Preserving corporate reputation through response strategy and communication channel. An experimental study towards a crisis scenario

Research Paper

An experimental study towards a crisis scenario:

Preserving corporate reputation through response strategy and communication channel

Julien Bruhl

Graduate School of Communication

University of Amsterdam

25th June, 2021 words: 3185

1. Introduction

This spring, when COVID-19 peaked, Bottega Veneta decided to have a fashion show at a club in Berlin. Thus, celebrities were invited and even had a party afterwards. However, the label denied commenting on the accusations of acting unethically (Weiss, 2021).

This example illustrates one of the few existing crisis response strategies, namely the denial strategy (Claeys et al., 2010). Also, the communication channel is an essential aspect. Since Schultz et al. (2011) discovered that “the medium had an effect on the organisational reputation” (p. 24), our study grasps upon these aspects by analysing the perception of varying messages through a social media site and a press release. Here, “electronic word of mouth communications, which are able to reach a vast number of people in a short period of time” (Ismagilova et al., 2017, p. 20) might be relevant.

Crises threaten the organisation's attractiveness and trustworthiness (Coombs, 2007). Consequently, a solution is to prevent worse, with respect towards the manifold stakeholders.

RQ1: Do the communication channel and crisis response strategy influence the company's reputation after a crisis?

Research indicates twitter users attest companies a more favourable reputation than newspaper readers (Schultz et al., 2011). Thus, our first hypothesis is the following:

H1: Using twitter as a communication channel stronger mitigates the public reputation loss due to the crisis than a press release.

Another strategy applied by the organisation is dedicated to rebuilding (Claeys et al., 2010). As scholars emphasise the non-existent difference between rebuild and denial strategies, regarding the reputation (Claeys et al., 2010), we verbalised this hypothesis:

H2: The corporate reputation after the crisis does not differ for rebuilding and denial response strategies.

We perceive apology as a rebuild strategy (Coombs & Holladay, 2008). Interestingly, findings report that the reduction of negative word of mouth is even more pronounced in less engaging responses than the apology (Schultz et al., 2011). So, the third hypothesis is:

H3: Compared to reading a press release, the positive effect on the company’s reputation of using twitter as a communication channel is weaker pronounced for rebuilding communication strategies than for denying communication strategies.

Moreover, higher crisis involvement leads to either higher or lower post-crisis attitude, depending on a matching response strategy in line with the situational crisis communication theory (Claeys & Cauberghe, 2014). However, we cannot postulate, which response strategy matches the crisis, as the truth behind the crisis is not disclosed:

H4: The non-existent effect of response strategy on the organization's reputation after the crisis will not differ between participants with either a high or a low crisis-involvement.

2. Methods

Design

To analyse whether the examined dimensions affect the crisis reverberation, we built a survey-embedded online experiment. Our choice was a 2 (communication channel: twitter vs press release) x 2 (response strategy: denial vs rebuild) between-subjects factorial design with no control group.

Participants

We opted for a convenience sample, as there was not enough time and resources available for a random sampling technique. Even though our team had to exclude three of them due to not giving consent, being below 18 years old or not answering the question around age accurately, enough individuals participated. After these adjustments, our sample contained N = 214 participants. The average age was 29.23 years (SD = 10.16). Also, 58.9% were female, 36.9% were male, 1.4% non-binary, and 2.8% did not disclose their gender. Furthermore, we asked the participants about their education: 9.3% had a high school degree or lower, while 2.8% held a trade, technical or vocational degree, 3.7% possessed an associate degree, 52.8% had a bachelor degree and 31.3% a master degree or higher. Furthermore, everyone participated voluntarily, as there was no incentive given.

Procedure and stimuli

In the beginning, the ones involved in the study read a letter about the research project's motive and declared their consent. Due to the latter, they had to be 18 years or older. Then, the survey started with two moderator variables.

At this point, a BBC online article with the header ‘Fast food chain in trouble after offensive comments’ introduced the fictional scenario: An LGBTQ+ discrimination in a Francis' Fries restaurant, which is a fictive company, did occur, involving a branch manager insulting a customer and a video of the incident circulated on social media. We aspired to

choose a news site, which most of the participants knew. Subsequently, Qualtrics randomly assigned all individuals to one of the four groups. On the one hand, participants could see a screenshot from twitter with two paragraphs and one picture. On the other hand, they encountered a letter to the customers published on the website’s news section, including the corporate logo. For both stimuli, a ten-second timer was built in to prevent the participants from preceding too quickly. The primary objective was to offer the participants a global brand’s narrative, which was developed around a start-up aesthetic. As participants should get a stimulus from a publicly accessible media environment, twitter and a website communique were the selected communication channels. These independent variables in the denial condition stated that the video was fake and defended the employees. By contrast, in the rebuild condition, not only an apology was offered, but also the employment relationship with the offender was terminated (see Appendix).

Ensuingly, the participants answered questions regarding another moderator variable and commenced to fill out the questionnaire concerned with the dependent variables. In the next section, two manipulation checks were presented, and we collected information on the demographic variables. Ultimately, everyone was debriefed and had the opportunity to give feedback in an open question.

Manipulation checks

As our foundation are the independent variables, we opted to notice technical and interpretative issues. So, we asked two dichotomous questions about the response strategy and the communication channel. Since we examine the relationship between two categorical variables, we must conduct two chi-squared tests on a contingency table. For our first one, the data satisfy the conditions since there are zero cells (<20%) counted less than five, and no cell has a frequency of zero. Chi-square requires a table with at least three columns or rows. Therefore, we conduct Fischer’s exact test, p < .001, which shows that our manipulation worked. This is a strong effect, Goodman & Kruskal’s tau = .72. Participants in the denial condition stated way more often (94%) that Francis' Fries denied the incident than those in the rebuild condition (9%). At the same time, individuals in the former condition had a lower score (7%) compared to those in the latter condition (92%) when they were asked whether the restaurant admitted their responsibilities. In the second test, the data satisfy the conditions too since there are zero cells (<20%) counted less than five. Also, no cell has a frequency of zero. Fischer’s exact test proves our manipulation was successful, p <.001. The relationship is strong, Goodman & Kruskal’s tau = .66. Those who saw a tweet stated way more often that twitter was the medium they were exposed to (89%) than those in the press release condition (7%). Moreover, participants who saw a response on twitter (11%) answered having seen a press release less in comparison to individuals reading the press release (93%). This means that the manipulation was perceived as intended.

Measures

We built a seven-point Likert scale for the moderator variable (1 = strongly disagree, 7 = strongly agree) crisis-involvement (M = 5.03, SD = 1.38, Cronbach’s a = .83). Therefore, two items were derived from a pre-existing scale, namely “This crisis is not relevant to me. This crisis is important to me” (Kopalle & Lehmann, 2001, as cited in Lee & Atkinson, 2019, p. 183). The additional statements 'I worry about this crisis happening to me' and ‘I am not interested in this crisis' were formulated by the researchers. In order to avoid acquiescence reverse coding was applied. At the same time, these four items had their theoretical foundation in Freedman (1964), who postulates involvement is a person's “interest in, concern about or commitment to a particular position on an issue” (p. 290). We evaluated this scale through a factor analysis. The Kaiser-Meyer-Olkin measure for sampling adequacy was above the minimum criterion of 0.5 (KMO = .74), and Bartlett's test of sphericity shows the correlations between variables to be significantly different from zero, x2 (6) = 266.92, p < .001. Based on Kaiser's criterion, we retained one extracted factor, which had an eigenvalue of 2.39 and explained more than half of the variance (60%) in the individual items. As the third item had a factor loading below .4, a decision was made to remove it as a measure of the underlying construct.

We operationalised our dependent variable through a seven-point Likert scale as well. Our team aimed to measure the corporate reputation with a meshed up seven-point Likert scale ranging from 1 = strongly disagree to 7 = strongly agree (M = 3.32, SD = 1.24, Cronbach's a = .92). Here, five items were presented to the participants and the first three were worded this way: “I trust this company. The company maintains high standards in the way it treats people. I admire and respect this company” (Fombrun et al., 2000, p. 253). They were succeeded by “Under most circumstances, I would be likely to believe what the organization says” (Coombs & Holladay, 1996, as cited in Coombs & Holladay, 2002, p. 174). Ultimately, “The company is trying to mislead me” (MacKenzie & Lutz, 1989, as cited in Newell & Goldsmith, 2001, p. 238) completed the scale. Here, a factor analysis was conducted. The Kaiser-Meyer-Olkin measure for sampling adequacy was above the minimum criterion of 0.5 (KMO = .88), and Bartlett's test of sphericity shows the correlations between variables to be significantly different from zero, x2 (10) = 790.56, p <.001. Based on Kaiser's criterion, we retained one extracted factor, which had an eigenvalue of 3.82 and explained more than half of the variance (76%) in the individual items. As all items had factor loadings well above .4, all items were retained as measures of the underlying construct. Our decision to combine existing items covering trustworthiness, ethical considerations, admiration, credibility and the individual relationship with the company was motivated by aiming to capture diverse perspectives.

Plan of analysis

To obtain results for the hypotheses, we conduct two independent samples t-tests. They will allow us to measure the influence of the dichotomous independent variables communication channel and crisis response strategy on the numerical dependent variable corporate reputation. Moreover, these two categories of each predictor variable are statistically significant independent samples that we want to compare regarding their outcome variable. Also, the researchers carry out a two-way analysis of variance to measure the interaction effect between the predictors communication channel and response strategy, and its impact on the outcome variable corporate reputation. When testing hypothesis number four around the moderation of the continuous predictor crisis-involvement, we execute a multiple regression because response strategy is a dichotomy, and corporate reputation is measured on an interval level.

3. Results

Effect of communication channel on corporate reputation: twitter vs press release (H1) Each of the two compared samples consists of more than 30 participants (N = 214), so the relevant assumption is met for the independent samples t-test with communication channel as a predictor and corporate reputation as an outcome variable. Hypothesis 1 stated that twitter as a communication channel is more beneficial for the organisation. However, the restaurant reputation is on average higher among participants in the press release condition (M = 3.40, SD = 0.12), compared to the ones in the twitter condition (M = 3.23, SD = 1.26).

However, the mean difference (of 0.17) is not statistically significant, t (212) = 1.02, p = .308, 95% CI [-0.16, 0.51], and constitutes a small effect, d = 0.14. In line with H1, results do not confirm that twitter positively influences corporate reputation.

Table 1

Group Differences for Reputation Between Groups in the Twitter Condition or in the Press Release Condition

Abbildung in dieser Leseprobe nicht enthalten

Effect of response strategy on corporate reputation: rebuild vs denial strategy (H2) Again, our relevant assumption is fulfilled since each of the two compared samples consists of more than 30 participants (N = 214). According to Hypotheses 2, there is no difference between using a rebuild or a denial response in terms of reputation restoring. So, our team runs an independent samples t-test with the independent variable response strategy and the dependent variable corporate reputation. The company's reputation is on average more favourable among participants in the rebuild condition (M = 3.81, SD = 1.21), compared to the ones in the denial condition (M = 2.83, SD = 1.06). The mean difference (of 0.98) is statistically significant, t (212) = 6.30, p < .001, 95% CI [0.68, 1.29], and represents a strong effect, d = 0.87. Consequently, we reject the second hypothesis.

Effect of interaction between communication channel and response strategy on corporate reputation (H3)

In this case, we conduct a two-way analysis of variance with the predictors communication channel and response strategy, as well as the outcome variable corporate reputation. The subgroups that are compared here are more or less of equal size, and all include more than 30 cases, so the relevant assumptions are satisfied. In the rebuild and twitter condition, there are n = 53 participants, which is an equal number of those exposed to a press release with a rebuild strategy. At the same time, n = 52 participants saw a tweet denying the accusations and n = 56 individuals were assigned to the press release in the denial condition. Following Hypotheses 3, there is a positive effect on corporate reputation positive using twitter in a denial condition, compared to a press release. This specific effect of twitter should increases less strongly in a rebuild condition. So, there should be a visible interaction. The two-way analysis of variance shows a statistically not significant interaction effect of response strategy and communication channel, F (1, 210) = 0.78, p = .379, n2 = 0.00. Thus, 0% of the variance in corporate reputation is explained by the interaction between response strategy and communication channel. The size of this effect is very weak. The analysis shows that those in the rebuild condition combined with the twitter condition reported less positive reputation (M = 3.78, SD = 1.18) than the ones who saw the press release in the same response condition (M = 3.84, SD = 1.25), yet when the response denied the crisis participants that were shown the tweet had the most negative perception of the firm (M = 2.66, SD = 1.08), while the readers of the press release had a somewhat higher reputation score (M = 2.99, SD = 1.04). Therefore, we reject the hypothesis of an interaction between the two predictors.

[...]