Lessons from the Consumer Response to the 2007 Toy Recalls This Draft: October 2009
University of Maryland
University of Maryland and NBER
Rotman School of Management,
University of Toronto
Abstract: In 2007, the Consumer Product Safety Commission (CPSC) issued 276 recalls of toys and other children's products, a sizeable increase from previous years. Many of these recalls involved the industry’s largest firms and most popular brands. We use this period of recall activity to investigate how consumers respond to information about product safety. Because recalled toys must be removed from the market, this setting offers an opportunity to examine whether – and to what extent – recall announcements have spillover effects to non-recalled items. Using the most comprehensive data available for this industry, we document changes in toy sales in the months following the recall announcements. The data reveal four key findings. First, for manufacturers who had recalls in 2007, unit sales of the types of toys involved in the recall fell, relative to their toys not related to the recalls. Second, we find no evidence that these manufacturers’ sales of toys not related to the recalls were negatively affected, indicating an absence of within-manufacturer spillovers. Third, units sales of toys produced by manufacturers who did not experience any recalls fell by about 25% in 2007, suggesting there were industry-wide spillovers. Finally, recalls of toys that were part of a brand appear to have had either positive or negative effects on sales of other toys bearing that brand, depending on the nature of the toys involved. We discuss what these patterns could imply about how consumers draw inferences from recall announcements and point to possible implications for both firm strategy and public policy.
We gratefully acknowledge the helpful comments of Severin Borenstein, Jonathan Guryan, Judy Hellerstein, Ginger Jin, Arik Levinson, Soohyung Lee, Nuno Limao, and Abigail Wozniak as well as seminar participants at the University of Maryland, Rotman School of Management, the Energy Institute at U.C. Berkeley, Georgetown, UC-Davis, UC-Irvine, and University of Michigan. We thank Danny Kim at NPD for answering our questions about the toy sales data and Kevin Mak at the Rotman Finance Lab for his assistance with assembling the stock price data. Molly Reckson provided capable research assistance. Financial support from the AIC Institute for Corporate Citizenship at the Rotman School of Management is gratefully acknowledged.
In many markets, consumers have imperfect information about product attributes and demand is based in part on expectations of product quality. The case for government intervention in such markets, and the optimal nature of any such intervention, will depend on how consumers form expectations about product quality. Yet, to date, we have little empirical evidence on how consumers actually form and update expectations of product quality. For example, does information about the quality of one set of products cause consumers to draw inferences about the quality of others? At what level do these inferences take place? Are consumers more likely to draw inferences between products that share a manufacturer, characteristic or brand? How large are the resulting externalities?
In this paper, we attempt to shed light on some of these issues by investigating how consumers responded to information about product safety, an attribute about which they were likely imperfectly informed. Our setting is the U.S. toy industry. In 2007, the Consumer Product and Safety Commission (CPSC) issued 276 recalls of toys and other children's products, as compared to 152 such recalls in 2006, 171 in 2005, and 121 in 2004.1 This series of product recalls is noteworthy for several reasons. First, it represents a greater than 80 percent increase in the number of recalled children's items from 2006 to 2007 and a much larger increase than that which is observed in other categories over this period. Second, it ultimately resulted in the passage of new federal legislation – the Consumer Product Safety Improvement Act – in early 2008. Finally, in surveys and interviews conducted at the time of the recalls, consumers clearly indicated an intention to change their behavior in response to these recalls. For example, in a Harris Poll of 2,565 adults in the United States conducted in October 2007, 33 percent of respondents said that they would buy fewer toys during the 2007 holiday season due to recent safety recalls and 45 percent said they would avoid toys from China.2
Using the most comprehensive data available for this industry, we document how these recalls affected toy sales in the months following the recall announcements. It is important to note that in contrast to, say, product ratings, recalls do not provide consumers with information about the safety of products available in the market since any products that are actively selling when they are recalled are immediately removed from retailers’ shelves. Moreover, in many cases, recalls are issued for products that are no longer active in the marketplace.3 As a result, to the extent that there is a consumer response to a recall, this would indicate that consumers are using the information contained in the recall announcement to update their expectations of the safety of other products in the market. Any demand response that we measure can thus be considered a “spillover effect”. The goal of our empirical analysis is to document the level at which these spillover effects are observed in order to understand how consumers draw inferences about product safety.
Two features of our setting make an examination of consumer expectations and spillover effects particularly interesting. First, the majority of the 2007 toy recalls involved risks associated with a common industry practice of producing in China and many related specifically to the use of paint with high concentrations of lead. This feature raises the possibility that consumers took these announcements as information about the safety of an industry-wide practice (rather than as information about the safety of any particular manufacturer’s toys) and increases the likelihood that non-recalling firms might also experience demand losses.4 Second, licensing and branding are extremely common in the toy industry, with licensed products accounting for approximately one quarter of toys in the industry (Clark, 2007). Brands (such as Fisher-Price’s “Laugh and Learn” line) and trademarked characters (such as “Dora the Explorer”) are often shared across different types of toys as well as across toys produced by different manufacturers. Not only does this create another level at which consumers may draw inferences about safety but it also raises the possibility that imperfect information may prevent consumers from accurately acting upon the inferences that they draw. For example, following a recall of certain toys produced by Mattel, consumers may infer that all Mattel toys are less safe. However, if consumers do not know that toys produced under the Fisher-Price brand are, in fact, produced by Mattel, they will not be able to accurately act upon that inference. While we will not be able to test between imperfect information and various levels of inferences as explanations for the patterns we observe, we discuss the implications of each for both policy formulation and firm strategy.
Our empirical analysis uses data on monthly Infant/Preschool toy sales from January 2005 to December 2007 inclusive.5 Our empirical approach attempts to account for several important institutional features of the toy industry. In particular, the fact that toy sales are highly seasonal means that any demand response to a recall at any point in the year is most likely to occur at Christmas. However, the fact that the popularity of any particular toy or type of toy may be short-lived means that what is popular one Christmas may not be popular the following year. It is thus very difficult to establish a single appropriate counterfactual level of sales that is clearly superior to alternative counterfactual estimates. We therefore carry out several complementary analysis which, taken together, describe the patterns in the data. We begin by documenting differences in total toy sales across our three years of data. Then, we investigate the relationship between having a recall during 2007 and Christmas 2007 sales. Data limitations prevent us from estimating our models at the level of the individual toy so we instead follow standard industry practice and classify toys into “categories” (groupings of similar toys) and “properties” (groupings of toys that share a common brand or trademark) and estimate the impact of recalls on sales at the manufacturer-category level as well as the property-category level. Finally, we conclude by carrying out in-depth examinations of the largest and most widely publicized recalls from 2007.
Several key findings emerge from our analysis. First, Christmas season sales of the types of toys involved in the recalls were reduced. The results of our OLS regressions indicate that – relative to their categories that did not experience recalls – manufacturers’ unit sales in categories that did have recalls were lower by about 30 percent. In addition, in the three recalls that we investigate in detail, the manufacturer’s Christmas season sales in the affected category-property fell substantially. Thus, consumers appear to have used the information contained in the recall announcements to update their expectations about the safety of similar toys produced by the manufacturer. This suggests that a process of standards and recalls – such as that which regulates toy safety – can impose costs on firms in the form of reduced demand. These costs will provide at least some incentive for firms to invest in product safety.
Second, we find no evidence that a manufacturer’s recall of one type of toy had any negative impact on its sales of other types of toys. This suggests that either consumers did not draw inferences from a manufacturer’s recall of one type of toy about the safety of unrelated toys produced by that manufacturer or that, perhaps due to the prevalence of licensing and branding, they did not know which toys were produced by which manufacturer. If the latter is true, then the current process of recalls may needs to be supplemented with additional information provision that enables consumers to better identify which toys are produced by whom. Of course, manufacturers may have incentives to limit association between their brands and publicize any recalls that do occur under a particular brand rather than the manufacturer name.
Alternatively, the fact that manufacturers’ unrelated toys were not affected might indicate that large diversified toy manufacturers made investments in rebuilding their overall reputation (to offset any negative inferences consumers may draw) or took steps to shift demand to their brands or product lines that were not involved in recalls (to exploit consumer’s imperfect information). We present some descriptive evidence on firm diversification in relation to this conjecture. If this is the case, then the extent to which recalls will provide large firms with incentives to invest in safety will depend on the extent to which these firms have to undertake costly investments to prevent losses in unaffected categories.
Third, unit sales of Infant/Preschool toys produced by manufacturers who did not experience any recalls were about 25 percent lower at Christmas 2007 than in 2005, with no measurable change in 2006.6 Thus, the recalls appear to have had negative spillovers to the industry as a whole. Consistent with consumers’ claims in surveys and in the media, this suggests that the specific recalls that took place led consumers to draw inferences about the overall safety of toys in the market. The presence of negative externalities indicates that, in the absence of government intervention, incentives for safety provision in this industry would not be sufficient.7
Finally, with respect to the role of licensing, we find that recalls of branded toys can have positive or negative effects on the demand for other toys that share the brand. We hypothesize that the degree of similarity between the recalled toys and other toys in the property may affect the direction of the response. Specifically, when toys are very similar, consumers are both more likely to draw inferences about the safety of other toys in the property as well as more likely to be imperfectly informed about which toys were actually involved in the recall. In contrast, when the toys are less similar, strong tastes for a particular brand may lead to positive shifting within the property.
This paper contributes to two related literatures. First, it is closely related to an existing – though mostly 20 year old – literature that measures the stock market response to recalls.8 The stock market response reflects the total costs that recalls impose on firms.9 Much, though not all, of this literature focuses on drug and automobile recalls due to the high frequency of recalls in these industries. This literature includes Jarrell and Peltzman (1985), Pruitt and Peterson (1986), Hoffer, Pruitt and Reilly (1988), Dranove and Olsen (1994), Barber and Darrough (1996), and Chu, Lin and Prather (2005). With the exception of Hoffer, Pruitt and Reilly (1988), all of the papers find statistically significant negative stock price reactions to the recalls. Several of the papers compare the estimated drop in shareholder wealth to estimates of the direct costs of the recalls and find that the former exceeds the latter. They speculate that this excess loss is due to a loss of “goodwill”; this provides indirect evidence that the consumer response to recalls may be significant. Crafton, Hoffer and Reilly (1981) and Reilly and Hoffer (1983) directly measure the demand response to automobile recalls.10
Second, this paper is related to a growing empirical literature that investigates the effects of government-mandated information disclosure programs. Information disclosure policies represent an alternative way to address the problems that arise from informational asymmetries and they take a variety of forms. Economists have studied the impact of information disclosure policies on consumer and firm behavior in a variety of contexts, including restaurant hygiene grade cards (Jin and Leslie, 2003); nutritional labeling requirements (Mathios, 2000); mercury and fish consumption advisories (Shimshack et al. 2007); SEC financial disclosure requirements (Greenstone, Oyer, and Vissing-Jorensen, 2006); and environmental safety contexts, such as requirements on community water suppliers to disclose information on chemicals in drinking water (Bennear and Olmstead, 2008). Fung, Graham, and Weil (2007) and Winston (2008) review and synthesize this research and the conditions under which information disclosure programs affect consumer and/or firm behavior in ways that achieve the underlying policy objectives.11
The remainder of this paper is organized as follows. Section II provides relevant background information. Section III describes the data. In Section IV, we carry out our empirical analysis of the consumer response to the recalls. In Section V, we present additional considerations including an empirical examination of the stock market response to the 2007 toy recalls. We also take up the question of whether consumers responded to the “Made in China” aspect of the recalls. We conclude in Section VI with a discussion of the implications of our findings for both policy formulation and firm strategy.
II. Institutional Background
A. Toy Industry Basics
In 2005, the U.S. toy industry generated $21.3 billion in retail sales.12 At both the manufacturer and retailer levels, the industry is dominated by a small number of large firms. At the manufacturer level, Mattel and Hasbro together account for roughly 30 percent of the market.13 The remaining firms are considerably smaller, with the third largest firm accounting for less than four percent of the market and the tenth largest firm accounting for just over one percent of the market.
For analysis purposes, the toy industry is classified into 11 “supercategories” which are broad groupings of toys with similar uses or purposes. Examples include “Action Figures and Accessories”, “Infant/Preschool” and “Youth Electronics”. Supercategories are further subdivided into finer categories. The Infant/Preschool supercateogry which we focus on is the largest in the industry, accounting for slightly more than 14 percent of total industry sales in 2005 (about $3.2 billion). It is divided into 13 finer categories such as “Preschool Vehicles” and “Infant Plush”. The four largest firms in the Infant/Preschool market are Mattel, Leapfrog, Hasbro, and RC2, which account for 27.3 percent, 10.3 percent, 6.6 percent, and 5.5 percent, respectively. The remainder of the Infant/Preschool market is served by firms accounting for small percentages.
Branding and licensing are quite common in the toy industry. A “property” refers to a set of toys that share a common brand. The property includes all toys produced by the owner of the brand as well as all toys produced by firms who have licensed the rights to use the brand. Broadly speaking, one can distinguish between two types of properties. The first type encompasses a brand that is owned by a toy manufacturer and used on some set of that manufacturer’s toys. The manufacturer may license that brand to other toy manufacturers – but often does not – and/or may license that brand to firms producing other types of consumer products (for example, bicycles or children’s’ furniture). Mattel’s “Laugh & Learn” brand is an example of this type of property. Mattel’s Fisher-Price division produces approximately 20 different infant toys under the “Laugh & Learn” brand. The second type of property encompasses a brand that is owned by a firm outside of the toy industry and that is licensed to one or more toy manufacturers. In this case, the property would include all toys which use the licensed brand or trademark and may include products from several different manufacturers. Examples include “Spiderman”, owned by Marvel Entertainment and “Dora the Explorer”, owned by Nickelodeon. In some cases, a single toy manufacturer may obtain the exclusive rights to a license; in other cases, it may be shared by several different manufacturers.
B. The Recall Process
The recall process is initiated through one of three channels: a complaint made to the CPSC; a complaint made to the company whose product is in question; or a field sample or investigation.14 When the CPSC receives a consumer complaint or is notified of a complaint made to a manufacturer, they immediately launch an investigation; if the content of the complaint is confirmed, the agency sends a letter to the company initiating a recall process. Manufacturers, importers, distributors, and retailers are required to report to the CPSC under Section 15 (b) of the Consumer Product Safety Act (CPSA) “within 24 hours of obtaining information which reasonably supports the conclusion that a product does not comply with a safety rule issued under the CPSA, or contains a defect which could create a substantial risk of injury to the public or presents an unreasonable risk of serious injury or death, 15 U.S.C. § 2064(b).”15
The large increase in the number of recalled toys and children’s products observed in 2007, as compared to earlier years, is unique to this category of products. Panel A of Table 1 reports the number of recalls per year in major categories of consumer products from 2004 through 2007. The number of toy recalls was 30, 31, and 38, respectively, for 2004, 2005, and 2006. That number jumped to 82 in 2007. For children’s products the numbers are 42, 64, and 56, with a jump up to 130 in 2007. The other categories do not show such a discrete increase in 2007.
A. Recall Data
We collect details about the toy recalls that took place between 2004 and 2007 from the CPSC website. For each recall, the CPSC website lists the date of the recall, the product name, the number of units recalled, the importer, manufacturer, and/or d16istributor, a description of the hazard, details about any reported incidents or injuries, a description of the product to assist in identifying recalled items, details about where and when the item has been sold, the typical price, where the item was manufactured, and a picture of the item.
Panel B of Table 1 shows changes in the characteristics of toy recalls from 2004 to 2007. There are two interesting patterns to note. First, although the majority of recalls in each year involve toys made in China, there is a noticeable increase in 2007, when 95 percent of recalls involved toys manufactured in China. Second, there has been a change in the types of safety hazards leading to recalls. Prior to 2007, 13 percent of recalls were due to lead paint and 49 percent were due to choking; in 2007, these numbers were 52 percent and 20 percent, respectively.17
The concern about lead paint in children's toys is largely driven by the fact that young children often put toys in their mouths and are thereby exposed to the lead content of paint. Lead is a powerful neurotoxin that interferes with the development of the brain and central nervous system as well as the kidney and blood-forming organs. Lead poisoning in children is generally associated with behavioral problems, learning disabilities, hearing problems and growth retardation.18 The federal legislation enacted in 2008 requires that surface lead, as in paint, must drop below 90 parts per million by August 2009, compared to the existing statutory level of 600 parts per million.19
B. Sales Data
We combine the recall data with data on the U.S. sales of toys in the Infant/Preschool toy supercategory from January 2005 through December 2007. We purchased this data from the NPD Group, self-described as the “single source for toy market research in the U.S., Europe, and Australia.” The NPD data is based on a panel of more than three million consumers.20 The panel is comprised of two sets of consumers: (1) an online panel of consumers who are instructed to record all of their purchases; (2) a panel of consumers who have scanners in their homes who are supposed to scan everything they buy. From these two panels, NPD generates a toy level dataset with both actual data from the panels (e.g. the number of transactions observed for each toy each month, the average price paid) as well as projected monthly unit and dollar sales figures (for the country). It is the latter measures that we use in our empirical analysis. After dropping observations for which no manufacturer information is available, our dataset includes data from a total of 156,524 transactions and 10,847 unique items over the full period.
There are three important features of our data. First, the data are generally not reliable at the item level. Because the data are based on a sample of consumer purchases and because the toy industry is highly fragmented at the product level, most of the toys in the dataset are only involved in a small number of transactions (NPD cautions against drawing inferences from cells with fewer than 35 transactions). In fact, the majority of toys have zero transactions in any given month.21 Because NPD does not keep records of market exit, we are unable to determine whether zero transactions indicates that no consumers chose to purchase that toy in a given month or if the toy was no longer supplied.22 In the toy industry, new toys are introduced frequently and current toys are either replaced or updated with new features so exit may be an important consideration. For these reasons, we have no choice but to aggregate the item level data over time and/or groups of items. In particular, we focus our regression analysis on sales at the level of the manufacturer-category and manufacturer-property. In our case studies, our level of analysis is generally the manufacturer-property-category.
Second, our data does not include consumer level variables. Therefore, though it would be interesting to explore consumer responses by retailer type or consumer demographics, we are unable to do so in this paper. Third, toy sales are highly seasonal. Roughly half of toy sales occur in the form of Christmas season purchases. An event-study type methodology is thus inappropriate for analyzing this data because the demand response to a recall will likely not occur immediately. We instead focus our analysis on fourth quarter sales, which include purchases made in October, November, and December of a given year.