'You may have a cancer-causing virus and not even know it' Fear appeals in online news
First Monday

You may have a cancer-causing virus and not even know it Fear appeals in online news by Brad Love and Michael S. Mackert



Abstract
Most explanations of health risks are presented as “fear appeals,” messages that attempt to arouse fear to modify personal behavior. The central goal of this research is to assess fear appeal performance in online news while examining several research–design issues. Looking at digital news content as fear appeals will expand the scope of research into a new area, and more effective measurement and induction will avoid some common pitfalls.

An online experiment produced data indicating that perceived threat and efficacy relationships developed as expected in online news and that improvements in research–design artifacts strongly affect observed variable relationships. Fear appeal and efficacy messages included in digital media content seem to be effective means to altering individual behavioral intentions.

Contents

Introduction
Literature review
Hypotheses
Method
Results
Discussion
Conclusion

 


 

Introduction

Digital media are now central to how individuals receive health information (Hodgetts and Chamberlain, 2006; Moriarty and Stryker, 2008; Fox and Brenner, 2012), and many of these mediated discussions of health issues are presented as “fear appeals” (Witte, et al., 2001), messages that attempt to arouse fear to get individuals to behave a certain way. Fear appeals persuade by initially exploiting emotions to imply or state that not following some behavior will lead to a negative situation such as illness, financial loss, or embarrassment.

One useful tool in studying fear appeals is the Extended Parallel Process Model (EPPM), which predicts when persuasive messages will be effective — and not — at directing change in individuals (Witte and Allen, 2000). The EPPM predicts that when individuals feel empowered to make change, they will be more likely to take action to remove a problem. However, when feelings of fear are greater than empowerment, individuals will ignore the threat and remove relevant thoughts or deny the problem.

Mass media content, particularly health news, regularly contains fear–inducing and empowerment messages. Media outlets often use threatening messages to induce fear and get attention through page views, for example. At the same time, media outlets also include empowerment– and efficacy–promoting messages to help viewers take protective action.

Despite ample research on fear appeal messages, few studies examine messages in the context of mass media (Hong, 2011), and fewer still examine digital media. A key objective of this paper is to test digital news as a context for the fear and efficacy messages so central to fear appeal research.

A second objective is to study methodological issues regarding the variables key to the EPPM and fear–appeal research. Building from meta–analyses by Boster and Mongeau (1984), as well as Witte and Allen (1990), this research expands the range of induction conditions by adding digital media and improves variable measurement.

Thus, this research aims to: (1) improve understanding of key fear appeal variables by expanding their applied context into digital media; and, (2) address methodological issues that may have influenced previous findings. It is not a test of the EPPM theory in line with intervention–focused reports. Instead, it serves as a call to augment how future fear appeal research is applied and measured in the digital world. The rest of this paper reviews relevant literature on fear appeal research, methods including changes in measurement and inductions, results, and a discussion of the implications for researchers and practitioners.

 

++++++++++

Literature review

Fear appeals and the Extended Parallel Process Model

Fear–appeal proponents argue most “logical and reasoned messages fall on deaf ears” [1] because individuals do not make decisions about risk issues on an entirely rational basis (Weinstein, 2001). Using emotions to get attention can improve exposure to information and affect behavior change (Morrongiello, et al., 2009) because an effective message must engage both viewer attention and processing (Grant Harrington, et al., 2006). At its simplest, any fear appeal contains two basic components, a threat and a suggested response (Witte, et al., 2001). The threat describes negative consequences, and the response specifies how to avoid the threat. These types of arguments can be found in many places from doctors talking to patients (“Stop smoking, or you’ll develop lung cancer.”) to businesses soliciting customers (“Buy our software to prevent computer viruses.”) (Witte, et al., 2001).

Not every fear appeal is successful, of course. What makes for effective use of fear appeals and what leads to failure has been the debate of much scholarship during the last five decades with few well–supported conclusions (LaTour, et al., 1996). A pair of effects have been consistent and reported regularly, however–perceived threat and perceived efficacy both influence people’s intentions (Sutton, 1982).

A consistent connection exists between perceived threat and following recommended actions (Sutton, 1982; Boster and Mongeau, 1984; Witte and Allen, 2000). Perceived threat is an individual’s belief that a danger or harm is possible. It is measured through two dimensions: perceived susceptibility and perceived severity (Witte, et al., 2001). The second prominent effect links perceived efficacy and adherence to suggested actions (Witte and Allen, 2000). Perceived efficacy and is measured through response efficacy (perceptions of effectiveness of the promoted action) and self–efficacy (individual beliefs about the ability to complete the proposed response) (Witte, et al., 2001). In general, perceived efficacy shows stronger links to following suggested actions than perceived threat.

One means of incorporating these ideas is Witte’s (1992b) EPPM, which works to improve the theoretical understanding of how fear appeals influence behavior. The EPPM explains the relationships among efficacy, threat, and outcomes, building upon older models to predict the successes and failures of fear appeals while re–integrating fear as a central variable (Witte and Allen, 2000). The EPPM — following the parallel response literature — predicts individuals use one of two paths to reduce fear when they perceive a threat: one to control the actual danger causing the threat, the other to control the emotional fear produced by the threat (Witte, et al., 1996). The first results in action intended to alleviate the problem (removing the danger); the second (fear control) only removes thoughts or recognition of the threat.

The primary application of the EPPM has been in theoretically guided health campaigns (Stephenson and Witte, 2001). The model has been tested in multiple contexts, including breast cancer (Kline and Mattson, 2000), cardiovascular disease (McKay, et al., 2004), HIV/AIDS prevention (Witte, 1992a; Witte, et al., 1998), and skin cancer (Stephenson and Witte, 2001) with populations segmented by education level (Gore and Bracken, 2005), ethnicity (Murray, et al., 1998), occupation (Witte, et al., 1998), and geography (Witte, et al., 1998). Data across the range of studies are consistent with the EPPM, according to Witte and Allen’s (2000) review.

Perceived efficacy in fear appeals

Perceived efficacy is the key factor determining which path an individual follow — danger control to remove the threat or fear control to ignore it (Witte and Allen, 2000). To get individuals on the danger control track where they take action against a threat, perceived efficacy must be greater than perceived threat. The audience must believe “they are easily, feasibly, and effectively able to avert a serious and relevant threat from occurring by adopting the recommended response.” [2]

When the efficacy element does not overpower the initial fear appeal, individuals opt to not adopt a suggested response to remove threat either because they feel incapable of completing the protective action or because the suggested action will not be effective. Rather than concerning themselves with eliminating the threat, individuals in the fear control process put their efforts into avoiding thoughts about the threat or minimizing the issue (Brehm, 1966). Combining low perceived efficacy and a high–perceived threat results in fear control (Witte, et al., 1996). An inverse relationship between efficacy messages and fear control responses (r = -.11) supports the hypothesis that a weak efficacy message leads to fear control responses (Witte and Allen, 2000). Raising efficacy is a key component of creating a desirable, action–oriented danger control response of self–protective action.

Perceived fear in fear appeals

As fear appeals get stronger, greater threat drives action, and high–threat messages result in larger effects than low–threat messages, or “the stronger the threat in a message, the more motivated individuals appear to be to process the message” in some way [3]. Also, fear control responses are inversely linked (r = -.18) with danger control reactions, matching the EPPM prediction that those defensively avoiding a recommendation will not be inclined to follow a message’s suggestions (Witte and Allen, 2001).

Fear appeals and mass media

Fear appeal research has focused almost exclusively on educational materials — brochures, posters, and videos — designed explicitly to communicate health information. The usage of media content such as news articles or TV news stories has been rare (Hong, 2011). However, health news content through media channels often serves a key information–providing purpose (Nell, 2002) and does so in an emotionally arousing way to maximize attention (Grabe, et al., 2000; Leask and Chapman, 2002; Dunlop, et al., 2008). As such, fear appeal research generally fails to address what may be the most common source of fear appeals, considering most Americans read online news consistently (Edmonds, et al., 2012).

Studying media content such as online news expands fear appeal experiments into a relatively unexplored area, one that readers potentially view differently than health education. Materials such as brochures have an obvious agenda regarding reader behavior, which can impact responses. Other media channels such as news outlets do not necessarily share the same biases, as industry standards expect information to be objective instead of persuasive (Schudson, 2001; Niven, 2003).

This research includes digitally presented text because such content plays an essential role in real–life behavior–modification campaigns. In the context of unmatchable corporate advertising budgets, non–profit organizations depend upon media content driven by health communication efforts to expand message reach. Through this avenue of independent content, organizations seeking positive health change can earn attention equivalent to paid–for messages, particularly in a media environment where online content is so easily shared. Additionally, the news context is appropriate because the opening paragraph of much content uses fear appeals to gain attention for the story as a whole or because the content–producer feels a fear message is the best way to get audience attention (Hofstetter and Dozier, 1986; Grabe, et al., 2000).

Results from this research should help health communication professionals and healthcare practitioners understand better message development to change health behavior. Knowing people’s likely responses to behavior–change messages will help produce media content to more effectively target audiences. While previous findings might be extended to mass media content, research to empirically verify that extension has been absent.

Fear appeal research designs

In examining fear appeal research, Boster and Mongeau’s (1984) meta–analysis described several noteworthy design artifacts. One artifact involves how the number of fear induction conditions relates to outcomes; in earlier fear appeal research with attitude as the dependent variable, studies with more levels of fear manipulation reported larger fear manipulation–attitude correlations (r=.31, df=23, p=.07 in Boster and Mongeau, 1984).

In addition to factors involving the number of fear appeal messages, the number of items used to measure the outcome variable and the fear manipulation–outcome correlation are positively related (r=.36, df=21, p=.05 in Boster and Mongeau, 1984 for fear manipulation–attitude). Looking at the same issues through another method, Boster and Mongeau (1984) also found standardized regression coefficients even larger than the zero–order correlations when the fear manipulation–attitude relationship was regressed onto type of experiment design and number of items in the measurement instrument (Β=-.24 for type of experiment and Β=.39 for number of items, R=.43).

Study design type, number of measurement items, and number of fear induction conditions historically have been responsible for part of the statistical relationship between fear induction messages and outcome variables (R=.50 in the multiple correlation examination by Boster and Mongeau, 1984, for methodological artifacts within the fear manipulation–dependent variable).

 

++++++++++

Hypotheses

The hypotheses examine the suitability of perceived efficacy and perceived threat to perform in a mediated, text–based news context. In Witte and Allen’s (2000) meta–analysis, response efficacy and self–efficacy produce correlations with the dependent variables (attitude, behavioral intention, behavior change) ranging from .12 to .17 [4]. Perceived efficacy and behavioral intention were correlated at .17.

To demonstrate that mediated news content can serve as a suitable medium for the application of this essential EPPM variable, H1 predicts the simulated online news articles employed here will induce a perceived efficacy–behavioral intention correlation (r=.17) that matches or exceeds the effects found in fear–appeal–focused meta–analyses.

H1: Higher perceived efficacy will associate positively with higher behavioral intention to take self–protective action at a level equal to or greater than prior research.

Similarly, the range of meta–analyses show stronger fear appeals lead to greater persuasion toward danger control responses (Witte and Allen, 2000). The strength of fear manipulation–dependent variable relationships remains relatively consistent across meta–analyses (Boster and Mongeau, 1984; Mongeau, 1998; Witte and Allen, 2000). To demonstrate perceived threat can perform as predicted within a mediated news context, H2 predicts the articles used here will induce a perceived threat–behavioral intention correlation in line with effects observed in prior research, specifically at .11 or greater.

H2: Higher perceived threat will associate positively with behavioral intention to take self–protective action at a level equal to or greater than prior research.

Another key test of the suitability of text–based, digital news content is its ability to induce differing levels of perceived threat. If digital news content is to serve as a suitable medium for persuasive appeals, it must be able to induce different levels of perceived threat.

H3: Threat perception will vary across three treatment conditions in line with expectations for the constructed news articles focusing on cancer, viruses, and medical research.

Additionally, it is worth noting prior meta–analytic work has discussed a number of research design artifacts that could significantly influence the fear induction–dependent variable relationship as well as the efficacy–dependent variable correlation (Boster and Mongeau, 1984). Specifically, the number of treatment conditions and the number of items used to measure the dependent variable affect the strength of the resulting correlations, due in part to restriction in range from little variance among treatment conditions and measurement error.

This study also serves to examine the effects of increasing the number of treatment conditions from the common low–fear/high–fear dichotomous induction design. Furthermore, this research employs a previously validated, six–item measure of behavioral intention in place of the often brief, less reliable dependent variable measures discussed in the meta–analyses by Boster and Mongeau (1984) and Witte and Allen (2000). The objective with these changes is to demonstrate an increased number of treatment conditions and improved measurement of key variables will correct for some attenuation in variable relationships.

 

++++++++++

Method

Participants were exposed to one of five simulated text–based, online articles intended to invoke a level of threat around a health topic ranging from high to non–existent. The three subject areas were:

  • cancer (breast cancer for females and prostate cancer for males)
  • viruses (human Papillomavirus or food poisoning)
  • clinical trials

Multiple articles were employed to promote variance among participant responses to help reduce attenuation in key correlations from restriction in range. Restriction in range from little variance in both the number of treatment conditions and responses to the treatments can attenuate the relationship between the fear manipulation and the outcome variable (Boster and Mongeau, 1984). The restriction in range from minimal treatment conditions may result in an improperly estimated correlation of the relationship between the appeal message and the dependent variable. A correlation based only on extreme reactions may not be accurate to how people respond to fear–appeal messages in real life, particularly the wide–ranging messages in mass media channels.

Placing the messages within digital news articles opens the context of fear appeal research from prior work on educational materials, which rarely match how individuals access health–related information. Well more than half of all Americans read daily text–based news in some form (Edmonds, et al., 2012), making the medium a logical environment for studying information processing.

Using messages with different levels of perceived fear will also eliminate message topic as a possible explanation for effects. Additionally, to improve design and measurement from prior fear appeal research, messages and instruments were extensively pilot tested. The pilot test sample was an international convenience sample from an online community used for science fiction literature.

Online samples were employed to broaden the sample from typical fear–appeal research on students or those living in high–risk environments such as areas with elevated disease rates. In other contexts, Internet–based samples have been used to approximate other data collection methods including mail or in–person surveys (Hewson, et al., 2003; Leece, et al., 2004; Baker, et al., 2007).

Stimuli material design according to the EPPM

The simulated news articles addressed the essential components of fear appeal messages. According to the EPPM, a message must have a component addressing each of the following perceived areas: susceptibility, severity, self–efficacy, and response efficacy. A statement of the reader’s susceptibility to the health problem and one concerning the severity of the health problem made up the threat component intended to produce a reaction.

The opening sentences focused on severity and susceptibility, following newswriting norms. Opening sentences are intended to grab reader attention and summarize the article’s content while focusing on a news value (Burns, 2002). Efficacy components were also incorporated into each article, focusing on preventative behavior and intentions to gather further information. The efficacy components served to aid in the development of behavioral intention measures and complete the EPPM model requirements. Each article followed the same format on a paragraph–by–paragraph basis.

Multiple articles were employed to provide a significant amount of variance across levels of perceived threat, health topics, and respondent reactions. Based on initial planning and confirmed by pilot testing, the five articles in the final data collection were intended to produce three induction conditions with maximum variance across the susceptibility and severity aspects of threat, one focused on cancer (breast and prostate), one concerning viruses (HPV and food poisoning), and one control group reading about non–threatening medical research.

Article attributes

Each article was manipulated to achieve approximately equal length and reading level. The pieces avoided technical terms or multi–syllabic words to increase readability and match journalistic conventions (Knight, 2003). The resulting five–paragraph articles had a mean length of 262 words, ranging from 233 to 280. The Flesch–Kincaid Grade Level (ninth grade) and Flesch Reading Ease (55.8) measurement scales both indicated easier–to–read texts than actual news content. The ninth–grade reading level ranks equal with mass–market publications such as TV Guide or Time magazine (DuBay, 2005) while a level below front–section stories in the Houston Chronicle (Meyer, 2004), New York Times (Jung, 2003), and Washington Post (DuBay, 2005).

The Flesch Reading Ease score is calculated using average sentence length and number of syllables per 100 words, ranging from 1 to 100. Scores show the experiment’s content slightly easier to read than the Wall Street Journal’s top stories (48.3), USA Today (47), New York Times (36.2) (Jung, 2003), and newspaper editorials generally (48) (Murphy, et al., 1994).

Instrument selection

Measurement instruments come from prior studies in which they have demonstrated accepted levels of reliability and validity. All were pre–tested using a different online sample. In addition to collecting demographic information, the following instruments were used as measurement instruments:

  • Perceived Threat: A two–part survey composed of perceived severity and perceived susceptibility (Witte, et al., 2001)
  • Perceived Efficacy: A two–part survey on perceived self–efficacy and perceived response efficacy (Witte, et al., 2001)
  • Behavioral Intention: An adapted version of an instrument measuring intentions with six items according to the recommended behaviors contained in each article (Wilson and Lankton, 2004)

Data collection

The data collection came from an online experiment using a sample provided by an organization specializing in online surveys. Participants were randomly assigned to one of the message conditions and exposed to only one article. The order of survey questions was consistent. The only change was the treatment article and wording changes in the accompanying items. An example constructed news article and sample survey items are shown in the Appendix.

 

++++++++++

Results

Descriptive statistics

A total of 1,105 usable responses (N=1,105) out of 6,268 e–mail requests with one follow–up were collected in exchange for entry into a prize drawing. Participants’ ages ranged from 19 to 81(M=45). The sample was 69 percent female, and 70 percent of respondents were married or in committed relationships. The sample consisted of 57 percent Americans, 16 percent Canadians, seven percent New Zealanders, and five percent Britons. Individuals reported ethnicity as white 82 percent of the time, Asian nine percent, black four percent, and multi–racial three percent. The online sample was considered appropriate because the project’s concern involved digital media messages. In addition, one project goal was to examine fear appeal messages outside a student sample or individuals in a high fear environment.

To arrive at the 1,105 usable surveys, 205 initial responses were eliminated. Cases were eliminated because of extreme incompleteness and suspicious Internet Protocol addresses (e.g., multiple responses linked to repeat addresses).

In the usable data, maximum likelihood estimation was used to fill in missing data points (Croy and Novins, 2005). Maximum likelihood estimation is considered by statisticians to be superior to most methods of data completion such as listwise deletion, mean substitution, or pairwise deletion (SPSS, 1999). The combined data from all treatment conditions were scrutinized for normality through visual examinations of P–P plots, Q–Q plots, and histograms to ensure data points fell approximately in a straight line, which was the case.

Measurement instrument assessment

The assessment of the measurement instruments occurred in three phases: 1) item analysis; 2) reliability assessment; and, 3) construct validity. Principal Component Analysis verified unidimensional relationships for item analysis. Cronbach’s alpha coefficients for all constructs were higher than the recommended minimum of .70 (Nunnally, 1978). Table 1 presents the alpha coefficients for each index along with descriptive statistics. Construct validity was assessed through convergent and discriminate validity (Cronbach, et al., 1972; Cook and Campbell, 1979). Researchers demonstrate convergent validity with factor loadings on the intended constructs greater than twice the standard errors (Anderson and Gerbing, 1984), as was the case here (.007–.067). Discriminant validity illustrated each measure did not correlate very highly with other measures from which it should differ by inspecting the confidence intervals surrounding each pairwise correlation estimate to ensure the intervals did not contain 1 (Anderson and Gerbing, 1988).

 

Descriptive statistics and Cronbach Alphas for measurement instruments
 
Table 1: Descriptive statistics and Cronbach Alphas for measurement instruments.

 

Comprehension and believability

Another important check on the validity of the data is to consider how well readers understood and believed the article content. Comprehension was measured using four true–false items that were similar across articles and focused on the information common to all articles: frequency of the issue, preventative measures, risk factors, and severity information. Items in this instrument were answered correctly 92 percent of the time. Such a high score supports the idea readers considered the information contained in the articles.

Article believability was measured using four items across seven–point Likert–type response scales. All five groups responded to the same four items. Based upon the item scores and mean across the individual items (5.68 out of 7; standard deviation: 1.2), respondents found the treatment stories to be believable.

Data were combined into a single data set for analysis because the specific message conditions were not a focus in this research. Combining data from different conditions promotes variance in responses, one objective here, and avoids some biases due to subject matter. Similar studies might focus on the message conditions to examine what makes for an effective fear appeal, but the goal of this study was to examine how individuals mentally react and process the information presented in this medium, as opposed to how to best induce effective conditions.

Hypotheses testing

H1: Higher perceived efficacy will associate positively with higher behavioral intention to take self–protective action at a level equal to or greater than prior research.

H1 predicted perceived efficacy would be positively associated with behavioral intention, as predicted in prior fear appeal studies, and do so at a level equal to or greater than existing research. Results show efficacy did have a positive correlation with behavioral intention and that the relationship was greater than indicated in existing meta–analytic work (a correlation of .52 at p=.01, two–tailed, compared to .17 in Witte and Allen, 2000). H1 was supported.

Just as prior research has reported, this outcome fits the idea that increasing your belief in your own abilities can be an effective means of persuasion. That is, if an individual perceives that a recommended behavior will be effective and it is achievable, the person will more likely undertake that behavior.

H2: Higher perceived threat will associate positively with behavioral intention to take self–protective action at a level equal to or greater than prior research.

H2 predicted higher levels of perceived threat would associate with increased levels of behavioral intention to undertake a danger control response. Perceived threat did have a statistically significant relationship with behavioral intention and did so at a level higher than reported in meta–analyses (r=.51 at p=.01, two tailed, compared to r=.11 in Witte and Allen, 2000, which is in line with other perceived threat–dependent variable relationships in Boster and Mongeau, 1984, and Mongeau, 1998). These data are consistent with the predictions of H2.

This result supports the idea that for some individuals, feeling threatened by a particular topic relates to constructive intentions to undertake self–protective behaviors. For example, diabetes could become more threatening to a person as he recognizes risk factors in his life such as advancing age, weight gains, and some blurred vision. As this individual perceives diabetes to be an increasing threat to his well–being, his intention to make behavioral change regarding diabetes care could also increase.

H3: Threat perception will vary across three treatment conditions in line with expectations for the constructed news articles focusing on cancer, viruses, and medical research.

Brown–Forsythe and Welch statistics checked the manipulations of the individual message conditions because of the non–homogeneity of variance and unequal sample sizes among the clinical trial–, virus–, and cancer–message groups. Both manipulation statistics (Brown–Forsythe: 107.62, p<.001; Welch: 88.31, p<.001) were statistically significant, offering evidence that the perceived threat scores for three treatment topics differed according to the intended exposure. The two dimensions of perceived threat were measured by three seven–point Likert items from Witte, et al. (2001). Examining the group means for the perceived threat index also indicates effects in the predicted direction (Mclin trials=20.86, Mvirus=29.39, Mcancer=31. 72). The cancer articles have higher threat scores than the virus articles, which have higher scores than the clinical trials control condition. Thus, H3 was supported by the data.

Additionally, Games–Howell, Dunnett’s T3, and Dunnett’s C provided further analysis to verify differences among the groups — similar to the way other analyses may employ Tukey’s HSD. Tukey’s HSD was not appropriate here because of the violations of parametric analysis’ assumptions that can be expected in an experimental project where sample sizes differ and the induction results in multiple conditions for the treatment. The post hoc Games–Howell, Dunnett’s T3, and Dunnett’s C tests confirm that the treatment groups differ at the p<.001 level for perceived threat scores. The Games–Howell test, in particular, is worth noting because it is the most powerful of the three and is strongly recommended for comparing groups with unequal sample sizes (Toothaker, 1993). The outcomes of the tests are recorded in Table 2 and can be interpreted as evidence of successful manipulation of perceived threat in response to the articles.

 

Post hoc tests for group means on perceived threat index
 
Table 2: Post hoc tests for group means on perceived threat indexa.
a Mean difference is significant at the .05 level.

 

Post hoc analysis

It is worth noting the correlations reported in this study are not only greater than prior meta–analytic research, as predicted, but considerably higher. While research has found fear manipulation–behavior correlations as high as .69 and fear manipulation–attitude correlations of .63 (Boster and Mongeau, 1984), such relationships are not common across the literature; thus, the strong relationships found in this study merit further examination to understand factors possibly driving the correlations.

As discussed earlier, existing work on fear appeals has noted methodological decisions can dramatically affect statistical outcomes in fear–appeal research. Boster and Mongeau (1984) calculated a multiple correlation between methodological artifacts and the fear manipulation–attitude correlations to estimate how much variance in the fear manipulation–attitude relationship is due to type of design, number of items used in measurement, and number of levels of the independent variables. The resulting R equals .50, indicating artifacts drive an extensive amount of the difference in the correlation between fear manipulation and the common dependent variable of attitude.

Similar to those methodological issues that were a central concern in this study, it appears individual experience with the subject matter of the fear appeal articles also drove much respondent reaction, in part explaining the high correlations. About 28 percent of respondents reported experience — whether personal, family, other, or some combination of those — with the condition specific to their induction article. While some exposure was to be expected across the sample considering the frequency of the health issues employed here, the percentages of prior experiences do warrant noting because they suggest the possibility that many respondents were primed to react strongly to the exposure due to pre–existing involvement.

For comparison, when individuals reporting experience with the topic of their exposure article were removed from the data set and correlations were re–calculated on the remaining sample of 800 people, the relationships change substantially. While still higher than reported in comparable meta–analyses (as predicted in H1 and H2), the relationships are not as strong with a perceived threat–behavioral intention correlation of .22 (instead of .51 for the full data set) and an efficacy–behavioral intention correlation of .42 (instead of .52 in the full data set). Both H1 and H2 still have ample support, but the impact of individuals reporting experience is an interesting artifact.

 

++++++++++

Discussion

Health messages are consistently available to the public through a number of media outlets, often in response to publicity campaigns, leaving journalists to act as de facto health educators. The ability of mediated messages to carry out this public health task can be improved by theories and message–design strategies. This research furthers digital media research by contributing theoretically and practically to knowledge of how individuals react to health messages embedded in digital mass media. The stated goals of the project covered two aspects of fear appeal research. First, this paper describes an examination of central fear–appeal variables in a mediated news context to examine a different message environment. Second, it assesses relevant methodological issues.

The first goal of this paper was to examine perceived threat and efficacy in a media context, opening possibilities for examining their induction in a new and forward–looking environment. Prior research employed brochures and posters as educational tools, with a few projects using videos or live seminars, despite the fact that so much of the information individuals receive comes through other media. Testing in the media environment showed that prior outcomes were mirrored in a digital news setting.

Health–focused organizations cannot always deliver their own messages unfiltered, making it essential to consider how the raw materials — often press releases or interview responses — can be structured to influence content in a way that promotes effective health outcomes. Primary materials can work within the reporting process by offering content promoting action–oriented outcomes. Releases and educational materials for reporters must focus on key elements of an effective message. With the EPPM, for example, this means each press release needs a perceived threat and a stronger perceived efficacy element. Web site content for health organizations could follow the same guidelines.

The second goal of this research was to improve the level of measurement and experimental design from prior fear appeal research, which happened in two ways. The first development involved adapting and pilot testing previously used instruments to achieve higher Cronbach’s Alpha reliability scores and reduce some common attenuation in measurement. Reduced alpha scores have been common in this research area. Perceived severity, susceptibility, self–efficacy, and response efficacy have typically reported alphas in the high .7s, for example. The instruments in this project improved their Cronbach Alpha reliabilities almost .10 over their normal results to .89 on average.

The second area of methodological improvement comes from having multiple messages to induce multiple levels of perceived threat, a tactic advocated by Boster and Mongeau (1984). In their work, studies with more levels of fear manipulation reported stronger relationships between perceived threat and the dependent variable. Despite this observation, few projects have moved beyond employing dichotomous low– and high–fear conditions. In this research, the perceived threat–behavioral intention correlation is .51 (p<.01). The comparable correlation in the Witte and Allen meta–analysis (the only one computing these relationships) is .11.

Along similar lines, the perceived efficacy–behavioral intention correlation in this study was significantly greater than in prior research. Witte and Allen (2000) report efficacy correlations with the common dependent variables as being between .12 and .17. Here, the perceived efficacy correlation with the dependent variable is .52 (p<.01). This further supports the previously found strong relationships between perceived efficacy and the common fear appeal dependent variables.

This study has several limitations worth considering in future research. One challenge with fear appeal research is that it is often conducted in experimental settings and assumes that the perception of threat or emotional fear is the predominant outcome of exposure (Kreuter and McClure, 2004).

Another challenge is an individual’s readiness to accept behavior–modification suggestions. No shortage of research has demonstrated that real behavior modification happens over time as individuals move through behavioral and psychosocial stages with differing characteristics (Prochaska, et al., 2002). Presence in a particular stage of behavior change could influence reaction to a fear appeal (Cho and Salmon, 2006). Potentially, as Maibach and Cotton describe (1995), fear appeals could aid in motivating individuals to work through stages of change by strengthening their plans to engage in preventative behavior.

 

++++++++++

Conclusion

This study investigated fear appeals in digital media. We examined the suitability of a commonly used fear appeal theoretical model, the Extended Parallel Process Model, and also tested several options for improved methodology, including more precise variable measurement and broader sampling than in prior research.

In studying these issues, we found that variables essential to fear appeal research — perceived threat and efficacy — maintained their previously documented relationships when employed in online health news. Understanding the opportunities for online use of fear appeal messages can help content producers and public health communicators develop more effective messages.

Additionally, by improving the level of measurement, expanding the number of treatment conditions, and broadening the survey sample, this study demonstrated the impact of methodological issues on variable relationships. Due to pre–testing, alpha reliability for variables improved by .10 over prior uses. Employing more induction levels in the treatment conditions also played a role in finding a .51 perceived threat–behavioral intention correlation, well above the strength of most relationships in the literature. The same is true for the perceived efficacy–behavioral intention correlation at .52.

This research serves as a powerful call to action to explore different contexts where fear appeals are present and can influence health behaviors. As this research demonstrates, there remains a need to further investigate methodological and measurement issues to continue advancing fear appeal research. A better and broader understanding of the mechanisms by which fear appeals influence behavior will benefit researchers and practitioners in a range of fields as well as the general public that increasingly depends upon online media for essential health information. End of article

 

About the authors

Brad Love, Ph.D., and Michael S. Mackert, Ph.D., are faculty members in the Department of Advertising and Public Relations at the University of Texas, based in the Belo Center for New Media. In the department, Love is an assistant professor, and Mackert is an associate professor.
Direct correspondence to: lovebrad [at] utexas [dot] edu

 

Notes

1. Witte, et al., 2001, p. 1.

2. Witte, et al., 1996, p. 318.

3. Witte and Allen, 2001, p. 604.

4. Witte and Allen, 2000, see p. 599 for the correlations between fear, efficacy and common DVs.

 

References

James C. Anderson and David W. Gerbing, 1988. “Structural equation modeling in practice: A review and recommended two–step approach,” Psychological Bulletin, volume 103, number 3, pp. 411–423.
doi: http://dx.doi.org/10.1037/0033-2909.103.3.411, accessed 21 January 2014.

James C. Anderson and David W. Gerbing, 1984. “The effects of sampling error on convergence, improper solutions, and goodness–of–fit indices for maximum likelihood confirmatory factor analysis,” Psychometrika, volume 49, number 2, pp. 155–173.
doi: http://dx.doi.org/10.1007/BF02294170, accessed 21 January 2014.

Jack W. Brehm, 1966. Theory of psychological reactance. New York: Academic Press.

Lynette Sheridan Burns, 2002. Understanding journalism. Thousand Oaks, Calif.: Sage.

Hyunyi Cho and Charles T. Salmon, 2006. “Fear appeals for individuals in different stages of change: Intended and unintended effects and implications on public health campaigns,” Health Communication, volume 20, number 1, pp. 91–99.
doi: http://dx.doi.org/10.1207/s15327027hc2001_9, accessed 21 January 2014.

Thomas D. Cook and Donald T Campbell, 1979. Quasi–experimentation: Design & analysis for field settings. Chicago: Rand McNally.

Lee J. Cronbach, Goldine C. Gleser, Harinder Nanda, and Nageswari Rajaratnam, 1972. The dependability of behavioral measurements: The theory of generalizability for scores and profiles. New York: Wiley.

Calvin D. Croy and Douglas K. Novins, 2005. “Methods for addressing missing data in psychiatric and developmental research,” Journal of the American Academy of Child & Adolescent Psychiatry, volume 44, number 12, pp. 1,230–1,240.
doi: http://dx.doi.org/10.1097/01.chi.0000181044.06337.6f, accessed 21 January 2014.

Sally Dunlop, Melanie Wakefield, and Yoshi Kashima, 2008. “Can you feel it? Negative emotion, risk, and narrative in health communication,” Media Psychology, volume 11, number 1, pp. 52–75.
doi: http://dx.doi.org/10.1080/15213260701853112, accessed 21 January 2014.

Rick Edmonds, Emily Guskin, Tom Rosenstiel, and Amy Mitchell, 2012. “State of the news media 2012: New devices, platforms spur more news consumption,” Pew Research Center (March 19), at http://www.pewresearch.org/2012/03/19/state-of-the-news-media-2012/, accessed 9 February 2013.

Susannah Fox and Joanna Brenner, 2012. “The social life of health information,” Pew Internet & American Life Project (12 July), at http://pewinternet.org/Reports/2012/Caregivers-online/Main-Report/The-social-life-of-health-information.aspx, accessed 12 July 2012.

Thomas D. Gore and Cheryl Campanella Bracken, 2005. “Testing the theoretical design of a health risk message: Reexamining the major tenets of the Extended Parallel Process Model,” Health Education & Behavior, volume 32, number 1, pp. 27–41.
doi: http://dx.doi.org/10.1177/1090198104266901, accessed 21 January 2014.

Maria Elizabeth Grabe, Shuhua Zhou, Annie Lang, and Paul David Bolls, 2000. “Packaging television news: The effects of tabloid on information processing and evaluative responses,” Journal of Broadcasting and Electronic Media, volume 44, number 4, pp. 581–598.
doi: http://dx.doi.org/10.1207/s15506878jobem4404_4, accessed 21 January 2014.

Nancy Grant Harrington, Derek Lane, Lewis Donohew, and Rick Zimmerman, 2006. “An extension of the activation model of information exposure: The addition of a cognitive variable to a model of attention,” Media Psychology, volume 8, number, pp. 139–164.
doi: http://dx.doi.org/10.1207/s1532785xmep0802_5, accessed 21 January 2014.

Claire Hewson, Peter Yule, Dianna Laurent, and Carl Vogel, 2003. Internet research methods: A practical guide for the social and behavioural sciences. London: Sage.

Darrin Hodgetts and Kerry Chamberlain, 2006. “Developing a critical media research agenda for health psychology,” Journal of Health Psychology, volume 11, number 2, pp. 317–327.
doi: http://dx.doi.org/10.1177/1359105306061190, accessed 21 January 2014.

C. Richard Hofstetter and David M. Dozier, 1986. “Useful news, sensational news: Quality, sensationalism and local TV news,” Journalism Quarterly, volume 63, number 4, pp. 815–820, 853.

Jaemin Jung, 2003. “Business news Web sites differ from newspapers in business content,” Newspaper Research Journal, volume 24, number 2, pp. 114–119.

Kimberly Nicole Kline and Marifran Mattson, 2000. “Breast self-examination pamphlets: A content analysis grounded in fear appeal research,” Health Communication, volume 12, number 1, pp. 1–21.
doi: http://dx.doi.org/10.1207/S15327027HC1201_01, accessed 21 January 2014.

Robert M. Knight, 2003. A journalistic approach to good writing: The craft of clarity. Second edition. Ames: Iowa State University Press.

Matthew W. Kreuter and Stephanie M. McClure, 2004. “The role of culture in health communication,” Annual Review of Public Health, volume 25, pp. 439–455.
doi: http://dx.doi.org/10.1146/annurev.publhealth.25.101802.123000, accessed 21 January 2014.

Michael S. LaTour, Robin L. Snipes, and Sara J. Bliss, 1996. “Don’t be afraid to use fear appeals: An experimental study,” Journal of Advertising Research, volume 36, number 2, pp. 59–67.

Julie Leask and Simon Chapman, 2002. “‘The cold hard facts’ Immunisation and vaccine preventable diseases in Australia’s newsprint media 1993–1998,” Social Science & Medicine, volume 54, number 3, pp. 445–457.
doi: http://dx.doi.org/10.1016/S0277-9536(01)00130-7, accessed 21 January 2014.

Pam Leece, Mohit Bhandari, Sheila Sprague, Marc F Swiontkowski, Emil H Schemitsch, Paul Tornetta III, P.J. Devereaux, and Gordon H. Guyatt, 2004. “Internet vs. mailed questionnaires: A randomized comparison,” Journal of Medical Internet Research, volume 6, number 3, at http://www.jmir.org/2004/3/e30/.
doi: http://dx.doi.org/10.2196/jmir.6.3.e30, accessed 21 January 2014.

Diane L. McKay, Judy M. Berkowitz, Jeffrey B. Blumberg, and Jeanne P. Goldberg, 2004. “Communicating cardiovascular disease risk due to elevated homocysteine levels: Using the EPPM to develop print materials,” Health Education & Behavior, volume 31, number 3, pp. 355–371.
doi: http://dx.doi.org/10.1177/1090198104263353, accessed 21 January 2014.

Philip Meyer, 2004. The vanishing newspaper: Saving journalism in the information age. Columbia: University of Missouri Press.

Cortney M. Moriarty and Jo Ellen Stryker, 2008. “Prevention and screening efficacy messages in newspaper accounts of cancer,” Health Education Research, volume 23, number 3, pp. 487–498.
doi: http://dx.doi.org/10.1093/her/cyl163, accessed 21 January 2014.

Barbara A. Morrongiello, Daniel Zdzieborski, Megan Sandomierski, and Jennifer Lasenby-Lessard, 2009. “Video messaging: What works to persuade mothers to supervise young children more closely in order to reduce injury risk?” Social Science & Medicine, volume 68, number 6, pp. 1,030–1,037.
doi: http://dx.doi.org/10.1016/j.socscimed.2008.12.019, accessed 21 January 2014.

J. Murphy, G. Gamble, and N. Sharpe, 1994. “Readability of subject information leaflets for medical research,” New Zealand Medical Journal, volume 107, number 991, pp. 509–510.

Victor Nell, 2002. “Why young men drive dangerously: Implications for injury prevention,” Current Directions in Psychological Science, volume 11, number 2, pp. 75–79.
doi: http://dx.doi.org/10.1111/1467-8721.00172, accessed 21 January 2014.

David Niven, 2003. “Objective evidence on media bias: Newspaper coverage of Congressional party switchers,” Journalism & Mass Communication Quarterly, volume 80, number 2, pp. 311–326.
doi: http://dx.doi.org/10.1177/107769900308000206, accessed 21 January 2014.

Jum C. Nunnally, 1978. Psychometric theory. Second edition. New York: McGraw–Hill.

Michael Schudson, 2001. “The objectivity norm in American journalism,” Journalism, volume 2, number 2, pp. 149–170.
doi: http://dx.doi.org/10.1177/146488490100200201, accessed 21 January 2014.

Larry E. Toothaker, 1993. Multiple comparisons procedures. Thousand Oaks, Calif.: Sage.

E. Vance Wilson and Nancy K. Lankton, 2004. “Modeling patients’ acceptance of provider–delivered e–health,” Journal of the American Medical Informatics Association, volume 11, number 4, pp. 241–248.
doi: http://dx.doi.org/10.1197/jamia.M1475, accessed 21 January 2014.

Kim Witte, 1992a. “Preventing AIDS through persuasive communications: A framework for constructing effective, culturally specific, preventive health messages,” International and Intercultural Communication Annual, volume 16, pp. 67–86.

Kim Witte, 1992b. “Putting fear back into fear appeals: The Extended Parallel Processing Model,” Communication Monographs, volume 59, number 4, pp. 329–349.
doi: http://dx.doi.org/10.1080/03637759209376276, accessed 21 January 2014.

Kim Witte and Mike Allen, 2000. “A meta–analysis of fear appeals: Implications for effective public health campaigns,” Health Education & Behavior, volume 27, number 5, pp. 591–615.
doi: http://dx.doi.org/10.1177/109019810002700506, accessed 21 January 2014.

Kim Witte, Gary Meyer, and Dennis Martell, 2001. Effective health risk messages: A step–by–step guide. Thousand Oaks, Calif: Sage.

Kim Witte, Kenzie A. Cameron, Maria Knight Lapinski, and Solomon Nyzuko, 1998. “A theoretically based evaluation of HIV/AIDS prevention campaigns along the Trans–Africa Highway in Kenya,” Journal of Health Communication, volume 3, number 4, number, pp. 345–363.
doi: http://dx.doi.org/10.1080/108107398127157, accessed 21 January 2014.

Kim Witte, Kenzie A. Cameron, Janet K. McKeon, and Judy M. Berkowitz, 1996. “Predicting risk behaviors: Development and validation of a diagnostic scale,” Journal of Health Communication, volume 1, number 4, pp. 317–341.
doi: http://dx.doi.org/10.1080/108107396127988, accessed 21 January 2014.

 

Appendix: Virus treatment article

 

Virus treatment article

 

 

Measurement instruments

 

 

Self-efficacy items

 

 


Editorial history

Received 10 February 2013; accepted 4 September 2013.


Creative Commons License
This paper is licensed under a Creative Commons Attribution–NonCommercial–ShareAlike 3.0 Unported License.

“You may have a cancer–causing virus and not even know it” Fear appeals in online news
by Brad Love and Michael S. Mackert.
First Monday, Volume 19, Number 2 - 3 February 2014
http://www.firstmonday.dk/ojs/index.php/fm/article/view/4368/3825
doi: http://dx.doi.org/10.5210/fm.v19i2.4368.





A Great Cities Initiative of the University of Illinois at Chicago University Library.

© First Monday, 1995-2017. ISSN 1396-0466.