The cloud and its silver lining: negative and positive spillovers from automotive recalls

Product recalls hurt the sales of non-recalled products in the category because of negative spillovers. Recently, there has been some evidence of positive spillovers from recalls on the sales of non-recalled products. We focus on spillovers from brand- (i.e., same brand), firm- (i.e., same firm, but not same brand), and country-level (i.e., same country-of-origin, but not same firm) recalls on the sales of non-recalled products. Furthermore, we examine how advertising and price of non-recalled products interact with brand-, firm-, and country-level recalls to affect their sales. We use data on 124 cars in the USA in 2006–2015. Results indicate that brand-level (country-level) recalls hurt (benefit) the sales of non-recalled products. Higher advertising and price of non-recalled products weaken the negative effect of brand-level recalls, while lower advertising and price strengthen the positive effect of country-level recalls. Finally, firm-level recalls result in positive spillovers when advertising is high.


Introduction
Product recalls hurt the sales of recalled products and of some non-recalled products in the category because of negative spillovers (Cleeren et al., 2013). Furthermore, as products in a category are substitutes, recalls may, sometimes, benefit the sales of non-recalled products.

Definitions
The term "product" (or "make-model" 1 ) refers to a specific car (e.g., Toyota Camry). The term "brand" (or "make") refers to a family of products with a brand name (e.g., Toyota). The term "firm" refers to the manufacturer (e.g., Toyota Motor Corporation). Finally, the term "COO" refers to the country-of-origin of the firm (e.g., Japan).
Using a product as unit of analysis, "brand-level recalls" are recalls of products of the same brand as the focal product, "firm-level recalls" are recalls of products of other brands of the same firm as the focal product, and "country-level recalls" are recalls of products of other firms from the same COO as the focal product. If Toyota Camry is the focal product, the recall of Toyota Corolla is a brand-level recall, the recall of Lexus ES is a firm-level recall, and the recall of Honda Accord is a country-level recall.

Main effects
We rely on two theories. First, consideration set formation theory (Hauser, 2014), which posits that consumers, when considering a purchase, first form a consideration set of products and then choose a product from this set. Second, accessibility-diagnosticity theory (Feldman & Lynch, 1988), which posits that, if a consumer thinks product A is informative about product B, she will use information about A to make inferences about B.
Brand-level Applications of the accessibility-diagnosticity theory to recalls (Roehm & Tybout, 2006) indicate that, when a recall is revealed, its effects are reflected not only on the recalled product but also on non-recalled products similar to it on aspects related to the recall (e.g., materials, plants, etc.) (Borah & Tellis, 2016). Products of the same brand share the same materials, plants (Hora et al., 2011), and esthetic features (e.g., BMW Kidney Grille). A recall will change the consideration set of consumers willing to buy the recalled product or a product similar to it. Consequently, consumers may exclude from (not include in) their consideration set (Palazzolo & Feinberg, 2015) the non-recalled product of the recalling brand, as they may consider it likely to incur recalls itself. Hence, brand-level recalls will decrease the sales of the non-recalled product. We do not formally hypothesize this relationship, given the extensive evidence for it in the literature (Liu & Shankar, 2015).

Country-level
A product's COO may affect consumers' evaluations of specific attributes (Johansson et al., 1985). As an example, German cars are considered fuelefficient, while US cars are considered powerful (Erickson et al., 1984). However, a product's COO influences consumers' purchase decisions only when information is unclear (Verlegh et al., 2005), serving as a heuristic under conditions of uncertainty.
When a product from a given COO is recalled, consumers who had previously included it in their consideration sets may no longer do so along with other products of the same brand. As a result, they will engage in a consideration set substitution process (Palazzolo & Feinberg, 2015). Consumers may rely on a heuristic to develop their new consideration sets (Maheswaran, 1994) and include products from the same COO as the recalled product, expecting them to be similar on desirable attributes (Johansson et al., 1985) to the excluded products. Anecdotal evidence supports this mechanism. In 2010, sales of Toyota make-models recalled for accelerator problems were down 13%. Toyota Camry midsize sedan, usually the best-selling car in the USA, recorded a 20% decrease being outsold by the equally Japanese Honda Accord (Bunkley, 2010). Hence: H 1 : The higher the number of country-level recalls, the higher the sales of a non-recalled product from the country.
We note that this logic runs contrary to Borah and Tellis (2016), who argue that consumers make similarity inferences for products from the same COO as they expect them to have similar processes of product development, so that country-level recalls will lead to negative spillovers. We suggest that this may not be the case as this mechanism is more likely to occur at the brand-level. Different firms from the same COO do not necessarily share suppliers nor processes, crucial aspects when consumers evaluate a non-recalled product's likelihood to incur recalls. Moreover, products from the same COO are often perceived as sharing attributes such as driving comfort (Erickson et al., 1984;Johansson et al., 1985), being substitutes in consumers' consideration sets.
Firm-level At first blush, it may appear that the mechanism discussed above for brand-level recalls also applies to firm-level recalls. The negative information of a recall may be regarded as diagnostic of non-recalled products of the same firm. Hence, firm-level recalls may decrease the sales of a non-recalled product. However, other reasons suggest the opposite. First, brands are often managed autonomously from firms (Kalaignanam et al., 2013), sharing neither ownership, production processes, nor materials. As a result, negative information of a recall may not spill over to non-recalled products of other brands of the same firm. Second, consumers may not be able to ascribe each brand to its firm. This is particularly true in the automotive industry (Gorzelany, 2019). If consumers are not aware of the branding structure in the category, they may assume the industry has only three levels, i.e., product, brand, and COO. As a result, they may (mis)interpret firm-level recalls as country-level recalls. Due to these countervailing mechanisms, we remain agnostic on the nature of the effects of firm-level recalls.

Interaction effects
We expect brand-level recalls to differ from country-level recalls on a key aspect. When a product of a given brand is recalled, consumers may consider negative information diagnostic not only of the recalled product but also of non-recalled products of the same brand, whose perceived quality will be negatively affected. Conversely, negative information may not be considered diagnostic of non-recalled products of different firms from the same COO. Such products' perceived quality will not be negatively affected (see H 1 ). Against this backdrop, we formulate the hypotheses for the interaction effects.

Brand-level
Advertising Advertising is a signal of quality and can differentiate the non-recalled product from the recalled products of the same brand (Xiong & Bharadwaj, 2013). As Roehm and Tybout (2006) note, when consumers are primed to consider differentiation across products, the recalled product is isolated from non-recalled products. Thus, advertising may help the non-recalled product to be retained in the consumer's consideration set, by signaling its superior quality with respect to the recalled products of the same brand. Hence: H 2adv : The higher the advertising of a non-recalled product, the weaker the negative effect of brand-level recalls on its sales.
Price As with advertising, price is also a signal of quality (Blattberg & Wisniewski, 1989). As recalls make safety more salient to consumers (Chen et al., 2019), and consumers grow less price-sensitive (Zhao et al., 2011), consumers may interpret a lower price, following brand-level recalls, as a signal of low quality as, as mentioned above, non-recalled products are expected to incur recalls when products of the same brand are recalled. Hence, we reason that, among products of the same brand as the recalled products, those with a higher price will be more likely to be included in (chosen from) the consideration set. A higher price, in fact, may be perceived as a signal that the non-recalled product is of higher quality compared to the recalled products of the same brand, weakening the negative effect of brand-level recalls. Anecdotal evidence supports this mechanism. In 2015, in the 2 months following Honda's recall of Accord and Civic due to airbag problems, discounts for the non-recalled Honda Fit were more than halved, resulting in a substantially higher price for consumers. Hence: H 2price : The higher the price of a non-recalled product, the weaker the negative effect of brand-level recalls on its sales.

Country-level
Advertising We expect opposing effects here. On the one hand, higher advertising may signal the quality of non-recalled products from the same COO as the recalled products to consumers who have grown more concerned about quality following recalls. On the other hand, however, inferred motivations may influence the effectiveness of advertising (Eagly et al., 1978). Higher advertising following country-level recalls may backfire as consumers believe this to be an opportunistic attempt to take advantage of the misfortune of the recalling brand (Cleeren, 2015;Cleeren et al., 2013;Siomkos et al., 2010), lowering product evaluations. This may explain why advertising for Honda Accord went down in the months following the 2010 Toyota recalls. We expect the second mechanism to outweigh the first as, as mentioned above, countrylevel recalls should not raise concerns regarding the quality of non-recalled products and their likelihood to incur recalls. 2 Consistent with this reasoning, we expect higher advertising to weaken the positive effect of country-level recalls. Hence: H 3adv : The higher the advertising of a non-recalled product, the weaker the positive effect of country-level recalls on its sales.
We do not expect this mechanism to hold for brand-level recalls. In the case of brand-level recalls, the non-recalled product is of the recalling brand and, as a result, cannot be considered opportunistic.
Price Again, we expect opposing effects. On the one hand, a higher price may signal the quality of non-recalled products from the same COO as the recalled products to consumers who have grown more concerned about quality following recalls. Furthermore, as inferred motivations may influence the effectiveness of price reductions (Campbell, 1999), a lower price following country-level recalls may backfire as consumers believe this to be an opportunistic attempt to take advantage of the misfortune of the recalling brand (Chen et al., 2019;Cleeren, 2015;Cleeren et al., 2013;Siomkos et al., 2010), lowering product evaluations. As an example, the executive VP of Honda's US subsidiary told dealers, after Toyota's 2010 recalls: "We will not react in a predatory way towards either Toyota or Toyota customers" (Simon & Jung-a, 2010). On the other hand, however, a lower price of the non-recalled product may strengthen the positive effect of country-level recalls, as consumers may choose, among products from the same COO as the recalled products, the one that offers superior value, i.e., a lower price. We expect this mechanism to outweigh the previous as country-level recalls, differently from brand-level recalls, should not raise concerns regarding the quality of non-recalled products and their likelihood to incur recalls. 3 Consistent with this reasoning, we expect a higher price to weaken the positive effect of country-level recalls. Hence: H 3price : The higher the price of a non-recalled product, the weaker the positive effect of country-level recalls on its sales.
For the reasons mentioned above regarding the main effect of firm-level recalls, we remain agnostic on the interaction effects of firm-level recalls.

Data
We use the US passenger cars industry as the context as numerous make-models can be targets of spillovers at any point in time. Passenger cars' recalls are highly publicized events resulting in consumers' awareness (PMI, 2015).
We collected data on passenger cars' recalls in 2005-2014 from the NHTSA. 4 To focus on recalls of (substantive) brands, we included 21 large brands. These brands constitute more than 90% of the sales of passenger cars in each year in 2006-2015.
We excluded make-models priced over US $150,000 because they are unlikely to be in the consideration set of mainstream shoppers (see Liu et al., 2017 for a similar logic). As only sufficiently large recalls are likely to affect perceptions, we excluded recalls involving less than 15,000 units. Smaller recalls typically involve batches produced in specific plants during circumscribed periods (see Cleeren et al., 2013 for a similar logic). When the database contained multiple recalls for the same make-model in a month, to avoid overcounting, we only considered one recall (Thirumalai & Sinha, 2011).
We collected monthly sales volumes for all make-models in 2006-2015 from Wards' Auto. Wards provides data on attributes, annual manufacturer-suggested retail list prices (MSRP), and monthly promotions. Furthermore, we collected data on make-model monthly advertising from Kantar.
Finally, for each month, we excluded all make-models that were recalled at least once in the previous (rolling) year. This allows us to distinguish the effects of other make-models' recalls from those of the focal make-model. The final sample is a panel of 4744 make-model-months in 2006-2015. For a list of make-models, see Table A1, Web Appendix A. There are 559 recalls in our sample. On average, each month, 4.37 make-models are recalled. Furthermore, each month, there are, on average, 0.24 recalls per brand, 0.35 per firm, and 0.89 per COO.

Measures
We provide the descriptions for variables in Table 1 and, in Table B1, Web Appendix B, the descriptives and correlations. All VIFs are below 10.

Estimation
As data are nested (make-model-month observations are nested in brands, which are nested in firms, which are nested in COOs), we estimate a random intercepts mixed model. Following prior research (Liu & Shankar, 2015;Liu et al., 2017), we consider attributes (e.g., horsepower, size) to be exogenous. In the automotive industry, in fact, attributes are redesigned when a new make-model is introduced, approximately every 5 years (Palazzolo & Feinberg, 2015). This assumption is reasonable, as firms cannot change the attributes of cars in the short-term (Sudhir, 2001).
Advertising and price are presumably endogenous as make-model-level omitted variables may affect both marketing mix and sales. Please see Web Appendix C for a discussion of how we mitigate endogeneity concerns using a control function approach (Petrin & Train, 2010).
where μs are parameter estimates, subscript t months, ε A1t s-ε A2t s the errors from equations c1-c2 (see Web Appendix C), and ε 1t observational error. The model includes random effects for COO, firm, brand, and make-model. Potentially endogenous variables are lagged by 1 month. Standard errors are bootstrapped (200 replications).

Results
We report a main effects' model in column 1, Table 2. In column 2 we include control variables. We report the results for the main model in column 3 (log likelihood = − 5078.75). Brand-level recalls decrease the sales of the non-recalled product (b = − 0.73, p < 0.05), while country-level recalls increase them (b = 1.62, p < 0.01) in support of H 1 . There is no effect of firm-level recalls (b = − 0.70, p > 0.10).
Advertising (b = 0.01, p < 0.05) and price (b = 0.06, p < 0.05) of the non-recalled product weaken the negative effect of brand-level recalls, in support of H 2adv and H 2price . Furthermore, advertising (b = − 0.02, p < 0.01) and price (b = − 0.13, p < 0.01) of the non-recalled product weaken the positive effect of country-level recalls in support of H 3adv and H 3price . Finally, the effect of firm-level recalls is positive when advertising is high (b = 0.04, p < 0.01). The interaction between price and firm-level recalls is not significant (b = 0.03, p > 0.10). We display the marginal effects for the interactions in Figure D1, Web Appendix D. We report robustness checks in Web Appendix E.

Discussion
The findings extend the recalls literature. First, while research focused on the effects of recalls on recalled products, we extend the narrower literature on spillovers, which has hitherto focused on negative spillovers (see e.g., Van Heerde et al., 2007). In doing so, we develop a comprehensive model of the effects of recalls by focusing on both negative and positive spillovers. Consistent with previous research (Liu & Shankar, 2015), the findings indicate that brand-level recalls reduce the sales of non-recalled products. Furthermore, country-level recalls increase the sales of nonrecalled products. In doing so, we identify a silver lining and, indeed, an opportunity for non-recalled products in the face of country-level recalls. The finding extends the literature on COO effects by showing that they play a different role than previously suggested. The results apparently contradict Borah and Tellis (2016), as discussed above; we conjecture that our focus on sales (vs. online chatter) may explain this divergence. Furthermore, our examination relies on a substantially different sample and timeframe.
Second, we depart from the literature, which focused on how recalled products can use their marketing mix. In doing so, we focus on how the marketing mix of non-recalled products can weaken (strengthen) negative (positive) spillovers.
Third, this research is the first to separately examine the effects of other products' recalls at multiple hierarchical levels in a category. While brand- (Roehm & Tybout, 2006) and firm-level (Zavyalova et al., 2012) recalls have been investigated, past research has not distinguished them. Distinguishing across brand-, firm-, and country-level recalls allows to disentangle how non-recalled products with different degrees of relatedness to the recalled product are likely to experience spillovers. We show that perceived similarity on processes of product development, which we expect at the brand-level, leads to negative spillovers, as consumers will consider a product that shares product development processes with the recalled product likely to have defects. Conversely, perceived similarity on relevant attributes (e.g., driving comfort), which we expect at the country-level, Log likelihood − 7337. 78 -5114.87 -5078.75 leads to positive spillovers, as consumers will consider a product with the same desirable attributes as the recalled product a good substitute for it. For managerial practice, managers are likely aware that brand-level recalls decrease the sales of a non-recalled product. However, higher advertising weakens such negative effect by signaling that the non-recalled product is different from the recalled one. Furthermore, a higher price shields the non-recalled product from the negative effects of brand-level recalls by reaffirming its quality (interestingly, the marginal effects show that, for expensive products, the effect of brand-level recalls can be positive). Such findings are useful when multiple products are offered under the same brand, as they allow managers to effectively weather the negative consequences of brand-level recalls.
Furthermore, managers should be cognizant that country-level recalls increase the sales of the non-recalled product. Lower advertising and price enhance the positive effect of country-level recalls on the sales of the nonrecalled product. This is useful when products from a given COO are characterized by attribute-related stereotypes (e.g., US appliances, Japanese electronics).
The findings show that, at the aggregate level, firm-level recalls do not affect the sales of non-recalled products. We reason that only some consumers know which firm owns a brand. For them, firm-level recalls display a negative effect, paralleling brand-level recalls. Conversely, for consumers who are not aware of the branding hierarchy in the category, firm-level recalls display a positive effect, paralleling country-level recalls. Interestingly, the marginal effects show that the effect of firm-level recalls turns from negative to positive as advertising increases. We argue that, from a tactical perspective, managers in the automotive industry should try not to emphasize the parent firm of their brand in their communication efforts, instead emphasizing its COO. If a make-model of another brand from the focal firm is recalled, in fact, make-models of the focal brand will be shielded by negative spillovers. Instead, they will experience positive spillovers, as the recall will be interpreted as a country-level recall.
The findings generate implications for policy-makers. Some recalls are particularly severe and may damage the economy of a country. We show that, even when a recall involves products from a given COO, nothing is lost as non-recalled products from the same COO can benefit from it.
Due to the peculiarities of the automotive industry, future research investigating the generalizability of our findings will be useful.