (message framing:
gain- or loss- framed messages) between subjects design was adopted in the current
study. Participants were randomly assigned into four conditions and were exposed to different warning labels. At the end, participants’ attitude, intention, perceived behavioral
control, subjective norm, descriptive norm, and future orientation were measured.
Procedure
Participants received a link to an online survey that was built in Qualtrics. They
were firstly presented with an informed consent page. In the first section of the survey,
participants answered demographic questions, including age, gender, education level,
nationality, and alcohol consumption amount. They were also asked if they prefer beer or
wine, or if they do not drink alcohol. If one type of alcohol is chosen, the following
alcohol bottles will match the preferred drinking type. If they identified as non-alcohol
drinkers, they were thanked and dismissed from the questionnaire. The participants were
then randomly assigned to four experimental conditions and were exposed to
corresponding messages on alcohol warning labels either on beer bottles or on wine
bottles (for examples of experimental material, see Figure 2 and Figure 3).
In each condition, four alcohol bottles with warning messages were presented.
After each bottle, personal relevance was measured. At the end of the exposure,
participants were asked to rank the four messages about how proper they are to be shown
four messages. In the next section, the dependent variables were measured, including
attitude and intention. Other variables, such as perceived behavioral control, injunctive
norm, descriptive norm, and future orientation were measured for testing the conceptual
model. At the end of the questionnaire, four alcohol warning messages were presented
again. Two manipulation questions were asked after each of the alcohol messages, to test
whether the participant identified the message as short/long framed, or as gain/loss
focused.
Figure 2. Example beer bottle. Figure 3. Example wine bottle.
Experimental Materials
Elicitation. Sixteen students (M age = 24.50, SD = 1.59; 68.8% women) were asked to fill in a self-report form (see Appendix A) about the alcohol drinking related
short-term or long-term consequences. Consequences mentioned in these self-reports, as
(Churchill, Pavey, Jessop, & Sparks, 2016; Gerend & Cullen, 2008) and a study that
developed cancer warning statements for alcoholic beverages (Pettigrew et al., 2014),
were selected and used in 40 messages for the pre-test (see Appendix B).
Pretest. A total of 36 students (M age = 24.50, SD = 1.67; 66.7% women) were recruited for pretesting experimental materials. Participants in the pretest were asked to
fill in the characteristics of the 20 messages (either 20 gain-framed messages; either 20
loss-framed messages) regarding their valence, gain- or loss-frame, long-term or
short-term consequence focus, and believability (see Appendix C). The order of pretested
messages were randomized. At the end, four pairs of health consequences were chosen
based on a match of their absolute mean grade. Each pair contained one short-term
consequence and one long-term consequence. These four pairs fulfilled the requirement
of manipulation, and they had a non-significant difference on valence (both p > .287) and
believability (both p > .160) (see Appendix D). Based on these eight consequences,
sixteen messages (see Table 1) were framed as either gain or loss.
Table 1 Alcohol warning labels in the questionnaire.
Gain-frame Loss- frame
Short-term - Drinking less alcohol decreases your risk of getting into driving accidents
- Drinking less alcohol decreases your risk of getting alcohol poisoning
- Drinking less alcohol decreases your risk of getting blackouts
- Drinking less alcohol decreases your risk of getting impaired judgment
- Drinking more alcohol increases your risk of getting into driving accidents - Drinking more alcohol increases your risk of getting alcohol poisoning
- Drinking more alcohol increases your risk of getting blackouts
- Drinking more alcohol increases your risk of getting impaired judgment Long-term - Drinking less alcohol decreases your risk
of getting organ problems
- Drinking less alcohol decreases your risk of getting brain damage
- Drinking less alcohol decreases your risk of heart diseases
- Drinking more alcohol increases your risk of getting organ problems
- Drinking more alcohol increases your risk of getting brain damage
- Drinking more alcohol increases your risk of heart disease
- Drinking less alcohol decreases your risk of high blood pressure
- Drinking more alcohol increases your risk of high blood pressure
Measurements
Attitude. Attitude was measured by six questions (adjusted from Keer, van den Putte & Neijens, 2012): “Do you think reducing alcohol drinking is”: 1 = unpleasant to 7
= pleasant, 1 = valuable to 7 = worthless (reverse coded); 1 = nasty to 7 = nice; 1 =
useless to 7 = useful; 1 = beneficial to 7 = harmful (reverse coded); 1 = enjoyable to 7 = not enjoyable (reverse coded). A factor analysis showed that they load on one factor, and the reliability test yielded satisfactory result (M = 3.37, SD = .81, α = .92).
Intention. Intention was measured by an agreement level on three statements (based on Mollen, Engelen, Kessels, & van den Putte, 2016): “I am willing to reduce alcohol consumption”, “I intend to reduce alcohol consumption”, and “I will try to reduce
alcohol consumption”. Answers were made on a scale from 1 = strongly disagree to 7 =
strongly agree. Factor analysis showed that they load on one factor, and reliability test yielded satisfactory result (M = 3.57, SD = 1.58, α = .82).
Perceived behavioral control. Perceived behavioral control was measured by four questions: “I am confident that I can reduce my alcohol consumption”; “Reducing
my alcohol consumption is completely up to me”; “For me, reducing alcohol
consumption is easy” and “It is impossible for me to reduce my alcohol consumption”
(reverse coded). Answers were made on a scale from 1 = strongly disagree to 7 =
strongly agree. A factor analysis showed that two factors have an eigenvalue bigger than 1 and the Cronbach’s Alpha is low (α = .55). As suggested by the result, the reversed
from the scale and it yielded one factor and a good reliability (M = 5.72, SD = 1.05, α
= .75).
Social norms. Social norms were measured in two aspects: descriptive norm and injunctive norm (Hendriks, de Bruijn, Meehan, & van den Putte, 2016). The injunctive
norm was measured by three questions (e.g., “Most of the people who are important to me would accept it, if I drink less alcohol”) with answers ranging from 1 = strongly
disagree to 7 = strongly agree. Factor analysis showed that they load on one factor, and reliability test yielded satisfactory result (M = 4.28, SD = 1.25, α = .71). Descriptive norm was measured by one question, “Among all the people who are important to you, how
many of them drink regularly?”, with answers ranging from 1 = about 0%/none of them
to 7 = about 100%/all of them.
Future orientation. Future orientation was measured by five five-point Likert Scale questions (Strathman, Gleicher, Boninger, & Edwards, 1994). Example statements are: “I spend very little time thinking about how things might be in the future” and “I
would rather save my money for a rainy day than spend it right away on something fun”.
Answers are ranged from 1 = definitely false to 7 = definitely true. Factor analysis
showed that two factors have an eigenvalue bigger than 1 and the Cronbach’s Alpha is low (α = .60). As suggested by the result, one item (i.e., “I would rather save my money
for a rainy day than spend it right away on something fun”) was deleted from the scale
and it yielded one factor and good reliability result (M = 3.43, SD = .84, α = .71).
Analytical plan
Structural equation modelling (SEM) was used to test the hypothesis and to
answer the research question. Specifically, a path model was tested in IBM- SPSS Amos
21. The path model was chosen because it provides an effective way to model various
multiple regressions that contain direct effects, indirect effects and moderation effects
(Lei & Wu, 2007). As illustrated in Figure 1, the exogenous variables included the
message framing, time context, their interaction; as well as future orientation, and the
interaction of future orientation and time context. These variables were expected to have
an influence on attitude, which further influences intention. Meanwhile, intention was
influenced by perceived behavioral control, injunctive norm and descriptive norm. Before
testing the SEM model, all the variables were standardized in order to prevent
multicollinearity. After standardization, interaction term (i.e., time context × message
framing; time context × future orientation) were made and standardized for SEM
analysis.
Several control variables (i.e., personal relevance, and alcohol consumption) were
tested to check if they have association with attitude and intention. For example, a t-test
was conducted to test if participants in the beer or wine condition held a different level of
attitude towards reducing alcohol consumption. If this is the case, then alcohol type
would be included in the analysis by adding a path from alcohol type to attitude.
After constructing the SEM model, the first step was to achieve an acceptable
model fit. AMOS software reports several types of goodness-of-fit indices. Normally
higher values of CFI and lower values of RMSEA represent good model fit (Lei & Wu,
smaller than .95 and RMSEA should not be bigger than .06. After achieving a good
model fit, specific parameters were examined to test the hypothesis. If the conceptual
model and hypothesis (Figure 1) were true, the corresponding parameters in the SEM
model should all be significant. Besides, bootstrapping with a 95% confidence interval
was adopted to estimate the direct effect, indirect effect and total effect of message
characteristics (e.g., message framing and time context) on intention. If attitude is a
mediator as in Hypothesis 4, the indirect effect will be significant.
Results Randomization
A one-way ANOVA was conducted to test whether participants were randomly
assigned over experimental conditions. The result showed that there was no difference
between the four conditions on age (p = .458). Chi-square tests were conducted to check
the distribution of gender and education level. The result showed that gender (p = .313)
and education (p = .196) were not significantly different across conditions.
Manipulation check
A t-test was conducted with the short- versus long-term condition as the
independent variable. The dependent variable was the manipulation question “Do you
think the alcohol warning message on this bottle focuses on a short-term consequence or
a long-term consequence of alcohol drinking?”. The answer ranged from 1 = short-term
to 7 = long-term. Because each participant was exposed to four consequences and was
therefore asked four times this question, the average was taken as an overall
representative. The four experimental conditions were recoded into two conditions
expected, there was a significant different in the scores in the short-term conditions (M =
2.83, SD = 1.27) and long-term conditions (M = 5.50, SD = 1.42); t (164) = 12.80, p
< .001.
Two t-tests were conducted to check if the message framing was successfully
manipulated by taking message frame conditions as the independent variable. The four
experimental conditions were recoded into two conditions (gain-framed vs. loss-framed)
by combining short-term conditions and long-term conditions. The dependent variable
was the manipulation question “This alcohol warning label focuses on 1 = the advantage
of drinking less alcohol or 0 = the disadvantage of drinking more alcohol.” Because each participant was exposed to four consequences and was therefore asked four times this
question, the average was taken as an overall representative. As expected, there was a
significant difference in the scores in the gain condition (M = .62, SD = .40) and loss
condition (M =.10, SD = .24); t (136.31) = -10.05, p < .001.
Control Variables
Personal Relevance. Pearson’s correlation analysis showed that personal relevance has a significant correlation with attitude (r = .16, p <. 05) and intention (r
= .31, p < .001), thus it was included as a control variable in the analysis.
Alcohol Consumption. Pearson’s correlation analysis showed that alcohol
consumption was not significantly correlated with attitude (r = -.01, p = .919) or intention
(r = .05, p =. 500), thus it was not included as a control variable.
Alcohol Type. Participants in the same condition were either exposed to beer bottles or wine bottles based on their preference. In order to dispel this concern, two
intentions regarding reducing alcohol consumption. According to the results, the mean
score of attitude (p = .726) and intention (p = .818) did not significantly differ between
alcohol container types. Thus the alcohol type was not taken into consideration in the
further analysis.
Main Analyses
The descriptive statistics (Table 2) showed an overview of the measured
variables. In general, the mean scores of attitude (M = 3.37, SD = .81) and intention (M =
3.57, SD = 1.58) were on a median level compared to the measurement range. The
average perceived behavioral control (M = 5.72, SD = 1.05) was high compared to the
maximum range of 7, indicating participants believe they have good control of decreasing
alcohol consumption if they intend to. Personal relevance scored in the lower half of the
range (M = 2.94, SD = 1.47), which means that participants perceive the consequences
mentioned on the alcohol warning labels as not very relevant to themselves. Injunctive
norm (M = 4.28, SD = 1.25), descriptive (M = 4.36, SD = 1.54) norm and future
orientation (M = 3.43, SD = .84) all have mean scores that were on a medium level.
Table 2 Descriptive statistics.
Mean SD Minimum Maximum
attitude 3.37 .81 1 5
intention 3.57 1.58 1 7
perceived behavioral control 5.72 1.05 1 7 injunctive norm 4.28 1.25 1 7 descriptive norm 4.36 1.54 1 7 personal relevance 2.94 1.47 1 7 future orientation 3.43 .84 1.25 5
While evaluating the SEM model, the Variance Inflation Factor (VIF) statistic
to 1, meaning there was no multicollinearity. Besides, because all the questions were
forced-answer questions, there was no missing value in the dataset, which means the
standardization was not influenced by the listwise deletion in AMOS. The zero-order
correlations between variables were presented in Table 3.
The AMOS output indicated that the data violated the criteria of multivariate
normality, the multivariate kurtosis is 8.146, which is over the critical value of 6. AMOS
spotted several outliers in the data, which have Mahalanobis distance with p values
greater than .05. After examining the data, there was no legitimate reason to drop these
cases, thus they were kept in the data analysis.
Table 3 Zero-order correlations between variables.
1 2 3 4 5 6 7 8 9 10 11 1.attitude — 2.intention .47** — 3.perceived behavioral control .36** 0.07 — 4.injunctive norm .27** .43** .14 — 5.descriptive norm 0.01 -0.12 -.07 -.10 — 6.personal relevance .16* .31** -.07 .25** .23** — 7.future orientation 0.04 -.16* .15 -.05 -.08 -.18* — 8.time context -.02 .09 .01 .07 -.12 -.06 .04 — 9.message framing -0.06 -.13 .02 -.01 .02 .07 -.03 .02 — 10. interaction of 8 and 9 -0.05 -.08 .13 .02 -.03 -.01 .04 -.00 -.00 — 11.interaction of 7 and 8 -0.05 -.01 -.08 -.08 -.09 .02 .08 -.01 .04 -.03 — Note 1. * p < .05; ** p < .01.
Note 2. Message framing: gain frame was coded as 1, loss frame was coded as 0; time context: short-term was coded as 1, long-term was coded as 0.
The SEM model was constructed as showed in Figure 4. All the exogenous
variables were allowed to correlate freely. The model did not yield a satisfactory model fit (χ²(24) = 38.51, p < .05, χ2/df =.62, CFI = .88, RMSEA = .06, 90% CI [.02, .10]). The
control, and between attitude and injunctive norm. This suggestion makes sense because
these three variables are all predictors of intention according to TPB model. Thus the
covariance was added between the error terms of attitude and the other two variables.
Besides, because descriptive norm was also suggested as an additional predictor of
intention, the error term of descriptive norm was also allowed to covariate with the error
term of attitude. After this step, the model yielded a satisfactory model fit: χ²(21) = 7.93,
p = .995, χ2/df = .38, CFI = 1.00, RMSEA = .00, 90% CI [.00, .00].
The estimates of the SEM model showed that personal relevance was the only
significant indicator of attitude (b* = .18, p < .05). Thus Hypothesis 1.1, Hypothesis
2.1.1, Hypothesis 2.2.1 and Hypothesis 3.1 were all rejected. The results of structural
equation modelling was presented in Figure 4.
Regarding intention, gain- and loss-framed condition variable showed a
marginally significant negative effect on intention (b* = -.11, p = .057), indicating
gain-framed message (coded as 1) decreased intention of reducing alcohol consumption
compared to loss-framed message, thus Hypothesis 1.2 is rejected. All the other
experimental conditions or interaction terms did not yield significant result, thus
Hypothesis 2.1.2, Hypothesis 2.2.2 and Hypothesis 3.2 were all rejected.
Bootstrap with Bias-corrected percentile method showed that there was no
significant indirect effect that is mediated by attitude (all p > .18), thus the research
question was answered: attitude does not mediate the effect of message type on drinking
intention.
Injunctive norm (b* = .26, p < .001), attitude (b* = .39, p < .001), descriptive
(b* = -.13, p < .05) all significantly predicted intention. The direction of effect is along
with what has been expected for injunctive norm (i.e., the more important others support
reducing alcohol consumption, the higher the intention of reducing alcohol consumption),
for descriptive norm (i.e., the more people drinking in the surroundings, the less intention
to reduce drinking) and for attitude (i.e., higher attitude towards reducing alcohol
consumption predicts higher intention towards reducing alcohol consumption).
Figure 4. Estimations in structural equation model.
*The numbers on arrow represents standardized regression estimate. * p < .05; ** p < .01; ***p < .001; † 0.1 < p < .05. .39*** Message framing gain/loss short/long-term consequences Attitude Intention Interaction of long/short term consequence and future orientation Perceived Behavioral Control Injunctive Norm Interaction of gain/loss framing and
long/short term consequence Descriptive norm Personal Relevance Future Orientation -.11† -.08 -.07 -.05 -.03 .08 -.06 .05 . 03 -.13* -.15** .26*** -.07 .20** .18**
Additional Analyses
There are some other variables which might influence the conceptual model. For
example, Kingsbury, Gibbons and Gerrard (2015) suggested that a previous level of
drinking can moderate the effect of message type on drinking related outcome variables.
An additional analysis was conducted by using current drinking behavior as grouping
variable in AMOS. With different current consumption levels, participants might react
differently to the alcohol warning label. For example, light drinkers might believe there is
no reason to drink less because they are already in a low risk group. Meanwhile, people
who drink more alcohol might be influenced by the message in a greater extent because
the message make them aware of the health problems. After a median split (median of
behavior was 5 glasses of alcohol per typical week), the current drinking amount of
participants was used to group participants.
The model yielded a satisfactory model fit: χ²(42) = 43.79, p = .395, χ2/df = 1.04,
CFI = .99, RMSEA = .02, 90% CI [.00, .06], which means the model could be applied for
both groups. The result (Table 4) showed that the lower half (group 1) and higher half
participants (group 2) indeed react differently to the alcohol warning label messages. For
the group 1, gain-framed messages leads to less intention of reducing alcohol (b*= -.15, p
< .05), which has the same direction as showed in the main analysis without grouping
participants. However, this effect is not significant for the group 2. For group 2, regarding
time context, messages with short-term consequences showed advantages in changing participant’s intention of reducing alcohol (b*= .21, p < .05): the short-term condition has
higher intention to reduce alcohol consumption, which support hypothesis 2.1.2. The
(b*= -.22, p < .01), indicating gain-framed (coded as 1) and short-term (coded as 1)
message combination or loss-framed and long-term message combination leads to less
intention of reducing alcohol consumption. Considering the advantage of short-term
messages, short-term and loss-frame combination should be preferred because they lead
to more intention of reducing alcohol drinking. Besides, for both groups, attitude did not
mediate any indirect effect between message type and intention, which was along with
the previous main analysis.
Table 4. Estimations in SEM model, grouping by current alcohol consumption.
group 1 (lower half) N = 93 group 2 (higher half) N = 73
Path Estimate S.E. Estimate S.E.
attitude <--- message framing (gain) -.00 .10 -.10 .10 attitude <--- time context (short) -.00 .10 .02 .10 attitude <--- interaction of message framing and time context -.10 .10 -.03 .10 attitude <--- interaction of future orientation and time context -.08 .11 -.07 .11 attitude <--- future orientation .19† .11 -.05† .11 attitude <--- personal relevance .19† .11 .27* .12 intention <--- message framing (gain) -.15* .08 -.07 .09 intention <--- time context (short) -.02 .08 .21* .09 intention <--- interaction of message framing and time context .05 .08 -.22** .09 intention <--- interaction of future orientation and time context .13 .08 .10 .09 intention <--- future orientation -.09 .08 -.21* .09
intention <--- attitude .43*** .08 .34** .11
intention <--- injunctive norm .31*** .08 .23* .10 intention <--- descriptive norm -.22** .06 -.06 .10 intention <--- perceived behavioral control -.21* .10 .11 .10 intention <--- personal relevance .22* .07 .21* .10
Note. † 0.1 < p < .05; * p < .05; ** p < .01; ***p < .001.
In the next step, all the path parameters in the model were set to be the same
across two groups to test the equality of parameters. After the constraining, the model yielded good model fit: χ²(58) = 62.44, p = .322, χ2/df = 1.08, CFI = .97, RMSEA = .02,
90% CI [.00, .05]. The chi-square difference test did not yield a significant effect, (△χ²=
18.65, △df = 16, p = .287), thus the simpler model (model with equal path parameters
constrain) should be preferred. This result indicates that the model with free estimation of
parameters is not significantly better than the model with equal path parameter
constrains.
Conclusion and Discussion Message framing and time context
The first aim of this study was to investigate the effect of different alcohol
warning label types on alcohol related outcomes. It was expected that alcohol warning
labels with gain-framed health consequences that state the advantage of decreasing
alcohol consumption would lead to more desired changes, compared to labels with
loss-framed health consequences that state the disadvantage of increasing alcohol
consumption. It was also expected that messages with short-term health consequences
that state what might happen after alcohol consumption in a short time period would be
more effective, compared to messages with long-term health consequences that state what
might happen in a longer period. This difference was expected to be stronger when the
message is framed as a gain instead of a loss. Meanwhile, future orientation was expected
to moderate the effect of time context (i.e., short-term versus long-term context), the
advantage of messages with short-term consequences would be bigger if the individual
has a low future orientation.
The result showed that framing the message as gain or loss did not have a
significant effect on attitude. Contrary to what has been expected, gain-framed messages,
intention. Thus Hypothesis 1 was rejected. This finding is against framing theory
(Rothman & Salovey, 1997), which states that for health-affirming behaviors,
gain-framed messages worked more effectively than loss-gain-framed messages. It challenged the
framing theory by showing that the application of framing theory into the context of
alcohol warning labels should be carefully examined. Previous research (O'Keefe &
Jensen, 2009) showed that the effect of message framing depends on different behaviors,
and the current research showed that a loss-framed message has advantage in the alcohol
warning label context. Besides, the current analysis challenged the findings in a previous
meta-analysis (Gallagher and Updegraff, 2012), in which the researchers found that
message framing in general did not have any effect on attitude or intention. The
difference between the current research and this meta-analysis also reflects the fact that
the application of message framing should vary per health behavior. Practitioners should
keep this in mind when designing health messages that targeting different behaviors.
The additional analysis of the conceptual model showed that there’s no significant
difference between the model with group comparison and without group comparison,
indicating that current drinking behavior does not moderate the model paths. However,
when applying the same model to two groups of participants (i.e., higher half or lower
half regarding their drinking amount), the effect sizes and significant levels differ
between two groups. The significant effect of message framing was only found among
participants whose alcohol consumption was in the lower half, not in the higher half. The
result indicates that current drinking behavior might moderate the effect of message
framing on drinking intention. Most of participants in the current sample do not have
made in the analysis might not capture the difference between participants with two level
of drinking. This might be the reason of non-significant model comparison. Further
analysis could take this into account and recruit more heavy drinkers to make a better
comparison between heavy drinkers and light drinkers. Practitioners should consider their
target audience when designing health promotion messages. The differences between
heavy users and light users can also be applied to other health behavior, such as smoking,
drug usage, etc. Further research about communicating health promoting message can
take current usage level as a moderator in the model.
Apart from message framing, this study also manipulated time context of the
alcohol warning label messages. Contrary to what has been expected, short-term or
long-term consequences of alcohol consumption did not show any significant effect on attitude
and intention, neither did the interaction between message framing and time context, and
the interaction between time context and future orientation. Thus Hypothesis 2.1, 2.2 and
Hypothesis 3 were all rejected. This finding was consistent with what had been found by
Churchill, Pavey, Jessop, and Sparks (2016) and Bernstein, Wood and Erickson (2016).
Thus the current research showed that manipulating time context may not be an effective
way in changing alcohol related outcome variables.
Even though the multi-group comparison in additional analysis showed that
current drinking behavior does not moderate the model paths, the path coefficients in the
model and their significant levels are different between the two groups when the model
was tested separately. As mentioned before, the non-significant model improvement
might be induced by the low percentage of heavy drinker in the sample. The findings
the higher half group. This research finding raises attention of the role of current drinking
behavior. Because the alcohol warning messages mainly intend to encourage people to
drink in moderately, the heavy drinkers might be taken as the main focus of this strategy.
Attitude as a mediator
The second aim of this study was to test the mediation role of attitude. After
controlling for other TPB variables, attitude indeed positively predicted intention as
expected, namely a positive attitude towards reducing drinking will lead to a higher
intention of reducing alcohol consumption. However, due to the fact that message type
did not have any effect on attitude, the effect of message type on intention was not
mediated by attitude.
As presented in the result section, attitude towards reducing alcohol consumption
significantly predicted intention of reducing alcohol consumption and there was a strong
correlation between these two variables. In both of the main and additional analysis,
message types showed significant effects on intention but not on attitude. This indicates
that health messages do not necessarily need to change attitude in order to change
intention of a certain behavior. Intention of decreasing alcohol consumption could be an
outcome of cognitive thinking produced directly by being exposed to persuasive
messages. Attitude of decreasing alcohol consumption might be an outcome of all
previous alcohol related experience and it is hard to be changed by a manipulation in a
one-time experiment. Even though attitude is an important predictor of intention
according to TPB model, promoting attitude change does not need to be the priority in
health communication. If the health message has a desired effect on intention, the effect
By adding personal relevance as a control variable, this study also found that
personal relevance has a positive effect on intention and attitude, which means if the
participant perceived the negative health consequence as relevant to himself or herself,
meanwhile s/he will have a more positive attitude and a stronger intention towards
reducing alcohol consumption. Because this variable was not manipulated in the current
study, it is difficult to make a causal conclusion based on cross-sectional data. Thus
future research could investigate the effect of personal relevance on attitude and intention
by manipulating this variable. As suggested by this study, the alcohol warning messages
should include consequences that are more relevant to the young adults group. In this
way, the message can further lead to less positive attitude towards drinking and weaker
drinking intention.
Overall evaluations
The participants of this study were reached through a personal network, which
brought several advantages as well as limitations to the study. The research budget and
time restriction made it beyond the researcher’s ability to conduct a random sampling
process, thus the data was not collected with random sampling methods and it might not
be representative of the whole population. However, due to the fact that the participants
were from around 50 nationalities, the current study is more universally applicable
compared to previous studies of alcohol consumption. After all, most of previous studies
were conducted among western participants in a certain university or in a certain area.
Participants in the current study had more mixed backgrounds, which could be the reason
research could further explore this possibility by conducting cross-culture studies to test
the differences between people from various backgrounds.
Compared to previous research on the effectiveness of alcohol warning messages,
the current study produced less biased research findings because a pre-test was conducted
to select proper experimental material. Messages on alcohol warning labels in this study
were selected based on their believability and valence. In this way, messages in different
experimental conditions on average have similar believability and valance scores, which
ruled out the possibility that these two issues caused the change in intention or attitude.
Apart from benefitting the current study results, the alcohol warning messages can also
benefit future studies by providing several validated study material.
There are also some limitations in the current study. Firstly, the present findings
were based on message manipulation in some two-dimensional alcohol containers. The
effect of alcohol warning message might be different in a realistic situation when the
participants have a real alcohol bottle in hand. The intention of drinking might be more
intensely triggered when participants are facing actual alcohol bottles. Future research
could further explore the effect of message framing by conducting the experiment in a
more realistic situation. Secondly, other potential moderators of message framing were
not tested in the current research. For example, self-esteem (e.g., Covey, 2014); and
perceived risk (e.g., Quick & Bates, 2010). Future research can extend the current study
by testing the effect of moderators of the alcohol warning messages, which could help
with designing more effective and tailored alcohol warning messages. The third
limitation of this study is related to the features of the alcohol warning label. Unlike
to mass audience once they are released to the market. The current study only
manipulated message framing and time context of the message. A significant effect was
found of future orientation on intention, indicating that personalities play an important
role in formulating intention towards drinking less alcohol. Tailored health promotion
messages can put more emphasis on low future orientated individuals because they tend
to have less intention towards stop drinking. However in practice, this variable is less
relevant for alcohol warning label design because the warning labels are designed for the
whole population. Even though high future oriented people showed less intention of
drinking alcohol, it is impossible to specifically target these group when launching
alcohol warning label messages. Future research can explore more variables that can be
manipulated in warning messages, thus the effective ones could be adopted in alcohol
warning label design. The fourth limitation is that this study is a cross-sectional study
without measuring behavior change of the participants. Previous research showed that
there is a gap between intention and behavior (Sheeran, 2002), thus further research
should take this into account and measure behavior change that could be induced by
alcohol warning labels.
Overall, the current study explored the effect of message framing and time
context on the attitude and intention of reducing alcohol consumption in the context of
alcohol warning label, and it also tested the mediation role of attitude. The results showed
that loss-framed alcohol warning labels have marginally significant advantages compared
to gain-framed messages in changing participant’s intention of drinking. However, the
advantage was only showed among people who drink less alcohol. The additional
changing the intention among people who drink more alcohol. This study provided
evidence that alcohol warning label could lead to desired effects on participants’ intention
References
Ajzen, I. (1991). The theory of planned behavior. Organizational Behavior and Human
Decision Processes, 50, 179-211.
Albarracin, D., Johnson, B. T., Fishbein, M., & Muellerleile, P. A. (2001). Theories of
reasoned action and planned behavior as models of condom use: A meta
analysis. Psychological Bulletin, 127, 142-146.
Armitage, C. J., & Conner, M. (2001). Efficacy of the theory of planned behaviour: A
meta‐analytic review. British Journal of Social Psychology, 40, 471-499.
Bell, E., Zizzo, N., & Racine, E. (2015). Caution! Warning labels about alcohol and
pregnancy: Unintended consequences and questionable effectiveness. The
American Journal of Bioethics, 15(3), 18-20.
Bernstein, M. H., Wood, M. D., & Erickson, L. R. (2016). The effectiveness of message
framing and temporal context on college student alcohol use and problems: A
selective e-mail intervention. Alcohol and Alcoholism, 51, 106-116.
Churchill, S., Pavey, L., Jessop, D., & Sparks, P. (2016). Persuading people to drink
less alcohol: The role of message framing, temporal focus and
autonomy. Alcohol and Alcoholism, 51, 727-733.
Courtney, K. E., & Polich, J. (2009). Binge drinking in young adults: Data, definitions,
and determinants. Psychological Bulletin, 135, 142-156.
Covey, J. (2014). The role of dispositional factors in moderating message framing
effects. Health Psychology, 33, 52-65.
Gallagher, K. M., & Updegraff, J. A. (2012). Health message framing effects on attitudes,
Medicine, 43, 101-116.
Gerend, M. A., & Cullen, M. (2008). Effects of message framing and temporal context on
college student drinking behavior. Journal of Experimental Social
Psychology, 44, 1167-1173.
Godin, G., & Kok, G. (1996). The theory of planned behavior: A review of its
applications to health-related behaviors. American Journal of Health
Promotion, 11, 87-98.
Green, L., & Myerson, J. (2004). A discounting framework for choice with delayed and
probabilistic rewards. Psychological Bulletin, 130, 769-792.
Green, L., Myerson, J., Lichtman, D., Rosen, S., & Fry, A. (1996). Temporal discounting
in choice between delayed rewards: The role of age and income. Psychology and
Aging, 11, 79-84.
Greenfield, T. K., Graves, K. L., & Kaskutas, L. A. (1999). Long-term effects of alcohol
warning labels: Findings from a comparison of the United States and Ontario,
Canada. Psychology and Marketing, 16, 261-282.
Hardisty, D. J., & Weber, E. U. (2009). Discounting future green: Money versus the
environment. Journal of Experimental Psychology: General, 138, 329-340.
Hagger, M. S., & Chatzisarantis, N. L. (2005). First- and higher- order models of
attitudes, normative influence, and perceived behavioural control in the theory of
planned behaviour. British Journal of Social Psychology, 44, 513-535.
Hendriks, H., de Bruijn, G. J., Meehan, O., & van den Putte, B. (2016). Online and
offline conversations about alcohol: Comparing the effects of familiar and
Hermens, D. F., Lagopoulos, J., Tobias-Webb, J., De Regt, T., Dore, G., Juckes, L., ...
Hickie, I. B. (2013). Pathways to alcohol-induced brain impairment in
young people: A review. Cortex, 49, 3-17.
Hu, L. T., & Bentler, P. M. (1999). Cutoff criteria for fit indexes in covariance structure
analysis: Conventional criteria versus new alternatives. Structural Equation
Modeling, 6, 1-55.
Huh, D., Mun, E. Y., Larimer, M. E., White, H. R., Ray, A. E., Rhew, I. C., … Atkins, D.
C. (2015). Brief motivational interventions for college student drinking may not
be as powerful as we think: An individual participant level data meta‐
analysis. Alcoholism: Clinical and Experimental Research, 39, 919-931.
Johnston, L. D., O'malley, P. M., Bachman, J. G., Schulenberg, J. E., & Miech, R.A.
(2015). Monitoring the future national survey results on drug use, 1975-2015.
Volume II, College Students & Adults Ages 19-55. Institute for Social Research,
112, 279-287
Keer, M., van den Putte, B., & Neijens, P. (2012). The interplay between affect and
theory of planned behavior variables. American Journal of Health Behavior, 36,
107-115.
Kees, J. (2010). Temporal framing in health advertising: The role of risk and future
orientation. Journal of Current Issues & Research in Advertising, 32, 33-46.
Kingsbury, J. H., Gibbons, F. X., & Gerrard, M. (2015). The effects of social and health
consequence framing on heavy drinking intentions among college
students. British Journal of Health Psychology, 20, 212-220.
