A crisis by any other name? Examining the effects of journalistic “crisis labeling” on corporate perceptions

Framing and Attribution Theory

There is plenty of existing evidence that people are, consciously or not, “acutely sensitive to contextual cues” when “making decisions, forming judgments, or expressing opinions.” ² Media can subtly alter the messages their readers are exposed to every day, which can lead them to interpret these messages in a certain way. This concept is referred to as framing.

Framing research began in sociology, when Goffman referred to a “schematic of interpretation,” through which he believed individuals “organize and make sense of information or an occurrence” as a “frame.” ³ Frames call attention to particular aspects of the story they describe, thus drawing attention away from other aspects deemed less important. ⁴ This means certain words, phrases, or information will be either included or omitted in order to “fit” the story into a specific “frame” of reference that will heavily influence what the reader takes from it. Research suggests that frames are “at the core of the media” ⁵ and are “critical to the construction of social reality.” ⁶

Different frames have been shown to lead to different attributions, based on people’s perceptions of causality. ⁷ Iyengar named two types of attributions: causal attributions, which make judgments as to the origins of the issue, and treatment attributions, which judge who or what has the power to either improve or worsen the issue. ⁸ Taylor and Fiske, however, suggested another potential factor in attribution: “cognitive availability.” ⁹ They proposed that people assign causal responsibility by using the information that comes to mind most easily. The present study is based on this idea—that people will attribute causal responsibility to organizations if a story is framed using key literary devices (i.e., crisis labels).

Attributions Based on Language Used by News Media

Public relations researchers often discuss the role of responsibility attribution in a crisis, insofar as the perceived cause often affects the response strategy. ¹⁰ But is it possible that the media’s linguistic choices can affect perceived attribution? Although not much research has been published on the power of wording and syntax in crisis framing, ¹¹ some studies have argued that, yes, language can, indeed, have significant effects.

Researchers have found that news article structure and information order can affect perceived attribution. Hewstone, for example, found that the presence or absence of the actor, stimulus and circumstances in the first sentence of a story affected readers’ subsequent causal attributions. ¹² Similarly, Leets found that the first actor in a story was assigned responsibility significantly more frequently than the second actor. ¹³ Knobloch-Westerwick and Taylor also found that causal attributions were generally more consistent with the first argument presented, as opposed to the second. ¹⁴

Studies show that micro-choices in linguistics also affect perceived attribution. Henley, Miller and Beazley, for example, found that respondents showed more acceptance of violence against women after reading reports of it written in the passive voice, rather than the active voice. ¹⁵ Similarly, Knobloch-Westerwick and Taylor found that when individuals’ actions were described using the active voice, he or she was perceived as more responsible for those actions. ¹⁶

Two studies have specifically examined attribution effects of labeling. For example, Gamson and Modigliani looked at effects of labeling in the context of nuclear power. ¹⁷ Many media outlets initially framed nuclear power as evil and destructive, or the “devil’s bargain.” ¹⁸ The researchers found trends in public opinion corresponding to this media portrayal and concluded that media discourse could both reflect and contribute to the creation of cultural opinions of issues. Moreover, they argue that journalists and public relations practitioners have the unique ability to invent these “catchphrases” (i.e., “labels”) that will come to dominate the public discourse. ¹⁹ Similarly, Jones, Kanouse, Kelley, Nisbett, Valins and Weiner found that labeling could affect causal attributions when the language used to describe the event contained explicit attributions. ²⁰

Media Labeling and Corporate Perceptions

Corporate perceptions refer generally to the attributes a person assigns to a company. This study examines the effect of labeling on several perceptions, including trust, reputation and locus. Fombrun and Shanley found that media visibility played a key role in reputation judgments, ²¹ so this study seeks to replicate those findings and see whether they extend to other judgments. In all, this study seeks to explore the relationship between crisis labeling in news articles and readers’ corporate perceptions. In addition, it looks at whether individual differences, such as gender, moderate this effect. This leads to the study’s core questions:

Study Design

A between-subjects experiment was conducted offline. Participants (N = 88) were undergraduate students recruited from four communications classes at a large public university in exchange for extra credit. Most (n = 69) were female, 18 were male and one did not list a gender. Ages ranged from 20 to 25 (M = 20.93, SD = 0.89).

Stimulus

All participants read the same news article, which was adapted from a Dole Food press release about a spinach recall. ²² A news article, rather than a press release, was preferred for the purposes of ecological validity; public relations practitioners would be unlikely to explicitly associate a crisis with their own organization, but, as discussed in the introduction, crisis labeling is a common practice in the news media. The text was, therefore, revised to read like a news article. It was also rewritten to describe a fictional tomato recall crisis that involved a fictional company, “FreshVeg, Inc.”

Participants were randomly assigned to one of two conditions. In the experimental condition (n = 46), the news article repeated the company’s name all eight times the crisis was mentioned. In the control condition (n = 42), the news article mentioned the company’s name only once, at the beginning of the article, and then referred to it more generically. For example, a sentence in the experimental condition read, “Consumers should dispose of any tomatoes purchased during the FreshVeg, Inc. tomato crisis,” while the same sentence in the control condition read, “Consumers should dispose of any tomatoes purchased during the tomato crisis.”

Manipulation Check

At the end of the study, participants were asked to write the name of the crisis they had read about. For the control condition, answers were marked as correct if they mentioned “tomato.” For the experimental condition, answers were marked as correct if they mentioned “FreshVeg, Inc.” and “tomato.” The researchers accepted misspellings of “tomato” and mild variations on the company name.

Across conditions, 65 answered correctly, 17 answered incorrectly and six did not answer. However, analysis showed a significant difference between experimental (58.5 percent) and control (100 percent) conditions in terms of their likelihood of answering correctly, χ ² (1, N = 82) = 21.45, p < .001, Cramer’s V = .51. Implications are discussed in the “Discussion” section.

Measures

RQ1: Journalistic “Crisis Labeling” and Corporate Perceptions

The results of a one-tailed test showed that participants in the experimental condition (M = 4.46, SD = 1.11) were more likely to say they lost confidence in the product than were participants in the control condition (M = 3.86, SD = 1.28), t(86) = 2.35, p = .02, d = .50. Similarly, participants in the experimental condition (M = 4.09, SD = 1.53) were more likely to say they lost confidence in the company than were participants in the control condition (M = 3.83, SD = 1.41), though this was not significant, t(86) = 0.81, p = .42, d = .18. Further probing showed a moderately positive relationship between losing confidence in the product and losing confidence in the company (r = .53, p < .001, N = 88).

Several other variables showed directionally consistent but nonsignificant findings, such as labeling leading to lower perceptions, but others showed no differences at all. Interestingly, the average corporate reputation score was higher for the experimental group (M = 3.58, SD = 0.96) than for the control group (M = 3.38, SD = 0.92), but this was not statistically significant, t(86) = 0.98, p = .33. Table 1 shows means and t-test results for all corporate perception measures.

OPEN IN VIEWER

Table 1

Means and Standard Deviations of “Crisis Labeling” and Perceptions

Perception	Subdimension	No Labeling M (SD)	Labeling M (SD)	t	p	d
Company Trust	Lost Confidence	3.83 (1.41)	4.09 (1.53)	.81	.42	.18
	Trustworthy	3.86 (1.32)	4.04 (1.17)	.70	.49	.14
	Reliable	3.52 (1.25)	3.52 (1.15)	.01	.99	0
	Quality Doubts	4.62 (1.25)	4.70 (1.31)	.28	.78	.06
Product Trust	Lost Confidence	3.86 (1.28)	4.46 (1.11)	2.35	.02	.50
	Trustworthy	3.79 (1.26)	3.80 (1.11)	.07	.94	.01
	Reliable	3.50 (1.07)	3.63 (1.06)	.58	.57	.12
	Quality Doubts	4.48 (1.23)	4.61 (1.13)	.53	.60	.11
Reputation		3.38 (0.92)	3.58 (0.96)	.98	.33	.22
Locus		3.95 (1.43)	3.67 (1.37)	.93	.35	.20

Note. N = 88, df = 86. All perceptions measured from 1 (absence) to 7 (presence). Subdimensions for reputation and locus are not included because, as discussed in the text, both are essentially treated as single-item measures (reputation because it was measured by an index and locus because it was measured by a single item).

For participants in the experimental condition, those who correctly noted the crisis name (M = 3.96, SD = 1.30) were more likely to say the crisis was inside the company’s control than were the participants who misstated the crisis name (M = 4.88, SD = 1.45), t(39) = −2.14, p = .04, d = .67. All participants in the control condition correctly stated the crisis name, so this difference was not found among that group. In addition, this effect was not seen with measures of corporate reputation or trust.

RQ2: Moderating Effects of Individual Differences

The results of a one-tailed test showed that, across conditions, women (M = 4.68, SD = 1.16) were more likely to say they have doubts about product quality than were men (M = 4.00, SD = 1.14), t(85) = 2.23, p = .03, d = .59. Gender differences were not found for other measures of trust, or for the measures of reputation or locus. Other individual differences, such as age, were not tested because the sample lacked sufficient variation.

Theoretical and Practical Implications

This study builds on previous work that discusses the effect of language within the context of framing theory and attribution theory. As discussed earlier, Taylor and Fiske’s “cognitive availability” framework proposes that people assign causal responsibility based on information that comes to mind most easily. ²⁹ This study did not find significant relationships between journalistic “crisis labeling” and causal attribution.

These results, or lack thereof, have practical implications for journalists and public relations practitioners. If the proposed relationship does not exist, and journalistic “crisis labeling” does not affect corporate perceptions, then public relations practitioners do not need to be concerned with the media’s micro-frames. They might matter more in different contexts, but in the example studied here, where the crisis lacked responsibility, journalistic micro-frames were rather meaningless in changing the conversation. The lack of statistically significant findings in this study suggests that public relations practitioners might not need to be concerned with how media outlets label their crisis.

On the surface, these results contradict previous theoretical research, but, as discussed below, these contradictions can be explained by various methodological and practical considerations.

It could be that the labeling manipulation was not strong enough, based on the ineffective manipulation check. As stated, many participants in the experimental condition did not remember the label of the crisis presented in the news article. It could be that the manipulation was not strong enough to elicit recall. However, the article in the experimental condition mentioned the company name eight times; the readers likely processed or at least noticed the manipulation, even if they were unable to name the crisis in a free-recall format. Therefore, though possible, it seems unlikely that this would completely account for the null results.

The second possibility is that previous results were overstated. Perhaps a publication bias (the so-called “file drawer problem”), as discussed throughout social science research, ³⁰ also has occurred in micro-frames research.

It could also be that the labeling manipulation did not indicate causal responsibility. The stimulus clearly states that the crisis was not the organization’s fault: “It seems the (FreshVeg, Inc.) tomato crisis was due to contamination by runoff.” It could be that crisis labeling would have a larger effect if the company were said to be at fault. This could explain why, for example, reputation was higher in the labeling group than in the control; maybe participants saw the label as the company taking responsibility when they were not responsible, which in turn made them seem trustworthy. This explanation seems particularly likely based on the findings of Jones, Kanouse, Kelley, Nisbett, Valins and Weiner; as stated earlier, they found that labeling could affect causal attributions when the language used to describe the event contained explicit attributions. ³¹ The findings of the present study might be different than previous literature, then, because the language used here to describe the event did not contain explicit attributions.

Perhaps this type of labeling does not, by itself, make the relevant information (i.e., the association between the company and the crisis) easily come to mind. It could be that the effects of journalistic labeling require more time or engagement than allowed for in this study. Perhaps the effects of journalistic “crisis labeling” are not immediate, as tested for here, but could be seen with longitudinal tests. Maybe calling the crisis the “BP oil spill” in one article is not enough to affect most corporate perceptions; maybe it is the continued use of the term across mediated and nonmediated contexts that ultimately affects perceptions. This possibility makes sense in light of Gamson and Modigliani’s labeling study; as discussed, they found that media discourse could both reflect and contribute to the creation of cultural opinions of issues by creating these catchphrases. ³² Perhaps the real effect of journalistic labeling, then, is in that larger scale reflection and contribution, neither of which could be tested within the parameters of this project but could be in future studies.

Limitations

This study might have neglected to measure relevant factors, such as the nature of the crisis, the credibility of the news outlet, or the duration of the news cycle. The possible effect of such unmeasured factors is supported by the fact the attitudes toward the product and the company differed, both from each other and in their relationship to the manipulation. The manipulation affected only confidence in the product, not the company. Although perceptions of both were correlated with each other, this was not a perfect linear relationship. In fact, post hoc analysis shows the correlation between corporate trust and corporate reputation was actually stronger (r = .78, p < .001, N = 88) than the correlation between corporate trust and product trust (r = .53, p < .001, N = 88). This suggests something about the difference between corporate- and product-perceptions still needs to be further analyzed in future studies of their relationship with journalistic micro-frames.

This potential overlap between corporate and product trust measures might be the result of the methodological issues faced with the trust scales, as discussed in the “Method” section. The principal components analysis revealed that instead of grouping by content, the items grouped by language; that is, the positively worded items grouped together, and the negatively worded items grouped together. This suggests a methodological problem wherein participants were either ignoring content and responding the same way to every question, or they were confused by the wording of the questions. In either case, the problem led the researchers to separate the items for use as single-item indicators, rather than proper indices. A future study with more reliable measures could help determine the reliability of this study’s findings.

Similarly, it could be that the fictional nature of the company kept the relevant information (i.e., the association between the company and the crisis) from coming to mind easily. A company was fabricated here so as to control for preconceived attitudes, but it could be that those very attitudes are necessary for the presence of reputation judgments. Similarly, the use of a fake company might have made the crisis seem less important, as readers would necessarily be unaffected by it. Perhaps the true relationship lies in some sort of amplification effect, based on preexisting attitudes toward the company. A future study using an existing company could address this possibility.

This study was conducted with communication students, not a general undergraduate population, let alone a sample representing the general public. Results from such convenience samples cannot be generalized to any larger population. But the fact that crisis labels had a seemingly negligible effect even on communication students—people specifically trained in identifying source cues and messaging techniques—suggests that the effect of labeling might be overstated even here. In other words, if labels had such a minor effect with this seemingly high-involvement sample, then perhaps labels would be ignored completely by the average reader. Future research with a more diverse sample could test this possibility.