Clouds on the horizon: privacy failures and its effect on electronic word-of-mouth

Clouds on the horizon: privacy

failures and its effect on

electronic word-of-mouth

Introduction

● Privacy failures - events that compromise personal information growing (e.g. Choi et al. 2016) ● Customers are at risk of identity theft and fraud

● Companies can expect negative eWOM as a response to privacy failures

● Research shows that this can lead to loss in both short-term and long-term profitability (Goldenberg et al. 2007; Villanueva et al. 2008; Luo 2009)

Introduction

Research questions that the study set out to answer:

RQ1: How do events related to privacy failures trigger eWOM in terms of volume and valence? RQ2: How is the volume and valence affected by the event type?

RQ3: How is the volume and valence affected by the industry?

Literature review

● Privacy failure (e.g. hacking, insider) a significant threat to customer privacy (Martin and Murphy 2016) ● Some consequences to firms of anticipated negative eWOM followed by the privacy failure in the long term

is brand dilution, volatile stock returns and decreasing firm value (Balaji et al. 2016)

● However, not all customers engage in negative eWOM as research shows that only the highly dissatisfied customers tend to do so (Richins 1983)

Literature review

● Volume and valence are central parts of eWOM (You et al. 2015; Rosario et al. 2016) where volume is highly visible while valence (or sentiment) is more diagnostic / informative

● Volume and valence expected to vary by the privacy failure (event type) and which industry it occur with basis in attribution theory

● According to attribution theory, a firm-related privacy failure such as an insider breach should lead to higher volume and more negative valence

● This is because customers make attributions or causal inferences about the privacy failure which may influence their response (Vaidyanathan and Aggarwal 2003)

Methodology

● Text mining to study social media data from Facebook: benefit that consumers are studied in their natural environment

● Regression analysis for hypothesis testing: zero-truncated count models for volume and linear regression for valence

● Data collected from the Facebook application programming interface (Graph API): from the official FB pages of the 5 companies Target, The Home Depot, JPMorgan Chase & Co., AT&T and Evernote

Methodology

● Two DV’s of eWOM with volume as a non-negative discrete count variable (number of privacy failure-related FB comments) and valence as a continuous sentiment score from the comment

● DV’s are created based on the FB comments extracted from the API data

● Two IV’s: event type as dummy variable (0=hacked / 1=insider) and industry as m-1 dummy variables leading to three dummy variables financial (0/1), telecom (0/1) and technology (0/1)

Results

● Results consist of two parts: first text mining then regression analysis

● Before text mining: checking for inconsistencies, data cleaning and preparation

● After data collection: created data frames for each company and created so-called corpus or collection of all the FB comments for a one-year period for each company

● Text mining proceeds with further creating a term document matrix, then regular matrix

● This serve as foundation for making word clouds, word frequency counts, word associations and sentiment analysis

Results: text mining

● The filtering process resulted in 2378 privacy failure-related FB comments to analyze (negative eWOM) ● Valence significantly more negative for all companies in the filtered data compared to total one-year data ● Common words in the eWOM that companies received: hacked, breach, security, card, account, fraud,

personal, information, problem, issue, data, stolen, privacy, protect

● Highest volume of privacy failure-related comments to Target with 1591 comments overall, and mean volume comments of nearly 10 received to each company FB posts , while 2nd highest for AT&T (350 comments, mean comments per company FB post = 4)

Results: regression analysis

Valence modeled by ordinary least squares regression due to continuous variable

● Four models were estimated: a full model, a full model with interaction, an event type only model and an industry only model

● Multicollinearity issue - in the full model and full with interaction the industry telecom was not estimable and in the full model, event type insider had a significant positive effect on the valence compared to hacked

● The event model was not overall significant, the industry only model chosen for further interpretation ● The valence was not more negative for the insider breached firm AT&T which was expected (H2) not

supported

Results: regression analysis

Volume modeled by zero-truncated poisson (ZTP) and zero-truncated negative binomial distribution (ZTNBD) since

the count variable had no 0 observations

● Estimation issues due to multicollinearity, hence models including event type only and industry only were estimated

● Significant positive effect of event type insider on the volume compared to hacked (baseline), with an increase in 239% as compared to hacked (H1 supported by ZTNBD)

● However, the results showed that the industry only model had the lowest AIC (1113.502) and BIC (1113.625) as well as lowest log-likelihood value (-551.750) and was interpreted further

General discussion

● Customers are expected to engage in negative eWOM following a privacy failure due to risk of identity theft and fraud (Son and Kim 2008) - it was found that the eWOM was significantly more negative for the filtered FB-comments

● It was expected a higher volume for insider breach than hacked (H1) which was supported, however more negative valence for an insider breach as compared to hacked was not found (H2 not supported)

● Due to multicollinearity issues full models with interaction terms were not estimable, thus the hypotheses H3a, H3b, H4a and H4b which stated that the industry can either strengthen or weaken the effect of event type on volume and valence were not supported

● In the negative eWOM for the five companies several common words were found which may indicate that people do not discriminate between the various events or industry in which it occurred

Conclusion

Managerial implications:

● Companies must act swiftly after a privacy failure to show that they take responsibility

● Especially, insider breach had a significant effect on the volume compared to hacking - so in cases of a firm-related privacy failure the company can expect a higher volume

Limitations: reliability, validity and generalizability

● Although free access, API data such as from Facebook quite restrictive and limited

Conclusion

Limitations: reliability, validity and generalizability (continued)

● Also due to companies’ eWOM management - some may sensor and delete negative comments ● The filtering process based on certain keywords (e.g. hacked, breach) to get the most relevant

privacy-related FB comments may have lead to exclusion of some (less) relevant comments to analyze ● Omitted variable bias: low explanation of the variation in the DV’s by the IV’s for any of the models as

measured by R2 - other relevant predictors are not included in the model

Conclusion

● It is suggested to obtain more data from several types of privacy failures

● Take into account social influence, how peers in the network also trigger the negative eWOM further by commenting others negative comments and thus sparking a discussion

References

Balaji, M.S, Kok Wei Khong and Alain Yee Loong Chong. (2016). Determinants of negative word-of-mouth communication using social networking sites. Information & Management. 53(4), 528-540. Choi, Ben C.F, Sung S.Kim and Zhenhui (Jack) Jiang. (2016). Influence of Firm’s Recovery Endeavors upon Privacy Breach on Online Customer Behavior. Journal of Management Information Systems. 33(3), 904-933.

Goldenberg, Jacob, Barak Libai, Sarit Moldovan and Eitan Muller. (2007). The NPV of Bad News. International Journal of Research in Marketing. 24(3), 186-200. Lomborg, Stine and Anja Bechmann. (2014). Using API’s for Data Collection on Social Media. The Information Society. 30(4), 256-265.

Luo, Xueming. (2009). Quantifying the Long-Term Impact of Negative Word of Mouth on Cash Flows and Stock Prices. Marketing Science. 28(1), 148-165.

Malhotra, Arvind and Claudia Malhotra. (2011). Evaluating Customer Information Breaches as Service Failures: An Event Study Approach. Journal of Service Research. 14(1), 44-59. Martin, Kelly D. and Patrick Murphy. (2016). The role of data privacy in marketing. Journal of the Academy of Marketing Science. 45(2), 135-155.

Richins, Marsha L. (1983). Negative Word-of-Mouth by Dissatisfied Consumers: A Pilot Study. Journal of Marketing. 47(1), 68-78.

Rosario, Ana Babic, Francesca Sotgiu, Kristine De Valck and Tammo H.A Bijmolt. (2016). The effect of electronic word of mouth on sales: a meta-analytic review of platform, product and metric factors. Journal of Marketing Research. 53(3), 297-318.

Son, Jai-Yeol and Sung S. Kim. (2008). Internet Users’ Information Privacy-Protective Responses: A Taxonomy and a Nomological Model. MIS Quarterly. 32(3), 503-529.

Vaidyanathan, Rajiv and Praveen Aggarwal. (2003). Who is the fairest of them all? An attributional approach to price fairness perceptions. Journal of Business Research. 56(6), 453-463.

Villanueva, Julian, Shijn Yoo and Dominique Hanssens. (2008). The Impact of Marketing-Induced versus Word-of-Mouth Acquisition on Customer Equity Growth. Journal of Marketing Research. 45(1), 48-59.

Thank you for the attention!

Questions?

Master thesis and presentation by Kristian Reksen Svedal