Product Recall and Future Choices - An Exploratory Study Con Korkofingas and Lawrence Ang, Macquarie University Abstract Major product recall incidents involving established brands over the last few years have increased markedly. Although the direct costs have been evaluated in these cases (typically in the millions) the indirect costs to brand equity and subsequent loss of market share are harder to evaluate. Although many case studies and some limited theoretical research have examined the impact of product recall on some of the above measures, there does not appear to be a framework that can be useful for generalisation. This paper applies a simulated multi-stage choice based experiment to assess the impact of hypothetical product recall experiences on brand equity measures and importantly future brand choice. Contrary to existing evidence we find that product recall experience has greater negative impacts for established strong brands than weaker nonestablished brands. Additionally, attributes of product recall such as the seriousness of the problem and speed of initial action impact on pre and post recall differences in consumer evaluations of brand equity. Differences in brand equity evaluations for the established strong brand significantly impact on post recall choice. Introduction The year 2007 can be considered to be ‘annus horribilis’ for product recall. Fortune (2007) voted a series of high profile product-recall in 2007 to be the number one dumbest moments in business, all of which involved Chinese-made products or ingredients. This includes Mattel recalling 20 million toy items because of lead paint; 850,000 Barbie accessories and Sesame Street toys for the same reason. Other prominent product recalls over the years include: Bridgestone/Firestone recall of 6.5 million tyres in 2000; 30 million cans and bottles of Coca Cola in Belgium and France in 1999; Intel’s recall of its flawed microprocessor in 1994 (costing Intel $500 million); Johnson and Johnson’s withdrawal of 31 million bottles of Tylenol in 1982 (costing $1.5 million for Johnson and Johnson to buy back). A serious outcome of product recall is the loss of brand equity and market share leading eventually to a loss of shareholder value (Davidson and Worrell, 1992). For instance, in 1990 the French chic mineral water, Perrier was recalled due to traces of benzene, a known carcinogenic. With its key positioning of “purity” severely damaged Perrier never recovered, losing half its market share to other brands (Geyser and Klein, 1990). By 1994, the brand generated only $59 million in revenue compared to $118 million prior to the recall (International Manager, 1994). However, despite the large financial cost to companies, this area has received a relatively small amount of systematic research. The aim of this paper is to examine the impact of product recall characteristics on key variables such as brand equity and future purchase choice through a simulated product recall situation. Literature Review Examination of previous product recall crises, company responses and consequences as case studies provides one avenue of research in this area. The case of how Tylenol cleverly went about 0 handling its October 1982 crisis of cyanide poison is now well known (Keller, 2003, p. 337-341). Other studies (Weinberger and Lepkowska-White (2000), Smith et al. (1996), Rupp and Taylor (2002)) study previous product recall experiences to provide checklists of appropriate managerial actions during product recall circumstances. Further research focuses on impacts of product harm crises on a variety of performance measures. Impacts on company stock price were examined by Davidson and Worrell (1992), Govindaraj and Jaggi (2004) and Rhee and Haunschild (2006) while Van Heerde et al. (2007) decomposed aggregate brand sales to estimate product recall impact on marketing effectiveness. Recent work by Cleeren et. al (2008) used individual level scanner data to examine how own and competitive brand advertising and brand characteristics moderate the impacts of a product harm crisis on two brands of peanut butter. These above studies, although useful, do not provide generalisable principles that would allow managers to make effective decisions under different circumstances. Another stream of research attempts, generally through experimentation, to understand underlying behavioural responses providing an evaluation of possible alternative marketing actions that could be taken during product harm crisis. Dawar and Pilutla (2000) found that if companies stonewalled in their response to a product-harm situation, its brand equity will decrease, but only if consumers have weak positive prior expectations about the company should do. Ahluwalia et al (2000) found that the level of consumers’ commitment also has a strong moderating influence on whether their attitude towards the brand will change. Over three studies, they consistently showed the higher consumer commitment to a brand (loyalty), the less vulnerable to negative publicity. Bitner (1990) found that consumers are less likely to be unhappy if causes of service failure are attributed to the employees of the firm to the (rather than the firm itself) because this is perceived to be temporary (stability attribution). If consumers perceive a service failure to be due to external uncontrollable circumstances less blame is put on the firm. Furthermore, Folkes and associates (1984, 1986) found that if a consumer attributes a product failure to the firm rather than to himself/herself, the consumer is more likely to feel anger, generate negative word-ofmouth, and want a refund and apology (locus attribution). Griffin et al (1991) found where blame is attributed to the firm, consumers are also less likely to want to buy again from the firm. Keller (2003) argued the longer the delay in executing the product recall, the greater the chance the equity of the brand will be damaged by the negative publicity and word-of-mouth generated. Our study contributes to this experimental research stream in seeking to understand the behavioural process by proposing a two stage choice experiment. The experiment differs from previous experiments because it is comprehensive manipulating a number of relevant product recall attributes simultaneously. It also differs from previous research by using a second stage choice to predict the impact of product recall attributes on future market shares. This allows for evaluation of alternative actions on brand choice probability and hence market shares directly. Although, the validity of lab experiments is open to question, the advantage is an understanding of the general behavioural process and a direct link to predicted market shares. Theory and Method We modify the brand value of Brand j (BVj) representation of Keller and Lehmann (2006) to posit brand valuation given information at time zero (I0) as BVj|I0 = Σ βi Xji,0 – Pj,0 + Vj,0 + Sj,0 (1) 1 Where Xji = functional attribute i levels, brand j; Pj = Price j ; Vj = brand equity; Sj = inertia Consumers evaluate BVj for all j and choose brand with maximum BVj. Product recall (and associated m characteristics (PRm)) will possibly impact on Xji, Vj. Consumers will update their attribute perceptions to Xji, 1 = f1 (Xji,0 , PRm) and brand equity to Vj,1 = f2 (Vji,0 , PRm), leading to an updating of BVj for all brands. Assuming product recall leaves perceptions of functional attributes (Xji) unchanged, and price and inertia remain the same then any changes in brand evaluations and hence brand switching are likely to be due to changes in brand equity Vj (affected by product recall characteristics PRm) To assess the impact of product recall characteristics on brand switching a two-stage factorial experiment was designed. In the first stage, respondents were asked to make a choice between two branded Mp3 players; Apple and a less well known brand (Sansa). The Mp3 players were almost identical in all respects apart from brand and price. The second stage involved a hypothetical product recall experience with the brand chosen in the first stage. After the product recall experience, a further choice between the MP3 players was elicited. Both stages were conducted using a self-completed survey booklet. The initial scenario information indicated that the respondent was in the market for an 8 Gb MP3 player and only two choices were available; Apple ($250) and Sansa ($150). Respondents were shown photographs of the respective Mp3 players with information about player dimensions, features and prices. All player features (excluding price and brand) were identical (brand photographs almost identical). After viewing the photographs and information, respondents were asked to indicate a choice. Additionally, measures of brand equity for both brands were elicited. Following Aaker, (1991), Keller (1993) and Agarwal and Rao (1996), brand equity is operationalised as a multi-item scale by summation of; brand attitude (good/not good; like/not like); brand reliability (reliable/not reliable); brand trust (trustworthy/not trustworthy); and perceived quality (high quality/low quality). All items were rated on a 7-point scale with total brand equity being a summated score of the individual items (maximum total brand equity score = 35, minimum total score =5). The second stage of the experiment involved a hypothetical product recall experience with the respondent’s chosen brand. The product recall experience consisted of five relevant attributes for completeness. For simplicity, each attribute was assumed to have two levels (indicated in the table below) giving a full factorial of 32 combinations. Problem type levels (serious, not serious) were designed to be non-functional so perceptions of functional attributes were unaffected. Attribute Communication (commprob) Problem Type (typeprob) Problem Attribution (blameprob) Problem Resolution (repairtime) Speed Action (speedact) Level 1 Media Release Paint Flaking – Toxic External Supplier Three week Repair Delay of six weeks 2 Level 2 Personal email from firm Paint Flaking – Non-toxic Internal (to firm) Fault Same Day Repair Immediate The product recall experience was presented to respondents in their booklet as a typical product recall notice with each respondent receiving a randomly selected combination from the full factorial. Finally, respondents were told, after successful resolution of the recall problem and no further issues with the player, they had lost the player a few weeks later and they needed to make a choice between the only two Mp3 players available; Apple and Sansa. Both had identical features (including price) to the original choice circumstance. Post-product recall brand equity measures (identical scales to initial stage) were also elicited. The experiment was undertaken by 341 undergraduates (at least 10 repetitions of full factorial) and results generated by SPSS. Results For the initial choice, 245 (71.8%) of respondents chose Apple (A) despite the price differential. Of the 245 who initially chose A, 61 (24.9%) chose Sansa (S) on the second choice occasion. For the 96 initial S choosers, 28 (13.2%) switched to Apple. This preliminary evidence may suggest that the more well known brand is penalised more by the product crisis due to higher prior expectations. The total equity measures for both brands are given in table 1 below; TABLE 1- Average Pre and Post Recall Brand Equity Measures Choice Pre-Recall A Pre-Recall S Post-Recall A Post-Recall S 29.004 14.976 25.938 15.784 (N=245) Std. Dev 3.898 5.883 5.676 6.019 SANSA Mean 24.229 19.979 24.229 20.083 (N=96) Std. Dev 7.069 6.301 7.104 6.446 APPLE Mean * Significant at 1% Diff A (Po-Pre) Diff S (Po-Pre) -3.065* 0.80816* 0 0.104 For initial Apple choosers, brand equity has decreased significantly for Apple and increased for Sansa. For initial Sansa choosers, there is no significant change in brand equity for either brand. Additionally, for Apple choosers, variability of own brand equity has increased suggesting increased uncertainty in equity evaluation. Changes to mean and variance of total brand equity for the established brand may reflect a high pre-recall equity base and/or the perceived low likelihood (pre-recall) or that problems will occur with the brand or product. Further to the above analysis, a binary logistic regression determined that pre-recall brand equity evaluations were significant in explaining initial choices (results not shown due to space limitations). Following on, differences in brand evaluations for Apple choosers were regressed against product recall attributes to determine if product recall attributes were relevant drivers of changes in total brand equity. Apple was chosen because it had significant differences in brand equity evaluation resulting from product recall while Sansa did not. Results are shown in Table 2 below; From Table 2, product recall attributes that significantly impact on the equity difference for Apple (Apple choosers) are speedact and typeprob. Moving from level 1 (serious problem, speed of initial action = 6 weeks) to level 2 (less serious problem, speed immediate) for each variable (separately) positively impacts on the difference in brand equity (decreases the negative impact in 3 this case) according with expectations. From a managerial perspective, consumers appear to significantly penalise (in terms of brand equity) firms that delay announcements of recall. Table 2- Regression of Differences Total Brand Equity (Apple) for initial Apple choosers Variable (Constant) commprob typeprob blameprob repairtime Coefficients -2.209 -0.114 0.537 0.321 -0.092 Std. Error 0.251 0.251 0.251 0.251 0.251 speedact 0.558 0.251 -0.024 0.115 0.069 -0.0197 t -8.795 -0.452 2.137 1.279 -0.366 p-value 7.66E-17 0.651 0.033 0.202 0.715 0.119 2.223 0.027 Beta 2 R = 0.084, F = 2.334 (p-value 0.04) Finally, a binary logistic regression was performed on post recall choice to examine if difference in brand equity evaluations and other relevant variables would be significant in explaining brand switching. Results appear in Table 3 below; (For the initial Apple choice group- Apple focal brand). Table 3- Binary Logistic Regression for second stage A choice (initial A choosers) Variables Coefficients S.E. Wald Post_total_eq_A 0.055 0.044 1.556 Post_total_eq_S -0.118 0.034 12.324 Diff_total_eq_A 0.144 0.05 8.336 Diff_total_eq_S 0.051 0.044 1.336 Post_totaleq_var_S 0.579 0.486 1.415 Post_totaleq_var_A -0.388 0.309 1.58 Constant 2.257 1.391 2.631 Sig. 0.212 0 0.004 0.248 0.234 0.209 Exp(B) 1.056 0.888 1.154 1.053 1.784 0.678 0.105 9.554 The Nagelkerke psuedo R2 measure is 0.30 while the % correct predictions for the model is 82%. However, switching prediction is less accurate with 41% correct predictions. Although seemingly low, it provides superior prediction than the 24.9% switching from Apple to Sansa from preliminary tabular analysis. The two significant variables are the difference in brand equity evaluation for Apple and the post recall evaluation for Sansa and both are correctly signed. The variability of brand equity for both brands does not seem to impact on brand choice. Several other models incorporating the product recall attributes were tried with the difference in brand equity included and excluded. When difference in brand equity was included, these variables were not significant but were significant when brand equity difference was excluded. This suggests, the impact of these variables is accounted for by the difference in brand equity (unlikely to be collinearity due to design). Results from Table 2, 3 suggest that product recall attributes (speedact, probtype) influence brand equity difference, which is highly significant in explaining post-recall choice. 4 Conclusions and Recommendations for Further Research. The experiment has provided some interesting preliminary results on the impact of product recall on brand equity evaluations. The results suggest that brand equity, in particular for Apple the established brand, is impacted negatively by the product recall incident while the less established brand is not affected in the same way. These results are inconsistent with results from (Ehrenberg et al. 1990) Ahluwalia et al (2000), Dawar and Pilutla (2000) which suggest that strong brands are likely to be more immune to post recall impacts than weaker brands. Possibly, strong brands are expected to perform consistently and any product recall situation leads to an erosion of consumer’s trust and re-assessment of brand and reliability. Consumers may not have the same high expectations of weaker brands. Managers of strong brands may have more to lose from product recall than managers of weaker brands. With regard to the product recall characteristics, the seriousness of the problem and the speed of company action appear to be the significant drivers (for this study) of brand equity re-evaluation. In this study, Apple in particular, was penalised for a tardy response in informing consumers of the product recall. The seriousness of the problem was also relevant; consumers downgraded their evaluation of brand equity for Apple when the problem involved perceived harm to the consumer (toxic paint flakes) relative to the benign paint flaking. Further the difference in brand equity between pre and post recall circumstances was important in explaining switching behaviour. The implication for managers is that any delays in dealing with the product recall and serious product recall problems can have serious brand equity consequences and hence lead to lower future market shares (although not shown here the magnitude of the impact can be determined from the equations given). This experiment represents a preliminary step in exploring the use of a designed factorial experiment incorporating choice to evaluate the impact of product recall experiences on brand equity. There are a number of limitations to this study that naturally provide scope for further research. The particular product attributes were limited for this experiment and more realistic experiments can expand initial product profiles and number of choices. Additionally, the number of levels for product recall attributes can be expanded for generalised results. The impact of product recall may be different across different types of product categories so investigation of different categories is needed. The impact of product recall on brand equity may also be gradual and not be reflected in a simple two period choice simulation. A simulation involving multiperiod choices with possible information acceleration may provide useful insights. Given this was a preliminary experiment, a smaller than desirable sample was used. Stronger results may be achieved with larger sample size giving more repetitions of specific recall attribute combinations. Further focus on brand equity variability measures may provide worthwhile avenues for research. 5 References: Aaker, David A.1991 Managing Brand Equity: Capitalising on the Value of a Brand Name. New York: The Free Press. Agarwal, Manoj K. and Vithala R. Rao, 1996. An Empirical Comparison of Consumer-Based Measures of Brand Equity. Marketing Letters 7 3, 237-247. Ahluwalia, Rohini., Burnkrant, Robert E., and Rao Unnava, 2000. Consumer Response to Negative Publicity: The Moderating Role of Commitment, Journal of Marketing Research, 37 (2), 203-214. Bitner, Mary Jo, 1990. Evaluating Service Encounters: The Effects of Physical Surroundings and Employee Responses. Journal of Marketing 54 (2), 69-82 Cleeren, Kathleen., Dekimpe, Marnik. G., Helsen Kristian, 2008. Weathering Product Harm Crises. Journal of the Academy of Marketing Science 36, 262-270. Davidson, Wallace and Worrell, Dan L., 1992. The Effect of Product Recall Announcements on Shareholder Wealth, Strategic Management Journal 3, 467-473. Dawar, Niraj, and. Pilluta, Madan M., 2000. Impact of Product-Harm Crises on Brand equity: The Moderating Role of Consumer Expectations. Journal of Marketing Research 37 May, 215226. Ehrenberg, Andrew., Goodhart S.C., and Barwise, Patrick T., 1990. Double-jeopardy Revisited. Journal of Marketing July, 82-91. Folkes, Valerie S. and Kotos, Barbara., 1986. Buyers’ and Sellers’ Explanations for Product Failure: Who Done it? Journal of Marketing April, 74-80. Folkes, Valerie S., 1984. Consumer Reactions to Product Failure: An Attributional Approach. Journal of Consumer Research 10 March, 398-409. Fortune, 2007. The 101 Dumbest Moments in Business. 156 3, 147-160. Geyser, Stephen A. and Klein, Norman., 1990. The Perrier Recall: A Source of Trouble. Case 9Boston: Harvard Business School, 590-104 Govindaraj, Suresh and Bikki Jaggi., 2004. Market Overreaction to Product Recall Revisited – The Case of Firestorne Tires and the Ford Explorer. Review of Quantitative Finance and Accounting 23, 31-54. Griffin, Mitch, Babin, Barry J. and. Attaway, Jill S., 1991. An Empirical Investigation of the Impact of Negative Public Publicity on Consumer Attitudes and Intentions. Advances in Consumer Research. 18, 334-341. 6 International Manager 1994, Nestle Struggles to Pump up Perrier, p.3-4. Keller, Kevin., 1993. Conceptualising, Measuring, and Managing Customer-Based brand Equity. Journal of Marketing 57 1, 1-22 Keller, Kevin., 2003. Strategic Brand Management: Building, Measuring and Managing Brand Equity, Prentice Hall, Second Edition. Keller, Kevin L. and Lehmann, Donald R., 2006. Brands and Branding: Research Findings and Future Priorities. Marketing Science 25 6, 740-759 Rhee, M., and Haunschild, P. R., 2006. The Liability of a Good Reputation: A Study of Product Recalls in the U.S. Automobile Industry. Organisational Science 17, 101-117 Rupp, Nicholas G., 2001. Are Government Initiated Recalls more Damaging for Shareholders? Evidence from Automative Recalls 1973-1998. Economics Letters 71, 265-270. Smith, Craig; Thomas Robert J., and. Quelch, John A., 1996. A Strategic Approach to Managing Product Recalls. Harvard Business Review September/October, 102-112. Van Heerde, Harald., Helsen, Kristiaan and. Dekimpe, Marnik G., 2007. The Impact of ProductHarm Crisis on Marketing Effectiveness. Marketing Science 26 2, 230-245. Weinberger, Marc G. and Lepkowska-White E., 2000. The Influence of Negative Information on Purchase Behaviour. Journal of Marketing Management 16, 465-482 7