Message framing effects on food consumption: A social marketing perspective

Abstract

Across two studies, this research examines the effects of positive versus negative framing of health messages on food consumption, in the relatively short and long run. Specifically, it studies the effects of multiple exposures to health messages over time (Study 1), using daily food consumption diary data. It replicates and extends those findings in a cross-sectional, single exposure setting to tease out the individual effects of positive and negative framing with consumption intentions as the dependent variable (Study 2). With this combination of an externally and internally valid setup, our research provides robust evidence of the associations of message effectiveness and behavioral outcomes. Our findings suggest that negative messages are more effective than positive messages in the short term but importantly, in the long term as well, suggesting that transforming consumption in the long term can be a viable social marketing objective with the appropriate message tactics.

JEL Classification: M31, M37

Keywords

Food consumption health communication longitudinal effects message framing social marketing

1. Introduction

A wide range of factors have been blamed for the current obesity crisis. Among these factors, food marketing gets a big share of the blame (Seiders and Petty, 2004). Research in the area has shown that marketing tools such as advertising (Kim et al., 2009; Wansink and Ray, 1996), package size (Wansink, 1996), and stockpile-inducing promotions (Ailawadi and Neslin, 1998) have the capacity to non-trivially influence consumption of food and drink, typically with very little cognitive effort or awareness by consumers (see Block et al., 2011). This is especially problematic for children and young adults who are susceptible to food advertising (Young, 2003) which can lead to unhealthy food habits that can be carried into adulthood (Hughner and Maher, 2006). However, research in marketing, health communications, and consumer well-being shows that social marketing (Dibb and Carrigan, 2013) and traditional marketing tactics (Charry and Demoulin, 2012; Krishen and Bui, 2015) could help combat the adverse effects of not only marketing but also other factors that contribute to obesity (Hastings, 2007; Rothschild, 2010). Such tactics have been found to successfully promote positive health behaviors in consumers (e.g. Brennan et al., 2010; Godinho et al., 2016; Krishen and Bui, 2015) and reduce negative health behaviors such as smoking and drinking (Veer et al., 2008; Michaelidou et al., 2008).

Our studies build on the existing research in consumer behavior and health communications to identify how message-related factors might promote healthful behaviors in the domain of (un)healthful consumption and falls squarely in the field of social marketing as defined by McDermott et al. (2005). Specifically, we aim to test the effect of the framing of health messages (positive vs negative) on food consumption in the short and the long run. Study 1 examines the effects of multiple exposures to the health messages (positive and negative) over time, using self-reported daily food consumption diary data. Study 2 replicates and extends those findings in a cross-sectional, single exposure setting to tease out the individual effects of positive and negative framing with consumption intentions as the dependent variable. In doing so, we make three important contributions.

First, while medical as well as social researchers have studied the effects of various factors that contribute to the rise in obesity at the individual and societal levels (e.g. decreases in physical activity, workplace changes, overabundance of food, rising demand for and availability of processed foods, increased promotion of unhealthful food products), they have usually done so at a single point in time (episodically) rather than over time (dynamically). Obesity is not the result of any single, independent consumption transgression but rather of multiple transgressions over an extended time period, so this “snapshot” approach may be an incomplete one. We examine whether social marketing messages might positively influence consumption and prevent transgressions over a single (Study 2), as well as an extended time period (Study 1). As Khare and Inman (2009) point out, research on eating behavior has primarily relied on cross-sectional data and “the dynamic aspect of eating behaviour has not received much attention” (p. 234). Examining food consumption from a dynamic perspective can improve our understanding of the problem, its causes, and its potential solutions. Of note, our choice of time period of 12 days to gain insights into the relatively long-term impact on food consumption was made keeping with existing literature which has used a time period of 2 weeks to study habitual behavior in eating patterns (Childers et al., 2011; Khare and Inman, 2006, 2009). A leading marketing information company, the NPD group, also collects data annually on eating trends where they collect daily diary over a 2-week period (Khare and Inman, 2009). As our results show, the time period utilized is sufficient to capture the decay parameters of the messages.

Second, although researchers have studied message framing in the social marketing domain in the past, for example, sunscreen usage and breast self-examination (Keller and Lehmann, 2008; Meyerowitz and Chaiken, 1987; Rothman et al., 1993; Rothman and Salovey, 1997), these studies do not represent the real-world scenario where consumers get exposed to both positively and negatively framed messages; a scenario that a consumer is more likely to encounter in reality. Past studies have looked at the impact of positively versus negatively framed messages, but not at their commingled impact. In this research, we remedy this in a longitudinal context. Furthermore, we replicate and extend these findings in a controlled, internally valid experiment which examines the effect of various combinations of health messages (positive, negative, mostly positive, and mostly negative) on preference for unhealthy foods.

Third, research has shown that intention does not always translate into behavior and that it is more accurate to measure food consumption, rather than preference or consumption intention (Sheeran and Orbell, 2000; Weinstein et al., 1991). In their meta-analysis, Sheeran and Orbell (2000) show that intention accounts for less than 50% of behavior. Guare and Shulze (2001) find that more than half of consumers enrolled in diet programs regress within the first month, suggesting a huge gap between intentions and behaviors. By using an online diary panel in Study 1, we control for this possibility and focus on the effect of advertising and message framing on food consumption for each participant.

2. Literature review and theoretical development

2.1. Social marketing and health communications

Researchers in the fields of marketing, health communications, and consumer well-being suggest that marketing tools can be used to de-market factors contributing to obesity (Hastings, 2007; Rothschild, 2010). One such marketing tool which is often used is the advertising message, and more specifically, how different aspects of a message (e.g. vividness, temporal framing, negative vs positive framing) may alter the effectiveness of health messages (e.g. Churchill et al., 2014; Menon et al., 2002). For example, caloric and nutritional disclosures provided by restaurants in an easy-to-understand format positively affect consumers’ purchasing decisions (e.g. Howlett et al., 2009). Green and Peloza (2014) also find that various advertising appeals (e.g. cost savings vs societal benefit) effectively promote environmentally friendly consumption, depending on the consumption context (private or public). Churchill et al. (2014) show that a day-framed versus a year-framed message differentially influences the avoidance of high-calorie snacks. However, even motivated consumers may have difficulty engaging in long-term self-regulation (Guare and Shulze, 2001). Thus, it is important to establish whether social marketing can achieve long-term gains in consumer well-being. By suggesting that social marketing can strengthen intentions toward healthful food consumption and change relatively long-term habits, we respond to calls to treat obesity as a widespread social problem that requires novel ways to alter people’s behaviors (Batra et al., 2010; Cheater, 2020).

2.2. Message factors

Messages—especially health-related ones—provide information about risk. The level of perceived risk conveyed by a message depends on its format (Godinho et al., 2016; Keller, 2006); that is, whether it is framed positively or negatively. A positively framed health message, such as eating healthy food to live a longer, healthier life, highlights the benefits a person can achieve by undertaking self-protective healthful behavior. Conversely, a negatively framed health message highlights the consequences of failing to engage in self-protective behavior, such as chronic conditions (e.g. high cholesterol) and negative life events such as heart attacks and strokes. Keller (2006) also finds that people in a promotion focus are more concerned with self-efficacy (will I be able to perform the behavior?) while those in a prevention focus are concerned with response efficacy (will performing the behavior achieve the desired effect?). Further matching efficacy-type with the regulatory focus results in greater intentions to perform the recommended behavior (sunscreen usage). Overall, messages framed in terms of negative consequences (losses) usually result in stronger intentions to engage in healthy behaviors than those framed in terms of positive consequences (gains) (Godinho et al., 2016; Meyerowitz and Chaiken, 1987; Rothman and Salovey, 1997). In a meta-analysis of 60 studies, Keller and Lehmann (2008) conclude that negative (moderate fear) appeals can be effective across different groups; both low- and high-involvement participants are equally persuaded. In a recent paper, Krishen and Bui (2015) show that “fear” driven advertisements are more effective than “hope” driven advertisements on indulgent food consumption and exercise intentions.

Existing literature further indicates that whether a message is effective also depends on its source, audience, and content (e.g. Bertolotti et al., 2016; Muehling et al., 1991; Zhang and Buda, 1999). For example, message framing effects can be moderated by education, such that more (less) educated consumers respond more to positively (negatively) framed messages (Smith, 1996). Chang (2010) suggests that message framing (promotion vs prevention) effects might depend on whether an individual holds an independent or interdependent self-concept. Appendix 1 presents a comprehensive snapshot of the literature on the various message factors that have been examined. One thing that quickly becomes apparent from this review is that message and social marketing tactics have been tested across several domains, targeting a diverse set of outcomes; from intentions to drink alcohol to intentions to use a dental product. More recently, research has also started examining the effect of message framing in the domain of food consumption, and our study lies within this context.

While prior studies are insightful and help with the development of our research, there are three key differences between their findings and ours that emerge from this review. First, most measure intentions and not actual behavior. In our research, in addition to intentions (Study 2), we also use self-reported actual consumption (Study 1). Second, message factors have primarily been examined from the perspective of positive versus negative or gain versus loss framing (these are mostly used interchangeably in the literature) where participants are only exposed to a single message rather than multiple messages of the same or varying valence. This does not represent a realistic scenario where consumers are exposed to positive and negatively framed messages simultaneously. Our research addresses this shortcoming. Third, existing research employs a cross-sectional design to examine the effect of message framing and as such, cannot inform us about how they might impact behaviors over the long term. We provide evidence from both, a cross-sectional (Study 2) and importantly, a longitudinal study (Study 1).

Overall, our research contributes to the growing research on message framing effects (see Appendix 1) and extends it in important ways by examining the question of these effects on healthful consumption by (a) utilizing a more externally valid design in addition to a controlled, internally valid study and (b) over a short as well as an extended time period which allows for an estimation of longitudinal consumption patterns.

2.3. Longitudinal analysis of consumption

Human beings tend to discount the value of future rewards (Berns et al., 2007; Loewenstein and Thaler, 1989). Although it makes normative sense to value future rewards less than equivalent current ones (i.e. temporal discounting; Loewenstein and Prelec, 1992), excessive discounting of future rewards leads to suboptimal outcomes and creates inconsistent time preferences (Frederick et al., 2002). This tendency to value immediate gains at the expense of more valuable future gains underlies problems ranging from debt management to incomplete medical compliance to unhealthy eating and obesity. As Wansink (2006) and Rothschild (2010) argue, most people understand what is good or bad for them but succumb to immediate gratification rather than pursuing their long-term goals. For example, eating a piece of chocolate cake has immediate, tangible utility; exercising for long-term health and fitness has a distant, uncertain utility. In the domain of obesity, the full consequences of such inconsistent preferences become evident only when consumption is studied dynamically, over the long term, rather than as a single episode. That is, addressing obesity requires not just making gains but also maintaining the gains achieved. A cross-sectional approach cannot identify the best long-term strategies to help people regulate themselves and achieve long-term gains. This significant gap in our existing knowledge needs to be addressed.

In addition, the cumulative effect of multiple exposures to a message might be more effective than a single exposure, in terms of promoting positive behavioral changes (Dijkstra et al., 1999). We argue that multiple exposure represents a more likely moderator of the link between intention and behavior. With our longitudinal analyses, we account for the number of times each participant is exposed to negative and positive messages. We thus, argue that a social marketer can control three inputs to improve message retention: (a) the social marketing message, (b) message framing, and (c) message frequency. These factors might strengthen the link between message and behavioral intentions, as well as the link between intentions and self-protective behavior. Next, we present the two studies conducted for the current research.

3. Pre-test: establishing message framing

In total, 210 participants (M_age = 37.06 years, 36.19% females) from Amazon Mechanical Turk (MTurk) completed the pre-test in exchange for monetary compensation. The pre-test featured three positive messages, three negative messages, and an irrelevant message. Participants were randomly assigned to view and evaluate two messages from the total of seven messages. The pre-test focused on establishing cognitive and affective aspects of each message. Each message had food and/or body images, with negative messages focusing on negative consequences of obesity and positive messages focusing on the benefits of healthy diet. The irrelevant condition had a non-consumption-related message. An example of each of these conditions is presented in Appendix 2. Each message was evaluated on a six-item, 7-point scale (1 = strongly disagree, 7 = strongly agree): “ad is effective/positive/cheery/attractive/informational” and “ad focuses on benefits (gains).” The messages (positive, negative, irrelevant) were significantly different in terms of how positive and cheery (r = .76) they were (M_Neg = 3.06, M_Irr = 4.45, M_Pos = 5.01; F(2, 417) = 77.11, p < .0001), as expected. That is, the positive messages were evaluated most positively (5.01) and the negative messages most negatively (3.06), with the neutral/irrelevant message (4.45) in between the positive and negative messages in its evaluation; all pairwise comparisons were significantly different at p = .01. The positive messages also displayed a higher level of perceived attractiveness, information, and benefits (p < .05) but importantly, all messages were evaluated as similarly effective (see Table 1). Furthermore, demographics collected in the survey such as age, gender, ethnicity, and primary language were not significant predictors in message evaluation. Thus, the pre-test establishes that the stimuli were successful in manipulating the perceived message framing in terms of its positivity.

Table 1.

Pre-test: Evaluation of messages.

Measure	Negative	Irrelevant	Positive	F(2, 418)
Cheery	3.06^a,b	4.45^c	5.01	77.11**
Attractive	4.80^a	4.70^c	5.45	8.82**
Information	3.12^a,b	4.95^d	5.46	100.54**
Effective	4.88	4.90	4.89	0.01
Focus on benefits	3.12^a,b	4.75^d	5.37	88.49**

Different from irrelevant at p < .01.

Different from positive at p < .01.

Different from positive at p < .05.

**Significant at p < .01.

4. Study 1: longitudinal impact of health messaging on food consumption

Previous literature has focused on independently estimating the effects of messages that are either positively or negatively framed. Our obejctive in this study is to examine and estimate message effects in a more externally valid setting where participants are exposed to both types of messages along with irrelevant messages multiple times and in random order, and where message impact on participants’ food consumption is measured over a relatively, extended time period.

4.1. Model development

With the objectives of the study in mind, the methodology adopted in the current research allows us to tease out the effectiveness of positive and negative message framing, regardless of the order and frequency of exposure to these messages. In line with the advertising and communication literature, we measure message effectiveness with the decay factor associated with them which captures the longevity of the message’s effect on the dependent variable. Thus, while we trade-off a little on the internal validity, we gain a lot in external validity and generalizability.

Box and Jenkins (1970) describe a dynamic time series as a combined transfer function noise model. We use this as the basis for our estimation. However, instead of sales, our dependent variable is the consumption of a particular product, which we represent as $C_{c t f}$ , where C is the reported consumption by individual c of food category f¹ at time t. In our setup, the consumption of a food category is captured using the number of servings reported by the subject.² In parallel with the advertising sales relationship, we frame our model as follows

C_{c t f} = α_{c} + \sum_{a = 1}^{A} λ^{- 1} (L) b (L) M_{a (t - i)} + N_{t}

(1)

The compact notation $λ^{- 1} (L) b (L)$ captures the transfer function relationship between C_ctf and $M_{a (t - i)}$ . Instead of using traditional dollar (or sometimes unit) values for the advertising expenditure, $M_{a (t - i)}$ is a binary variable that represents the presence (1) or absence (0) of a message M, of the type symbolized by a, at time t – i, where i is a stepped operator (can also be specified as the potency of the message if measures are available). In this case, $N_{t} = \emptyset^{- 1} (L) θ (L) {(1 - L)}^{d} w_{t}$ , where $ϕ (L)$ is the autoregressive process of order p, $\emptyset^{- 1} (L)$ is the moving average process of order q, and ${(1 - L)}^{d}$ is the difference operator of order d. Therefore, L is the standard backshift operator for the estimation. White noise is captured by $w_{t} ~ (0, s_{w}^{2})$ (McCleary and Hay, 1980). Because message is not product specific and repeats across participants, it does not take a subscript. As the equation shows, if a participant is not exposed to any message, the consumption term degenerates to N_t $N_{t}$ which represents smoothed, time-varying consumption.

The introduction of the term M_a(t-i) is extremely important to the model specification because of two reasons. First, it allows us to tease out the effect of exposure to a message for a single individual and second, it allows us to study this effect longitudinally for each message type. We can therefore disaggregate complex scenarios such as watching two positive messages one day and two negative ones the following day. Given the random and complex nature of exposures to messages in the real world, this model in its current formulation is able to capture this variation. It allows for a scenario that is more complex than has been studied in controlled experimental settings. Hence, rather than analyze different message types individually, we use one single combined model to estimate the effects of all three. Also, in a more complex scenario, the value of the variable could be replaced with other measures such as time of exposure, quality of message, and so on.

The flexibility of our model also allows us to specify either a direct or a distributed lag model. A distributed lag model is preferable, because it enables us to include the lagged value of the dependent variable, consumption. Including the dependent variable is preferable because we collected the data over a period of 12 days, and we expect current consumption (both positive and negative) to depend on previous consumption. Thus, assuming that the disturbance term follows a general ARMA process, we can redefine the model³ as

C_{c t f} = \sum_{j = 1}^{m} λ_{j} C_{c t f - j} + \sum_{i = 0}^{I} \sum_{a = 1}^{A} b_{i} M_{a (t - i)} + N_{t}

(2)

where $C_{c t f - j}$ is the consumption of food category f at time t – j, m is the number of lagged consumption terms, and i represents the number of time periods being analyzed. We therefore define the first two components of Equation (2) as the transfer function component, with $N_{t}$ as the noise function component. The transfer function component, $\sum_{j = 1}^{m} λ_{j} C_{c t f - j} + \sum_{i = 0}^{I} \sum_{a = 1}^{A} b_{i} M_{a (t - i)}$ can be represented as $v (L) M_{a t}$ . Here, $v (L)$ becomes the polynomial operator that relates food-category-specific consumption to advertising, the general form of which can be represented by $v (L) = (b (L) / λ (L)) L^{b}$ (Hanssens et al., 2001). Furthermore, $λ (L)$ is the rth-order polynomial operator, $L^{b}$ is the bth-order dead-time operator, and $b (L)$ is the sth-order polynomial operator. The two sets that need specification are (p d q) for the noise component function and (r s b) for the transfer function component.

The percentage implied duration interval for the time length of the message effect is provided by (Clarke, 1976)

l^{*} \leq 1 + \frac{\log (1 - h) - \log [\frac{b_{1} + λ b_{0}}{b_{1} + λ b_{0}}]}{\log λ}

(3)

With this calculation, we can determine the decay factor of the message in question. Thus, to calculate the half-life of a message, we set h = .5. $l^{*}$ depicts the number of time units required for half of the effect of the message to disappear. We calculate the half-lives to illustrate message type effectiveness as well as report decay at 10% intervals.

4.2. Procedure and data collection

We collected responses from 116 individuals (45.6% women) from an online diary-based study. We informed all participants that they would have to fill an online diary at least 10 times over a 12-day period, with mandatory responses on the first and last days. A total of 92 participants satisfied this condition; they were retained for the analyses. The choice of around a 2-week period was based on existing literature (Khare and Inman, 2009), as we discuss previously. In addition, as our results validate later, the choice was well above the estimated half-life of all types of messages and hence did not require any extrapolation.

We updated the survey daily and collected participants’ responses. The survey on day 1 differed from those on other days, in that we asked participants about their demographics (gender, age, income, and ethnicity), weight, and height, so that we could calculate their body mass index (BMI), on day 1. We also collected information about whether they were currently in the process of dieting and if English was their first language. We used these factors as control variables. On the first day of the survey, we also explained the size of the unit “serving” and types of food groups to participants. On all days, we collected data about the number of servings consumed on the previous day in six food categories: (a) vegetables, (b) fruits, (c) grains or starchy foods,⁴ (d) dairy products, (e) high-quality proteins, and (f) high-fat foods.

To avoid demand effects and response bias, we asked participants to follow a sequence: report food consumption for the previous day, complete a filler task, and view the messages (in three of the four conditions, as outlined below). After completing the detailed food log, and completing the filler task, participants were exposed to one of four possible conditions. In the first condition, participants saw a negatively framed social marketing message focusing on the consumption of unhealthy food products, obesity, and its negative consequences. In the second condition, they viewed a positively framed message focused on healthy, balanced eating. In the third condition, participants viewed a social marketing message that was irrelevant to the task at hand (with varied framing—positive, negative, and neutral; for example, focused on drug abuse, skin cancer, environmental safety, and subliminal advertising). Finally, in the fourth condition, participants did not view any message. Thus, on a given day, each participant could see either a message (positive, negative, or irrelevant: see Appendix 2) or no message. Allowing subjects to view different types of messages makes our model development and estimation more rigorous and generalizable.

Over the next 11 days (no messages were shown on day 1), 89 participants were exposed to a random selection of each of the four conditions (3 participants were dropped because of incomplete responses). A total of 965 responses were recorded during this period and analyzed. On average, each participant was exposed to 8.59 messages across this time period. Without weighting the frequency of exposures, the average number of messages viewed across all participants over the 12-day time period of data collection was as follows: 2.76 positive messages, 2.91 negative messages, and 2.92 irrelevant messages, on average.

Similar to the pre-test, we evaluated how positive and cheery each message was perceived to be, on a seven-point scale (1 = strongly disagree, 7 = strongly agree) using two items (r = .82). We averaged the items to form a composite score for each message type (positive, negative, and irrelevant). The results show that the three message types prompted different perceptions of their positivity (M_Neg = 2.13, M_Irr = 3.92, M_Pos = 5.11), and each condition differed significantly (p < .01) from the other two.

4.3. Analysis

From Equation (2), the variable a can take three possible values. If a = 1, it indicates an exposure to a positively framed message; 2 indicates a negatively framed message; and 3 indicates an exposure to an irrelevant message. Technically, a participant might not have watched a message in a certain period, so a = 4 should be iterated. However, the lack of a message leads Equation (1) to degenerate into noise term only, so iterations were restricted to 1–3.

To make the analyses tractable, we averaged the six food categories into one composite measure of total healthful consumption, with high-fat foods reverse coded.⁵ The Cronbach’s alpha for the six categories was .62, but it improved to an acceptable .73 after we deleted dairy products (which had the lowest correlation to the total). We therefore removed this product category from further analyses and used a composite mean measure of the remaining five categories as the dependent variable in the reported analyses. We should clarify that analyses conducted on the six food categories individually yielded similar estimates across the categories and therefore, for the sake of tractability, only the combined results are reported. (Table 2 displays half-lives for all categories individually.⁶ Irrelevant messages decayed to 50% in approximately a day for all categories and have not been included in the table.)

Table 2.

Study 1: Half-life estimates of messages by individual food categories.

Percentage decay	Positively framed	Negatively framed
Vegetables	3.82	7.43
Fruits	4.16	7.49
Grains and starchy foods	3.80	7.24
Dairy products	3.08	6.43
High-quality proteins	4.10	7.24
High-fat foods	3.75	7.60

4.3.1 Selection of order of decay

The selection of the iterative orders for all three (decay, specification, and disturbance) is based on the existing literature (Hanssens et al., 2001). The order of decay (r) was iterated from 1 through 3. However, for orders of decay greater than 1, the estimates of λ were non-significant, implying that the model specification was overly complicated. This was true for all three types of messages. The final analysis therefore contains the results for only the first level of decay; that is, the one based on a single lagged consumption term. This is in line with the findings from several authors such as Bass and Clarke (1972) who note that most advertising models do not demonstrate a decay order of greater than 1.

4.3.2 Selection of order of specification

The order of the (s) specification was also iterated from 1 to 3. For s = 2 or 3, all values for $\hat{b}$ were non-significant. Therefore, considering the significance of the parameter estimates, we report only the results from s = 1. However, for the irrelevant message type, no $\hat{b}$ was found to be significant, so the results for this set of messages do not include parameter estimates for ${\hat{b}}_{1}$ , unlike the other two types of messages.

4.3.3 Selection of order of disturbance

The autoregressive order for the disturbance term (p) was set at 2 for our analyses, in accordance with Weiss and Windal (1980). Although many advertising studies set (b), (d), and (q) to 0, we initially set b to 1, with the assumption that participants would demonstrate an effect of the advertising on their consumption after one period. We make this assumption because the time interval (i.e. 1 or 2 days) for our data collection is small and the memory of the event and hence its effect might be high. Although it conflicts with convention, this approach follows the suggestion of several authors (e.g. Moriarty, 1985) to make the model more robust. However, based on the estimation process, b was non-significant for all cases and all messages, so we re-ran the analysis with b reset to 0.

The final models for our analyses were (110) (200) for the positively and negatively framed message and (100) (200) for the irrelevant message. We added the demographic variables to the final set of estimations. None of the covariates, including age, gender, ethnicity, or BMI, was significant in any of the cases, and is not discussed further. Importantly, the results do not vary with or without the covariates in the estimation.

4.4. Results

As we reveal in Table 3, for the positively framed message, the first order of decay was statistically significant at ${\hat{λ}}_{1} = . 816$ . The first order of specification also was significant at ${\hat{b}}_{0} = . 280$ and ${\hat{b}}_{1} = . 254$ . The first-order autoregressive disturbance term $\hat{α} = . 392$ was also significant. The estimate for $\hat{β}$ was marginally significant at p = .07, and we maintained the second-order term, which yielded the best result compared with the other models we estimated. The overall model fit was acceptable.

Table 3.

Study 1: Parameter estimates for the three types of messages.

Parameters	Positively framed (110) (200)			Negatively framed (110) (200)			Irrelevant (100) (200)
	Estimates	SE	t-ratio	Estimates	SE	t-ratio	Estimates	SE	t-ratio
${\hat{λ}}_{1}$	0.816	0.193	4.228	0.906	0.441	2.054	0.317	0.149	2.128
${\hat{b}}_{0}$	0.254	0.113	2.248	0.160	0.031	5.161	0.468	0.230	2.035
${\hat{b}}_{1}$	0.280	0.128	2.188	0.072	0.028	2.571	–	–	–
$\hat{a}$	0.392	0.120	3.267	0.596	0.122	4.885	0.407	0.160	2.544
$\hat{β}$	0.191	0.098	1.949	0.384	0.126	3.048	0.467	0.125	3.736

The results were similar for the negatively framed messages. The estimates for the first-order decay were statistically significant at ${\hat{λ}}_{1} = . 906$ . A comparison of this estimate with the positively framed message reveals that the carryover effect is much stronger for the negatively framed message. The first order of specification was again significant at ${\hat{b}}_{0} = . 160$ and ${\hat{b}}_{1} = . 072$ . Both disturbance terms were significant as well, $\hat{α} = . 596$ and $\hat{β} = . 384$ . The model fit was thus, acceptable. The irrelevant message could only fit a (100) (200) model, and the order of specification had to be set to 0. The order decay was statistically significant, though substantially lower at ${\hat{λ}}_{1} = . 317$ , which implies that this message condition had the lowest carryover effect. This result is not surprising; the messages did not relate to food consumption.

4.5. Interpretation and representation of effects of advertising decay for messages

The values obtained from the estimation process allow us to investigate the decay of the three types of messages using Equation (3), such that we derive the implied duration interval for the effect of advertising. Although the equation applies well to the positively and negatively framed messages, we estimated the irrelevant message as a (100) (200) model. The value of ${\hat{b}}_{1}$ thus is 0, which reduces the equation to the Koyck (1954) duration interval model, in which

l^{*} \leq [\frac{\log (1 - p)}{log(λ)}]

(4)

Using Equation (4), we plotted the decay rates of the three sets of messages. The results are in Table 4, and we plot them in Figure 1. Note that the decay parameters are plotted based on the decay percentage rather than the conventional time elapsed. This has been done to keep the estimates for the three message types easily comparable. In turn, we can estimate the time periods at which certain levels of the messages’ effects have dissipated, using 10%–90% dissipation, in increments of 10%. The half-life of the three types of messages are highlighted in the table.

Table 4.

Study 1: Decay estimates for the three types of messages.

	Percentage decay	Positively framed	Negatively framed	Irrelevant
	0.10	1.07	1.39	0.63
	0.20	1.65	2.58	0.73
	0.30	2.30	3.93	0.84
	0.40	3.06	5.50	0.98
Half-life of the message	0.50	3.96	7.34	1.14
	0.60	5.06	9.60	1.33
	0.70	6.47	12.52	1.58
	0.80	8.46	16.62	1.93
	0.90	11.83	23.61	2.51

Figure 1.

Decay patterns for the three message types (in number of days to reach percentage decay).

Because we conducted the analysis on a daily basis, the results include days as the unit. We find that 10% of the message’s effect dissipates in 1.07 days for the positive message; the same level of dissipation takes about 1.39 days for the negative message. The difference across the three message types is even more evident for the most utilized measure of decay, the half-life, which was 3.96 for positive messages and 7.34 days for negative ones. That is, a positive message retains only half of its potency after 3.96 days whereas a negative message does so for almost double that time, 7.34 days. It is noteworthy that the length of the data collection is sufficient to capture the long-term effects of the messages. Furthermore, the method allows us to project the complete lifespan of the messages as shown in Figure 1.

Next, we present Study 2, a cross-sectional experiment with a different dependent variable, consumption intentions.

5. Study 2: teasing out message effects on consumption intentions

Study 2 seeks to build on the results of Study 1 in two important ways. One, replicate and extend the findings of Study 1 in an internally valid, controlled experimental setup with consumption intentions. Second and more importantly, to clearly tease out the effects of positive and negative messages by implementing a design with conditions where only one type of message is used. We do so by including experimental conditions where we manipulate the exposure of participants to all negative, all positive, mostly negative, and mostly positive messages.

5.1. Participants and experimental design

A one factor, four levels (message type: all negative (three negative ads), mostly negative (two negative ads and one positive ad), mostly positive (two positive ads and one negative ad), and all positive (three positive ads)) between-subjects design was implemented in study 2. Participants were recruited from MTurk in return for financial compensation. A total of 288 participants (M_age = 37.52 years, 33.68% females) completed the study.

5.2. Procedure

Each participant was randomly assigned to one of the four conditions described above consisting of a pre-determined set of advertisements. They were asked to complete two ostensibly unrelated surveys in the following order: (a) information evaluation of messages and (b) a study of food preferences. They were then asked questions regarding their eating habits and knowledge of food-nutrition. Following these, demographics were collected.

5.2.1 Manipulation checks for messages

As in the pre-test, participants evaluated the messages on a six-item, 7-point scale (1 = strongly disagree, 7 = strongly agree): “ad is effective/positive/cheery/attractive/informational” and “ad focuses on benefits (gains).” The message conditions (all positive, all negative, mostly positive, and mostly negative) were significantly different in terms of how positive and cheery (r = .75) they were (M_AllNeg = 2.82, M_MostNeg = 3.77, M_MostPos = 4.45, M_AllPos = 5.13; F(3, 286) = 29.24, p < .0001), as expected. All conditions were significantly different from each other, at p < .01. The messages were also different in terms of their perceived information and focus on benefits (p < .01) but no differences emerged on message attractiveness. Importantly, all messages were evaluated as similarly effective (see Table 5). Thus, the manipulation checks establish that the different conditions successfully manipulated the message framing, as intended.

Table 5.

Study 2: Evaluation of ads.

Measure	All negative	Mostly negative	Mostly positive	All positive	F(3, 286)
Cheery	2.82^a,b,c	3.77^b,c	4.45^c	5.13	29.24**
Attractive	5.25^d	5.51	5.55	5.82	1.95^ns
Information	3.08^a,b,c	4.00^d,c	4.66^c	5.78	34.13**
Effective	5.13	5.02	5.04	5.22	0.31
Focus on benefits	3.19^a,b,c	4.12^b,c	4.59^c	5.42	17.94**

Different from mostly negative at p < .01.

Different from mostly positive at p < .01.

Different from all positive at p < .01.

Different from all positive at p < .05.

**significant at p < .01.

5.2.2 Consumption preference for hedonic, unhealthy food products

Next, participants self-reported their consumption preference for two hedonic food items—chocolate and cookies. The products were chosen based on extant literature that has found them to be consistently perceived as hedonic and unhealthy (e.g., Garg et al., 2007; Govind et al., 2020; Wansink et al., 2003). Participants indicated their preference/liking for that product “right now” on a 7-point scale (1 = do not prefer it at all, 7 = strongly prefer it).

5.2.3 Demographic and control variables

These included age (in years), biological sex (1 = female, 2 = male), and ethnicity (Caucasian/African American/Hispanic/Asian/Others), along with other variables such as dieting status (1 = dieting and 0 = not dieting) and their weight (in pounds) and height (in feet and inches). A preliminary model included gender, age, ethnicity, and BMI as control variables but all were non-significant, p > .10. They are hence not discussed further. In addition, prior research shows that restrained versus unrestrained eaters respond differently to contextual factors such as affect and stress (Ward and Mann, 2000) that might be induced by the health messages. Our sample consisted of 82 dieters. We ran the analysis with and without dieters, and the results replicated. The results reported are thus inclusive of the dieters.

5.3. Results

The effect of the four message conditions on consumption preference for chocolates and cookies was analyzed.

5.3.1 Cookies

The effect of message type on the preference for cookies was found to be significant, F(3, 286) = 4.20, p < .01. In a pairwise comparison of the conditions, the difference between the two extreme conditions was found to be significant $(M_{All Negative} = 4.21 < M_{All Positive} = 5.23, p < . 01)$ . The mostly positive condition $(M_{Mostly Positive} = 4.80)$ was marginally different from the all negative condition (p = .05), but not different from the mostly negative condition or from the all positive condition. The mostly negative condition $(M_{Mostly Negative} = 4.42)$ was statistically different from the all positive condition (p < .05), but not different from the mostly positive condition or from the all negative condition. The order of the preference for cookies was thus in the expected direction. Details of the effects are provided in Table 6.

Table 6.

Study 2: Preference for unhealthy product across different message types.

Measure	All negative	Mostly negative	Mostly positive	All positive	F(3, 286)
Cookies	4.21^c	4.42^d	4.80	5.23	4.20**
Chocolates	4.25^b,c	4.54^c	5.02	5.10	4.05**

Different from mostly positive at p < .01.

Different from all positive at p < .01.

Different from all positive at p < .05.

**Significant at p < .01.

5.3.2. Chocolates

The preference for chocolates followed a similar pattern to that of cookies. The effect of the message condition was found to be significant, F(3, 286) = 4.05, p < .01, and pairwise comparisons of the conditions yielded results in the expected direction. Specifically, the difference between the two extreme conditions was found to be significant $(M_{All Negative} = 4.25 < M_{All Positive} = 5.10, p < . 01)$ $(M_{All Negative} = 4.25 < M_{All Positive} = 5.10, p < . 01)$ . Furthermore, the mostly positive condition $(M_{Mostly Positive} = 5.02)$ was statistically different from the all negative condition (p < .01), but only partially different from the mostly negative condition (p = .07) and not different from the all positive condition. The mostly negative condition $(M_{Mostly Negative} = 4.54)$ was statistically different from the all positive condition, but only partially different from the mostly positive condition (p = .07), and not different from the all negative condition. A comparison of the conditions is provided in Table 6.

Study 2 replicates and extends the results of Study 1 in a controlled, internally valid single exposure context with a different dependent variable—consumption preference for hedonic food products. The design implemented in this study allows us to address the potential concern in Study 1 where positive and negative messages are mixed and thus, enables us to tease apart the effect of message framing more cleanly. Overall, these results help establish the robustness of the findings in Study 1 by utilizing a new context and a new dependent variable.

6. General discussion

Obesity is a rapidly growing global epidemic (World Health Organization (WHO), 2015). The number of obese people worldwide has doubled since 1980 with more than 650 million obese adults and 42 million overweight or obese children (under 5) in 2014 (NCD Risk Factor Collaboration (RFC), 2016). Alarmingly, in recent times, this trend has spread even to developing countries. The obesity epidemic exacts heavy tolls on both individuals and society, with significant implications for their long-term welfare. At the individual level, obesity is a major risk factor for chronic diseases, including type 2 diabetes, cardiovascular disease, hypertension and stroke, and certain forms of cancer. These diseases (four of the top five costliest) impose a deep financial burden, in addition to affecting people’s health and quality of life (NCD RFC, 2016). At the societal level, obesity has played a large role in the exponentially rising costs of health care. For example, the health care expenditures in Australia rose from 8.7% of gross domestic product (GDP) in 2006–2007 to 10.3% of GDP in 2015–2016 (Australia Institute of Health and Welfare, 2018). Similarly, the United States has seen its public health care expenditures increase from 13.4% of GDP to 17.9% in a decade (1999–2009), whereas the expenditures rose from an average of 9% of GDP in the European Union in 2006 to about 13% in 2010 (Przywara, 2010). These rising costs, along with the motivation to improve overall quality of life, have prompted health care professionals and policy makers to start shifting their focus from cure to prevention.

Our research answers the call for consumer research improving consumer well-being and health, especially in the socially pertinent domain of food consumption and its negative consequences such as obesity that have started to affect the developing world in addition to hurting the developed ones (e.g. Block et al., 2011; NCD RFC, 2016). Even as policy makers take steps to address this important issue, researchers have argued that new ideas and approaches are needed for lasting change (e.g. Batra et al., 2010).

Our findings establish that in the short and the long term, consumption patterns can shift with repeated exposures. Specifically, our longitudinal data on the amounts of various food groups consumed find that negatively framed messages are almost twice as effective as positively framed messages in curbing the consumption of unhealthy food products (7.34 vs 3.96 days as their half-lives). Study 2 replicates and extends the findings from Study 1 with a controlled, cross-sectional experiment where participants were randomly exposed to one of the four different message conditions such that some saw only positive messages, some only negative, and others were shown predominantly positive or negative messages. This study also supports the efficacy of negative message framing (vs positive) in terms of influencing individuals’ consumption intentions. Overall, our results support prior as well as emerging research which has shown negative framing to be more effective across population groups including children (Charry and Demoulin, 2012; Krishen and Bui, 2015).

Furthermore, our research provides an interesting insight. Specifically, even though positive messages were rated higher on cognitive aspects such as information and benefits, negative messages were more persuasive in terms of downstream effects on intentions and behavior. Both studies support this result. This suggests that in the context of food consumption, affect may be a more powerful driver than cognition. This is an important finding and suggests avenues for future research to gain clearer insights into these relationships. Finally, our findings have important implications for stakeholders such as healthcare professionals who motivate patients to adopt healthier lifestyles, social marketers who communicate messages about appropriate health behaviors, and consumers who want to promote their own long-term well-being.

Our research makes several contributions. First, while a longitudinal approach with diary panels appears in other contexts, we adopt this dynamic approach to the study of the effects of social marketing messages on food consumption. Existing research has provided important insights into the phenomenon; however, it has undertaken a cross-sectional examination of the effect of message framing and as such, cannot inform us about how it might impact behaviors over the long term. By taking a longitudinal approach, our research allows us to specify the effects of multiple health message exposures over time.

Second, in Study 1, we include food consumption data, collected daily, rather than consumption preferences or intentions. Although intentions can lead to behavior, the effect is usually modest (Armitage and Conner, 2001; Godin and Kok, 1996; Sheeran, 2002). For example, several researchers have shown that intentions account for only 30% of the variance in future behavior, thus 70% is unaccounted for as measured in studies using the Theory of Planned Behavior. Another study that measured dietary behavior 6 years after measuring intentions found that intentions predicted only 9% of the variance in behavior variance (Conner et al., 2002). By studying actual food consumption, we are able to draw stronger conclusions about message effectiveness and behavioral outcomes.

Third, our research design across both studies focuses on effects (Weiss and Windal, 1980) of positively and negatively framed messages presented in combination, which is a more realistic approach because consumers are seldom exposed to uniform message framing (positive or negative). Existing research has manipulated message framing (positive vs negative or gain vs loss) independently across different individuals. While this allows for a more controlled setup, it diminishes external validity of the findings. Most individuals are exposed to both positively and negatively framed messages, almost simultaneously in the real world. Thus, our research enables us to assess message effectiveness in a more realistic setting where participants are randomly exposed to both positive and negative messages. Furthermore, we tease apart the distinct effects of positive and negative framing in this context, to establish their independent efficacies.

7. Limitations and Future Research

Our findings suggest that emphasizing the negative repercussions of failing to meet health goals might help health care professionals achieve effective changes in consumption habits. In addition, health care professionals need to repeat such messages over time. In this context, it would be interesting to evaluate the differential efficacy of various communication media, such as face-to-face interactions, information material in doctors’ offices, and direct mail, in additional research. Also, it is noteworthy that our sample includes the general population rather than special groups such as dieters, obese individuals, or those who are on a weight loss program. Future research should examine the efficacy of such messages across different groups to gauge the associated benefits.

Our results also establish that when consumers are exposed to a message campaign multiple times, social marketing messages have a positive impact on the nutritional content of food consumed. Although both negative and positive framings have impact, the negative version is more effective than positive framing, possibly because (a) negative message framing leads to stronger intentions (Study 2) which have stronger impacts on behavior, or (b) negative messages induce better retention which translates into greater self-protective intentions and behaviors. The scope of the current study did not allow us to test these potential drivers but this presents a fruitful direction of inquiry for future research. In addition, the interactive nature of the viewing order of messages might also be a factor that influences consumption. In the current research, the methods employed do not allow us to analyze these. Of note, even if this could be achieved, it would be of limited use as advertisers can seldom control the order that messages are viewed in, although they can control the order in which they are released.

As we discussed earlier, it can be argued that using self-report data might be prone to social desirability bias or might be inaccurate because of inaccurate recall. However, as Khare and Inman (2009) point out, if researchers pay attention and use “sound data quality checks and good reasoning” (p. 250), data collected using self-reports can be insightful. In the current research, the online diary provides explanation for serving size in various food categories along with relevant examples and collects several control variables to maintain the reliability of the data and judge its validity. Furthermore, it is important to note that despite its limitations, self-report data remain the best and the only viable option in measuring eating behaviors in a longitudinal context (Khare and Inman, 2009).

In the interests of tractability and practical implementation capacity, we collapsed our analysis across various types of food. We hope that additional research measures the effects of consumption of various food groups, and of various messages, in curtailing the consumption of specific food groups. Finally, future research could explore whether and how the relationship between intention and behavior is moderated by differential retention of messages, by collecting data pertaining to these variables, as well as implementing the relevant changes in the empirical analysis.

Footnotes

Appendix

Appendix 1.

Review of the literature on message factors.

Article	Message factors^a	Dependent variables	Main findings
Abhyankar et al. (2008)	Gain vs Loss Frame	Intention to Take MMR Vaccination	A loss-framed (vs gain-framed) message is more effective in increasing MMR vaccination Intention. This effect is stronger for women who have previously made the decision to vaccinate their child.
Agrawal and Duhachek (2010)	Guilt vs Shame Appeal	Likelihood to Binge Drink	When the ad frames others as sufferers (vs Observers), guilt (vs Shame) emerges. Compatible appeals (i.e. appeals that elicit the same emotion as being incidentally experienced) are less effective than incompatible appeals.
Agrawal et al. (2007)	Self- vs Others-Appeal	Self-risk of Having Hepatitis C	When people are primed with positive emotions, compatibility between emotion and appeal (e.g. happy with self-appeal, peacefulness with others-appeal) increases the effectiveness of messages. When people are primed with negative emotions, compatibility between emotion and appeal decreases the effectiveness of messages (e.g. sadness with self-appeal, agitated with others-appeal).
Albarracin et al. (2003)	Abstinence vs Moderation Appeal	Intention to Drink Alcohol	When there is no trial, two appeals have the same effect. However, people who are exposed to the abstinence appeal message who have tried the product also have a stronger intention to use the product in the future.
Apanovitch et al. (2003)	Gain vs Loss Frame	HIV Testing Intention	Loss-framed messages are more effective for people who are uncertain about what the outcome of the test would be. For people who are certain that the test would not find the presence of HIV, gain-framed messages are more effective in promoting testing.
Arora (2000)	Positive vs Negative Frame Source Credibility	Attitude and Intention to Obtain Dental Exam	There is a strong main effect of credibility on intention to take dental exam. Negative-framed messages are more persuasive than positive-framed messages.
Arora et al. (2006)	Positive vs Negative Frame Source Credibility	Attitude and Intention to Engage in Exercise	When the source credibility is low, positively framed messages are more effective than negatively framed messages. However, when the source credibility is high, their effectiveness is non-significant.
Bannon and Schwartz (2006)	Gain vs Loss Frame	Kindergarten Children Food Choice (Healthy vs Unhealthy)	Nutritional messages have significant effects on children’s short-term food choice (vs Control condition). However, no difference is found between gain- and loss-frame messages.
Bartels et al. (2010)	Gain vs Loss Frame	Attitude and Interest on Vaccine	When the risk associated with a health behavior is low, people respond more favorably to gain-framed messages. However, when the risk associated with the health behavior is high, people respond more favorably to loss-framed messages.
Berry and Carson (2010)	Gain vs Loss Frame	Intention to Engage in Healthy Exercise	Message framing does not affect the persuasiveness of exercise messages.
Bertolotti et al. (2016)	Health vs Well-being Frame Factual vs Pre-factual (“If . . . then”) Frame	Intentions to Eat Meat	Intentions to eat meat is reduced when the messages use the combination of pre-factual, well-being-focused messages or factual, health-focused framing.
Block and Keller (1995)	Positive vs Negative Frame High- vs Low-Efficacy Appeal	Intention to Comply with the Recommendations concerning STD and Skin Cancer	In the low-efficacy condition, negatively framed messages lead to a greater intention to follow the recommended behavior. In the high-efficacy condition, positively and negatively framed messages are equally persuasive.
Bosone et al. (2015)	Promotion vs Prevention Frame	Intention for a Healthy Diet	People exposed to “fitting” messages (e.g. a positive model and a promotion-framing, a negative model and a prevention-framing) are the most willing to engage in a healthy diet.
Brick et al. (2016)	Gain vs Loss Frame	Dental Flossing Behavior	Since US culture emphasizes individualism and approach orientation, a greater cultural exposure improves patient choices and memory for gain-framed messages. Individuals with less exposure to US culture show these advantages for loss-framed messages.
Broemer (2002)	Positive vs Negative Frame	Attitude and Intention for Exercising and Safe-sex Behavior	Highly ambivalent individuals are more persuaded by negative-framed messages. Individuals low in ambivalence are more persuaded by positive-framed messages.
Broemer (2004)	Positive vs Negative Frame	Attitude on Healthy Diet	Negatively framed messages are more persuasive when the symptom is easily imagined. Positively framed messages are more effective when the symptom is difficult to image.
Brug et al. (2003)	Gain vs Loss Frame	Attitude and Intention for Healthy Diet	No significant differences in attitude or intention to perform the preventive nutrition behaviors are found between the gain-frame and loss-frame conditions.
Chang (2007)	Positive vs Negative Frame	Intention to Use (Dental) Product	People show higher behavioral intention for new promoted products (prevention or detection) in the gain-framed ad. The framing effects are stronger in promoting new products than in promoting familiar products.
Cho and Boster (2008)	Gain vs Loss Frame	Attitude and Intention to Use Drugs	Loss-frame (vs gain-frame) messages are more persuasive for adolescents who report that their friends use drugs. Neither gain nor loss framing have a persuasive advantage for adolescents who report that their friends do not use drugs.
Churchill et al. (2014)	Day vs Year Frame	Self-reported Snack Consumption	Individuals with low levels of eating self-efficacy have lower levels of snacking when the message uses year (vs day) framing
Collymore and McDermott (2016)	Gain vs Loss (Fear, Disgust) Frame	Intention to Reduce Alcohol Consumption	Loss-framed messages, in particular those featuring disgust, are the most effective for increasing intention to reduce alcohol intake.
Cox and Cox (2001)	Gain vs Loss Frame Statistical vs Anecdotal Evidence	Likelihood to Have Mammogram (Detecting Breast Cancer)	Positive anecdotes are less persuasive than negative anecdotes. Framing with statistical evidence has non-significant effect among people.
Detweiler et al. (1999)	Gain vs Loss Frame	Intention to Use Sunscreen	People who read the gain-framed (vs loss-framed) brochures are significantly more likely to request sunscreen, intend to repeatedly apply sunscreen while at the beach, and intend to use sunscreen with a SPF of 15 or higher.
Duhachek et al. (2012)	Gain vs Loss Frame	Intention to Binge Drink, Viewing Time of Alcohol Drinks Ads, and Drinking Intention	Guilt appeals are more effective when paired with gain frames. Shame appeals are more effective when paired with loss frames.
Gallagher and Updegraff (2011)	Gain vs Loss Frame Intrinsic vs Extrinsic Appeal	Attitude on Exercise	Gain-framed messages would “fit” with intrinsic outcomes (e.g. satisfaction, enjoyment) and loss-framed messages would “fit” with extrinsic outcomes (e.g. appearance, health). However, NFC moderates these effects. When people have high NFC, such fit effect would occur, whereas when people have low NFC, non-fit effect would occur.
Gallagher and Updegraff (2012)	Gain vs Loss Frame	Meta-analysis	Gain-framed (vs loss-framed) messages are more persuasive to encourage illness prevention behaviors overall. No significant effect of framing is found when persuasion is assessed by attitude or intention, or among studies encouraging detection behaviors.
Gerend and Maner (2011)	Gain vs Loss Frame	Fruit and Vegetable Intake (Servings per Day)	Fearful people eat more servings of fruits and vegetables after exposure to a loss-framed (vs gain-framed) message. In contrast, angry people eating (marginally) more servings of fruits and vegetables after exposure to a gain-framed (vs loss-framed) message.
Godinho et al. (2016)	Gain vs Loss Frame	Intentions and Self-reported Consumption of Fruit and Vegetable	For prevention-focused individuals, loss-framed message is more effective. However, for promotion-focused individuals, loss- and gain-framed messages are equally effective. For individuals with low levels of baseline intention, neither gain- nor loss-framing is effective. However, for those with high levels of baseline intention, loss- (vs gain-) framing is more effective.
Goodall and Appiah (2008)	Gain vs Loss Frame	Attitude and Intention to Smoke	Adolescents have more favorable attitude toward the loss-framed (vs gain-framed warnings). Smokers exposed to the loss-framed messages have significantly lower intention to smoke in the future.
Hoffner and Ye (2009)	Gain vs Loss Frame Exemplary (Similar vs Not Similar)	Intention to Use Sunscreen	Both frames (vs Control) increase intention to use sunscreen. Planned SPF is higher for men in the loss frame, but it is unaffected by framing for women. The gain frame is more effective for people who are high in similarity with the model, whereas the loss frame is more effective for people low in similarity with the model.
Jones et al. (2003)	Positive vs Negative Frame Source Credibility	Intention, Attitude, and Exercise Behavior	Positively framed communication from a credible source is more effective to increase exercise intention and behaviors.
Jones et al. (2004)	Positive vs Negative Frame Source Credibility	Intention, Attitude, and Exercise Behavior	No effects of framing and credibility are found.
Keller (2006)	Self-efficacy vs Response Efficacy Appeal	Intention to Use Sunscreen	When self-efficacy features (perceived ease) are paired with promotion focus and when response efficacy features (perceived effectiveness) are paired with prevention focus, health messages are more effective.
Keller and Lehmann (2008)	22 Tactics (e.g. Fear, Framing, Source Credibility)	Meta-analysis	Message tactics have a significant influence on intention toward health-related recommendations, controlling for different individual personality.
Keller et al. (2003)	Gain vs Loss Frame	Intention to Have Mammogram	People induced with a positive mood are more persuaded by the loss-framed message. People induced with a negative mood are more persuaded by the gain-framed message.
Krishen and Bui (2015)	Fear vs Hope	Intention to Choose Dessert and Intention to Increase Daily Exercise	Fear appeals are more effective than hope appeals to reduce intention to indulge. There is no difference in exercising intentions between fear and hope appeals.
Latimer et al. (2008)	Gain vs Loss vs Mixed Frame	Exercise Behavior	Using gain-framed messages exclusively can increase the efficacy of exercise messages.
Lee and Aaker (2004)	Gain vs Loss Frame Promotion vs Prevention Focus	Brand Attitude (Different Health-related Brands)	Appeals presented in gain frames are more persuasive when the message is promotion focused. Loss-framed appeals are more persuasive when the message is prevention focused.
Mann et al. (2004)	Gain vs Loss Frame	Dental Flossing Behavior	When given a loss-framed message, avoidance-oriented people floss more than approach-oriented people. When given a gain-framed message, approach-oriented people floss more than avoidance-oriented people.
McCaul et al. (2002)	Reminder (Gain vs Loss Frame) vs Action Message	Vaccination Rates	Different framing is no more effective than providing a simple reminder. Action instructions has a better incremental effect on vaccination rates.
Meyerowitz and Chaiken (1987)	Positive vs Negative Frame	BSE Attitude, Intention, and Behavior	Negative-framed messages are more effective than positive-framed messages on women to engage in BSE.
Nisson and Earl (2016)	Approach vs Avoidance Frame Action vs Inaction Frame	Consumption of Healthy and Unhealthy Snack	Individuals viewing active-approach messages consume more food in general than those viewing active-avoidance messages. However, there is no difference in consumption for those viewing inactive messages.
O’Keefe and Jensen (2009)	Gain vs Loss Frame	Meta-analysis	Loss-framed appeals are only slightly, but statistically significantly, more persuasive than gain-framed appeals in promoting disease detection behavior.
Riet et al. (2010)	Gain vs Loss Frame Self-efficacy	Salt Consumption	Loss-framed messages are more effective to persuade people to decrease salt intake than gain-framed messages, but only when people have a high self-efficacy.
Rothman et al. (1993)	Positive vs Negative Frame	Study 1: Intention to Perform Skin Cancer Detection Study 2: Sunscreen Choice	Study 1: Negative-framed messages are more effective for women (high-involvement condition), whereas positive-framed messages are more effective for men (low-involvement condition). Study 2: More women request sunscreens with high SPF protection, while there is no significant difference among men.
Sherman et al. (2006)	Gain vs Loss Frame	Flossing Efficacy, Intention to Floss, Flossing Behavior	Health messages framed to be congruent with individuals’ approach/avoidance motivations are more effective in promoting health behaviors than health messages incongruent with approach/avoidance motivations.
Smith and Stutts (2003)	Long-term Health Fear vs Short-term Cosmetic Fear Appeal	Smoking Behavior	Average smoking declines when people are exposed to any type of antismoking fear appeals (vs Control). Short-term cosmetic fear appeals are more effective for men, but long-term health fear appeals are more effective for women.
Tykocinskl et al. (1994)	Positive vs Negative Frame	(Healthy) Habit of Having Breakfast	Self-discrepancy: (1) AO is the discrepancy between one’s actual self-concepts and the concepts one ought to have; (2) AI is the discrepancy between one’s actual self-concepts and the concepts one hopes to have. Positive-frame messages are more effective for AO people, whereas negative-frame messages are more effective for AI people.
Uskul et al. (2009)	Gain vs Loss Frame	Intention to Floss	British Caucasians (with a stronger promotion focus) are more persuaded by the gain-framed message. East Asian people (with a stronger prevention focus) are more persuaded by the loss-framed message.
Witte and Allen (2000)	Fear	Meta-analysis	Strong fear appeals and high-efficacy messages induce the greatest behavior change. Strong fear appeals with low-efficacy messages induce the greatest levels of defensive responses.
Zhao and Pechmann (2007)	Positive vs Negative Frame Promotion vs Prevention Focus	Intention Not to Smoke	For promotion-focused adolescents, a promotion-focused positively framed antismoking message is the most effective at persuading them not to smoke. For prevention-focused adolescents, a prevention-focused negatively framed antismoking message is the most effective.
Zhao et al. (2015)	Temporal (Long term vs Short term) Frame	Perceived Effectiveness of Smoking Warning Label	Among non-smokers, those high in CFCs respond more favorably to long-term framing, whereas those low in CFC respond more positively to short-term framing. Among smokers, short-term framing is more effective among high-CFC smokers, whereas among low-CFC smokers, the framing effect is not distinct.

MMR: Measles, Mumps and Rubella; STD: sexually transmitted disease; NFC: need for cognition; SPF: sun protection factor; BSE: breast self-examination; AO: actual: ought; AI: actual: ideal; CFC: consideration of future consequence.

The terms gain (vs loss) message framing and positive (vs negative) message framing have been used interchangeably in the literature.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Part of this research was funded through a research grant from the UNSW Business School.

ORCID iDs

Nitika Garg

Rahul Govind

Final transcript accepted 3 January 2021 by Andrew Jackson (Editor-in-Chief)

Notes

References

Abhyankar

O’connor

Lawton

(2008) The role of message framing in promoting MMR vaccination: Evidence of a loss-frame advantage. Psychology, Health and Medicine 13: 1–16.

Agrawal

Duhachek

(2010) Emotional compatibility and the effectiveness of antidrinking messages: A defensive processing perspective on shame and guilt. Journal of Marketing Research 47: 263–273.

Agrawal

Menon

Aaker

(2007) Getting emotional about health. Journal of Marketing Research 44: 100–113.

Ailawadi

Neslin

(1998) The effect of promotion on consumption: Buying more and consuming it faster. Journal of Marketing Research 35: 390–398.

Albarracin

Cohen

Kumkale

(2003) When communications collide with recipients’ actions: Effects of post-message behavior on intentions to follow the message recommendation. Personality and Social Psychology Bulletin 29: 834–845.

Apanovitch

McCarthy

Salovey

(2003) Using message framing to motivate HIV testing among low-income, ethnic minority women. Health Psychology 22: 60–67.

Armitage

Conner

(2001) Efficacy of the theory of planned behavior: A meta-analytic review. British Journal of Social Psychology 40: 471–499.

Arora

(2000) Message framing and credibility: Application in dental services. Health Marketing Quarterly 18: 29–44.

Arora

Stoner

Arora

(2006) Using framing and credibility to incorporate exercise and fitness in individuals’ lifestyle. Journal of Consumer Marketing 23: 199–207.

10.

Australian Institute of Health and Welfare (2018) How much does Australia spend on health care? Australia’s Health, Series No. 16, Chapter 2. Canberra: AIHW.

11.

Bannon

Schwartz

(2006) Impact of nutrition messages on children’s food choice: Pilot study. Appetite 46: 124–129.

12.

Bartels

Kelly

Rothman

(2010) Moving beyond the function of the health behaviour: The effect of message frame on behavioural decision-making. Psychology and Health 25: 821–838.

13.

Bass

Clarke

(1972) Testing distributed lag models of advertising effect. Journal of Marketing Research 9: 298–308.

14.

Batra

Strecher

Keller

(2010) Leveraging Consumer Psychology for Effective Health Communications: The Obesity Challenge. New York: M.E. Sharpe.

15.

Berns

Laibson

Loewenstein

(2007) Intertemporal choice—Toward an integrative framework. Trends in Cognitive Sciences 11: 482–488.

16.

Berry

Carson

(2010) Ease of imagination, message framing, and physical activity messages. British Journal of Health Psychology 15: 197–211.

17.

Bertolotti

Chirchiglia

Catellani

(2016) Promoting change in meat consumption among the elderly: Factual and prefactual framing of health and well-being. Appetite 106: 37–47.

18.

Block

Keller

(1995) When to accentuate the negative: The effects of perceived efficacy and message framing on intentions to perform a health-related behavior. Journal of Marketing Research 32: 192–203.

19.

Block

Grier

Childers

, et al. (2011) From nutrients to nurturance: A conceptual introduction to food well-being. Journal of Public Policy & Marketing 30: 5–13.

20.

Bosone

Martinez

Kalampalikis

(2015) When the model fits the frame: The impact of regulatory fit on efficacy appraisal and persuasion in health communication. Personality and Social Psychology Bulletin 41: 526–539.

21.

Box

GEP

Jenkin

(1970) Time-series Analysis Forecasting and Control. San Francisco, CA: Holden Day.

22.

Brennan

Dahl

Eagle

(2010) Persuading young consumers to make healthy nutritional decisions. Journal of Marketing Management 26: 635–655.

23.

Brick

McCully

Updegraff

, et al. (2016) Impact of cultural exposure and message framing on oral health behavior: Exploring the role of message memory. Medical Decision Making 36: 834–843.

24.

Broemer

(2002) Relative effectiveness of differently framed health messages: The influence of ambivalence. European Journal of Social Psychology 32: 685–703.

25.

Broemer

(2004) Ease of imagination moderates reactions to differently framed health messages. European Journal of Social Psychology 34: 103–119.

26.

Brug

Ruiter

Van Assema

(2003) The (ir)relevance of framing nutrition education messages. Nutrition and Health 17: 9–20.

27.

Chang

(2010) Message framing and interpersonal orientation at cultural and individual levels: Involvement as a moderator. International Journal of Advertising 29: 765–794.

28.

Chang

(2007) Health-care product advertising: The influences of message framing and perceived product characteristics. Psychology and Marketing 24: 143–169.

29.

Charry

Demoulin

(2012) Behavioural evidence for the effectiveness of threat appeals in the promotion of healthy food to children. International Journal of Advertising 31: 773–794.

30.

Cheater

(2020) The obesity epidemic: Time for a positive national healthy eating strategy. International Journal of Health Promotion and Education 58: 51–53.

31.

Childers

Haley

Jahns

(2011) Insights into university freshman weight issues and how they make decisions about eating. Journal of Consumer Affairs 45: 306–328.

32.

Cho

Boster

(2008) Effects of gain versus loss frame antidrug ads on adolescents. Journal of Communication 58: 428–446.

33.

Churchill

Good

Pavey

(2014) Promoting the avoidance of high-calorie snacks. The role of temporal message framing and eating self-efficacy. Appetite 80: 131–136.

34.

Clarke

(1976) Econometric measurement of the duration of advertising effect on sales. Journal of Marketing Research 13: 345–357.

35.

Collymore

McDermott

(2016) Evaluating the effects of six alcohol-related message frames on emotions and intentions: The neglected role of disgust. Journal of Health Psychology 21: 1907–1917.

36.

Conner

Norman

Bell

(2002) The theory of planned behaviour and healthy eating. Health Psychology 21: 194–201.

37.

Cox

(2001) Communicating the consequences of early detection: The role of evidence and framing. Journal of Marketing 65: 91–103.

38.

Detweiler

Bedell

Salovey

, et al. (1999) Message framing and sunscreen use: Gain-framed messages motivate beach-goers. Health Psychology 18: 189–196.

39.

Dibb

Carrigan

(2013) Social marketing transformed: Kotler, Polonsky and Hastings reflect on social marketing in a period of social change. European Journal of Marketing 47: 1376–1398.

40.

Dijkstra

De Vries

Roijackers

(1999) Targeting smokers with low readiness to change with tailored and nontailored self-help materials. Preventive Medicine 28: 203–211.

41.

Duhachek

Agrawal

Han

(2012) Guilt versus shame: Coping, fluency, and framing in the effectiveness of responsible drinking messages. Journal of Marketing Research 49: 928–941.

42.

Frederick

Loewenstein

O’Donoghue

(2002) Time discounting and time preference: A critical review. Journal of Economic Literature 40: 351–401.

43.

Gallagher

Updegraff

(2011) When ‘fit’ leads to fit, and when ‘fit’ leads to fat: How message framing and intrinsic vs. extrinsic exercise outcomes interact in promoting physical activity. Psychology & Health 26: 819–834.

44.

Gallagher

Updegraff

(2012) Health message framing effects on attitudes, intentions, and behavior: A meta-analytic review. Annals of Behavioral Medicine 43: 101–116.

45.

Garg

Wansink

Inman

(2007) The influence of incidental affect on consumers’ food intake. Journal of Marketing 71: 194–206.

46.

Gerend

Maner

(2011) Fear, anger, fruits, and veggies: Interactive effects of emotion and message framing on health behavior. Health Psychology 30: 420–423.

47.

Godin

Kok

(1996) The theory of planned behavior: A review of its applications to health-related behaviors. American Journal of Health Promotion 11: 87–98.

48.

Godinho

Alvarez

Lima

(2016) Emphasizing the losses or the gains: Comparing situational and individual moderators of framed messages to promote fruit and vegetable intake. Appetite 96: 416–425.

49.

Goodall

Appiah

(2008) Adolescents’ perceptions of Canadian cigarette package warning labels: Investigating the effects of message framing. Health Communication 23: 117–127.

50.

Govind

Garg

Mittal

(2020) Weather, affect, and preference for hedonic products: The moderating role of gender. Journal of Marketing Research. Epub ahead of print 10 June. DOI: 10.1177/0022243720925764.

51.

Green

Peloza

(2014) Finding the right shade of green: The effect of advertising appeal type on environmentally friendly consumption. Journal of Advertising 43: 128–141.

52.

Guare

Shulze

(2001) Weight loss recidivism and relapse prevention. In: Gumbiner

(ed.) Obesity. Philadelphia, PA: American College of Physicians, pp. 295–303.

53.

Hanssens

Parsons

Schultz

(2001) Market Response Models: Econometric and Time Series Analysis. Norwell, MA: Springer Science & Business Media.

54.

Hastings

(2007) Social Marketing: Why Should the Devil Have All the Best Tunes? Oxford: Butterworth-Heinemann.

55.

Hoffner

(2009) Young adults’ responses to news about sunscreen and skin cancer: The role of framing and social comparison. Health Communication 24: 189–198.

56.

Howlett

Burton

Bates

, et al. (2009) Coming to a restaurant near you? Potential consumer responses to nutrition information disclosure on menus. Journal of Consumer Research 36: 494–503.

57.

Hughner

Maher

(2006) Factors that influence parental food purchases for children: Implications for dietary health. Journal of Marketing Management 22: 929–954.

58.

Jones

Sinclair

Courneya

(2003) The effects of source credibility and message framing on exercise intentions, behaviors, and attitudes: An integration of the elaboration likelihood model and prospect theory. Journal of Applied Social Psychology 33: 179–196.

59.

Jones

Sinclair

Rhodes

, et al. (2004) Promoting exercise behaviour: An integration of persuasion theories and the theory of planned behaviour. British Journal of Health Psychology 9: 505–521.

60.

Keller

(2006) Regulatory focus and efficacy of health messages. Journal of Consumer Research 33: 109–114.

61.

Keller

Lehmann

(2008) Designing effective health communications: A meta-analysis. Journal of Public Policy & Marketing 27: 117–130.

62.

Keller

Lipkus

Rimer

(2003) Affect, framing, and persuasion. Journal of Marketing Research 40: 54–64.

63.

Khare

Inman

(2006) Habitual behavior in American eating patterns: The role of meal occasions. Journal of Consumer Research 32: 567–575.

64.

Khare

Inman

(2009) Daily, week-part, and holiday patterns in consumers’ caloric intake. Journal of Public Policy & Marketing 28: 234–252.

65.

Kim

Cheong

Zheng

(2009) The current practices in food advertising: The usage and effectiveness of different advertising claims. International Journal of Advertising 28: 527–553.

66.

Koyck

(1954) Distributed Lags and Investment Analysis, vol. 4. Amsterdam: North-Holland Publishing Company.

67.

Krishen

Bui

(2015) Fear advertisements: Influencing consumers to make better health decisions. International Journal of Advertising 34: 533–548.

68.

Latimer

Rench

Rivers

, et al.(2008) Promoting participation in physical activity using framed messages: An application of prospect theory. British Journal of Health Psychology 13: 659–681.

69.

Lee

Aaker

(2004) Bringing the frame into focus: The influence of regulatory fit on processing fluency and persuasion. Journal of Personality and Social Psychology 86: 205–218.

70.

Loewenstein

Prelec

(1992) Anomalies in intertemporal choice: Evidence and an interpretation. The Quarterly Journal of Economics 107: 573–597.

71.

Loewenstein

Thaler

(1989) Anomalies: Intertemporal choice. Journal of Economic Perspectives 3: 181–193.

72.

Mann

Sherman

Updegraff

(2004) Dispositional motivations and message framing: A test of the congruency hypothesis in college students. Health Psychology 23: 330–334.

73.

McCaul

Johnson

Rothman

(2002) The effects of framing and action instructions on whether older adults obtain flu shots. Health Psychology 21: 624–628.

74.

McCleary

Hay

(1980) Applied Time Series Analysis. Beverly Hills, CA; London: Sage.

75.

McDermott

Stead

Hastings

(2005) What is and what is not social marketing: The challenge of reviewing the evidence. Journal of Marketing Management 21: 545–553.

76.

Menon

Block

Ramanathan

(2002) We’re at as much risk as we are led to believe: Effects of message cues on judgments of health risk. Journal of Consumer Research 28: 533–549.

77.

Meyerowitz

Chaiken

(1987) The effect of message framing on breast self-examination attitudes, intentions, and behavior. Journal of Personality and Social Psychology 52: 500–510.

78.

Michaelidou

Dibb

Ali

(2008) The effect of health, cosmetic and social antismoking information themes on adolescents’ beliefs about smoking. International Journal of Advertising 27: 235–250.

79.

Moriarty

(1985) Transfer function analysis of the relationship between advertising and sales: A synthesis of prior research. Journal of Business Research 13: 247–257.

80.

Muehling

Laczniak

Stoltman

(1991) The moderating effects of ad message involvement: A reassessment. Journal of Advertising 20: 29–38.

81.

NCD Risk Factor Collaboration (RFC) (2016) Trends in adult body-mass index in 200 countries from 1975 to 2014: A pooled analysis of 1698 population-based measurement studies with 19·2 million participants. The Lancet 387: 1377–1396.

82.

Nisson

Earl

(2016) Regulating food consumption: Action messages can help or hurt. Appetite 107: 280–284.

83.

O’Keefe

Jensen

(2009) The relative persuasiveness of gain-framed and loss-framed messages for encouraging disease detection behaviors: A meta-analytic review. Journal of Communication 59: 296–316.

84.

Przywara

(2010) Projecting future health care expenditure at European level: Drivers, methodology and main results (No. 417). Directorate General Economic and Monetary Affairs (DG ECFIN), European Commission. Available at: https://ec.europa.eu/economy_finance/publications/economic_paper/2010/pdf/ecp417_en.pdf

85.

Riet

JVT

Ruite

Smerecnik

, et al. (2010) Examining the influence of self-efficacy on message-framing effects: Reducing salt consumption in the general population. Basic and Applied Social Psychology 32: 165–172.

86.

Rothman

Salovey

(1997) Shaping perceptions to motivate healthy behavior: The role of message framing. Psychological Bulletin 121: 3–19.

87.

Rothman

Salovey

Antone

, et al. (1993) The influence of message framing on intentions to perform health behaviors. Journal of Experimental Social Psychology 29: 408–433.

88.

Rothschild

(2010) Bringing a bit of social marketing to the problem of obesity. In: Batra

Strecher

Keller

(eds) Leveraging Consumer Psychology for Effective Health Communications: The Obesity Challenge. Armonk, NY: M. E. Sharpe, pp. 307–318.

89.

Seiders

Petty

(2004) Obesity and the role of food marketing: A policy analysis of issues and remedies. Journal of Public Policy and Marketing 23: 153–169.

90.

Sheeran

(2002) Intention-behaviour relations: A conceptual and empirical review. In: Hewstone

Stroebe

(eds) European Review of Social Psychology, Vol. 12. London: John Wiley & Sons, pp. 1–36.

91.

Sheeran

Orbell

(2000) Using implementation intentions to increase attendance for cervical cancer screening. Health Psychology 19: 283–289.

92.

Sherman

Mann

Updegraff

(2006) Approach/avoidance motivation, message framing, and health behavior: Understanding the congruency effect. Motivation and Emotion 30: 164–168.

93.

Smith

(1996) Framing in advertising and the moderating impact of consumer education. Journal of Advertising Research 36: 49–49.

94.

Smith

Stutts

(2003) Effects of short-term cosmetic versus long-term health fear appeals in anti-smoking advertisements on the smoking behaviour of adolescents. Journal of Consumer Behaviour 3: 157–177.

95.

Tykocinskl

Higgins

Chaiken

(1994) Message framing, self-discrepancies, and yielding to persuasive messages: The motivational significance of psychological situations. Personality and Social Psychology Bulletin 20: 107–115.

96.

Uskul

Sherman

Fitzgibbon

(2009) The cultural congruency effect: Culture, regulatory focus, and the effectiveness of gain-vs. loss-framed health messages. Journal of Experimental Social Psychology 45: 535–541.

97.

Veer

Tutty

Willemse

(2008) It’s time to quit: Using advertising to encourage smoking cessation. Journal of Strategic Marketing 16: 315–325.

98.

Wansink

(1996) Can package size accelerate usage volume? Journal of Marketing 60: 1–14.

99.

Wansink

(2006) Nutritional gatekeepers and the 72% solution. Journal of the American Dietetic Association 106: 1324–1327.

100.

Wansink

Ray

(1996) Advertising strategies to increase usage frequency. Journal of Marketing 60: 31–46.

101.

Wansink

Cheney

Chan

(2003) Exploring comfort food preferences across age and gender. Physiology & Behavior 79: 739–747.

102.

Ward

Mann

(2000) Don’t mind if I do: disinhibited eating under cognitive load. Journal of Personality and Social Psychology 78: 753–763.

103.

Weinstein

Sandman

Roberts

(1991) Perceived susceptibility and self-protective behavior: A field experiment to encourage home radon testing. Health Psychology 10: 25–33.

104.

Weiss

Windal

(1980) Testing cumulative advertising effects: A comment on methodology. Journal of Marketing Research 17: 371–378.

105.

Witte

Allen

(2000) A meta-analysis of fear appeals: Implications for effective public health campaigns. Health Education & Behavior 27: 591–615.

106.

World Health Organization (WHO) (2015) Obesity and overweight. Fact Sheet No. 311. Available at: http://www.who.int/mediacentre/factsheets/fs311/en/ (accessed 4 February 2015).

107.

Young

(2003) Does food advertising influence children’s food choices? A critical review of some of the recent literature. International Journal of Advertising 22: 441–459.

108.

Zhang

Buda

(1999) Moderating effects of need for cognition on responses to positively versus negatively framed advertising messages. Journal of Advertising 28: 1–15.

109.

Zhao

Pechmann

(2007) The impact of regulatory focus on adolescents’ response to antismoking advertising campaigns. Journal of Marketing Research 44: 671–687.

110.

Zhao

Nan

Iles

, et al. (2015) Temporal framing and consideration of future consequences: Effects on smokers’ and at-risk nonsmokers’ responses to cigarette health warnings. Health Communication 30: 175–185.