2. Motivation

Advertising a brand-name product on television, particularly national television, can be an expensive marketing strategy. To help maximize advertising benefits, firms may choose to engage in targeted advertising. Thus, targeted advertising can be an integral part of a firm's marketing strategy in industries where firms compete with differentiated products. Iyer, Soberman, and Villas-Boas (Reference Iyer, Soberman and Villas-Boas2005) show that targeted advertising can be even more valuable than target pricing in a competitive environment. Johnson (Reference Johnson2013) shows that improved targeting of advertising can raise the profits of firms. In a large-scale field experiment, Goldfarb and Tucker (Reference Goldfarb and Tucker2011a) show that targeting online advertising to specific website content increased consumer purchase intent. In a separate study, Goldfarb and Tucker (2011b) examine privacy regulations that restrict advertisers from collecting user data in order to target their advertisements. They find that such restrictions made online advertising far less effective at changing stated purchase intent. Other research on targeted advertising focuses on the price of advertising in the newspaper industry (Chandra, Reference Chandra2009) and the television industry (Goettler, Reference Goettler2012).

The breakfast cereal industry is ideal for examining targeted advertising for several reasons. First, breakfast cereal is a heavily consumed product in the United States, and competing firms in the industry rely on vast amounts of product differentiation. The product category includes several consumer segments (children, adults, etc.) and offers a wide variety of flavor profiles (chocolate, fruity, etc.) and nutritional content (high fiber, high sugar, etc.) within each segment. Consistent with such large-scale product differentiation, the breakfast cereal industry also employs a significant amount of television advertising. In fact, cereal companies spent $264 million on advertising in 2011 (http://cerealfacts.org). Advertising in the cereal industry has been documented to have an advertising-to-sales ratio four to six times higher than other food industries (Nevo, Reference Nevo2001). Cereal has also been consistently the second most advertised product (following fast-food restaurants) to children aged 2–11 from 2004 to 2015 (Frazier and Harris, Reference Frazier and Harris2016).

With regard to breakfast cereals, manufacturers recognize that sugar appeals to children. In fact, General Mills refers to a “sweetness threshold” of roughly 9 grams of sugar per serving, which is “the breaking point where the sugar level is so low that the sweetness is not enough for a kid to eat it on day two after trying it on day one” (Jargon, Reference Jargon2011, p. B1). Given the demand for sugary cereals, we might expect that sugar content would be a primary driver of what products are advertised, particularly to children. With regard to adults, however, sugar may not be an advantageous characteristic, as adults face a completely different choice set for managing their health. To this point, Schwartz et al. (Reference Schwartz, Ross, Harris, Jernigan, Siegel, Ostroff and Brownell2010) found that as the nutritional profile of a cereal became worse, the cereal was more heavily advertised to children. A heart-healthy breakfast cereal advertisement may fall on deaf ears if targeted to children, just as a Cocoa-covered, high-sugar cereal advertisement may not appeal to adults. In fact, Berning, Huang, and Rabinowitz (Reference Berning, Huang and Rabinowitz2014) find that the average nutritional content of advertised children's cereals is one with more fat and sugar and less protein.

In addition to nutritional quality differentiation, breakfast cereals utilize their brand packaging to an extent that few other products do. Breakfast cereal packaging is often covered in brand characters, promotional opportunities, nutritional claims, and other engaging marketing strategies. Product brand images such as Toucan Sam, the Trix rabbit, and Cap'n Crunch are identifiable characters that entice young consumers. Consequently, products that have certain brand profiles may be ideal products to advertise to specific target audiences. This is an issue that is not just isolated to food products as it has been documented that tobacco advertisers had been known to use characters to appeal to young audiences (Kelly et al., Reference Kelly, Slater, Karan and Hunn2000). This practice subsequently ended with tobacco manufacturers being subject to legal agreements and government policies prohibiting such marketing practices.

Brand profiles are also enhanced by product packaging that often includes other attractions. This includes things like games on the box, toys in the box, and other forms of brand enhancements. In particular, children's cereals often make use of what are called advergames, which integrate brand characters or identify with online video games (Lee et al., Reference Lee, Choi, Quilliam and Cole2009). In a study of 7- and 8-year-olds, Mallinckrodt and Mizerski (Reference Mallinckrodt and Mizerski2007) found that although the children understood the persuasive intent of a Froot Loops advergame, the children who played the game still showed greater preference for Froot Loops over other cereals.

Further, breakfast cereal packaging is used to promote product cobranding. This is a brand alliance strategy between two or more products being simultaneously offered to consumers (Geylani, Inman, and Ter Hofstede, Reference Geylani, Inman and Ter Hofstede2008). The expectation is that cobranding can help gain increased access to new markets and can signal reputation and quality (Rao, Qu, and Ruekert, Reference Rao, Qu and Ruekert1999; Rao and Ruekert, Reference Rao and Ruekert1994). Examples of cobranding include cereal boxes with television or movie themes, athlete endorsements, or cartoon endorsements. If a cereal company is attempting to attract consumers by cobranding its product, such cobranding efforts might affect the decision to advertise and who to target with such advertisements.

The focus of this research is to examine the relationship of observable product characteristics, including cereal packaging and cobranding, and the promotion of the product via television advertising. Although Schwartz et al. (Reference Schwartz, Ross, Harris, Jernigan, Siegel, Ostroff and Brownell2010) previously examined the nutritional quality of cereal advertised to children, they relied on a nutrition score, which aggregates various nutrients. Our analysis allows us to examine the influence of specific nutrients, in particular, sugar. In addition, Schwartz et al do not account for other types of product characteristics, such as packaging and cobranding. In their analysis of advertising, Berning, Huang, and Rabinowitz (Reference Berning, Huang and Rabinowitz2014) disaggregate the nutritional content of cereals, but they also do not consider other product characteristics. Finally, neither of the previous analyses accounts for intermittent pulse advertising, where firms strategically advertise in pulses rather than continuously. We discuss this in further detail in the following sections.

3. Data

We compile a unique data set to estimate our empirical model. For the cereal advertising data, we use Nielsen Media Research weekly national advertising data from 2006 to 2008 for 113 different breakfast cereals. These data include, by brand, advertising expenditures and GRPs broken down into three groups: ages 2–11 (children), ages 12–17 (teens), and ages 18 and up (adults). GRPs are a measure of the reach (percentage of the targeted segment seeing the advertisement) times the frequency (number of times viewed) of a television commercial as calculated by Nielsen based on measured viewing of participating Nielsen U.S. TV families. For example, if 40% of the target audience views an advertisement five times in a week, the GRPs for that segment are 200. Alternatively, if 100% see the advertisement twice, GRPs are 200.

Looking at the advertising GRPs, there are relevant differences in the advertising strategy for different products. Figure 1 provides an example of advertising GRPs to the children's group for three popular cereal brands. Honey Nut Cheerios illustrates a cereal that is pulse advertised continually during our data period. There are frequent spikes in advertising GRPs and occasional zeros in advertising. Honey Bunches of Oats shows a similar pattern of advertising over the entire data set, but with a greater frequency of zeros. This provides an important contrast regarding how two cereals can be advertised at different levels of intensity. Further, the zero advertising that we observe reflects a pause in advertising rather than a complete stop in advertising during our data period. Finally, Cap'n Crunch demonstrates pulse advertising as well, but only for 1 year before it is no longer advertised.

Figure 1.

Children's Advertising Gross Rating Points (GRPs) for Three Breakfast Cereals

We cannot distinguish between advertising that completely stops and advertising that stops during our data period and then resumes after our data period. For instance, certain cereals could have a break between periods when they are advertised, but we do not observe the continuance of advertising after the break. This should not affect our analysis, but limits the external validity of our results to the cereals we are examining during our time frame. However, this is relatively extensive and inclusive of national advertising in the United States.

We use the Nielsen data to identify breakfast cereals that are advertised from 2006 to 2008. Then, we use several other data sets to identify the characteristics of the products being advertised during that time. The first data set is the Mintel Global New Products Database (http://www.mintel.com/global-new-products-database). Mintel provides photographs of most new grocery store products in U.S. markets since 1985. This includes any changes to product packaging as well. This allows us to identify if a product has a brand character, a nutritional claim, or even some form of cobranding during our data set. It also allows us to capture nutrition information for a large share of products over time.

Although Mintel has product packaging information for all the cereals advertised in our data, nutrition information is sometimes lacking. For products for which Mintel does not have nutrition information, we rely on the U.S. Department of Agriculture, Agricultural Research Service National Nutrient Database (2006–2008) (https://ndb.nal.usda.gov/ndb/). This allows us to gather fairly accurate nutrition data.Footnote ² We also rely on the Canadian Nutrient File database provided by Health Canada (2010) (http://www.hc-sc.gc.ca/fn-an/nutrition/fiche-nutri-data/index-eng.php) and manufacturer's websites for the few remaining cereals that are not included in the other databases. Although the accumulated data may not reflect changes in the nutritional quality of breakfast cereals to the exact day, we do not perceive this as a major problem. Breakfast cereals do not have large changes in their ingredients. That is, a children's cereal with high sugar content does not suddenly become an adult cereal with high fiber. As such, the nutritional data we compiled represents the most accurate data available.

Our final data set includes the product, packaging, and nutritional profile of cereals advertised on television from 2006 to 2008. As cereals vary in both package and serving size, we normalize the nutritional data by grams. As shown in Table 1, there is a wide variance in the nutritional content of the breakfast cereals in our data set. The average serving size is about 35 grams, or just over 1 ounce.

Table 1.

Nutrition Profile of Breakfast Cereals in Data Set

Looking at the promotional data (Table 2), we can see that breakfast cereals are more heavily advertised to children than adults and teens. In addition, there is a tremendous range in advertising GRPs to all groups, especially children. A large number of cereals (46.5%) have nutritional claims on their box. Other common types of product promotion include a game on the box (29.4%), a toy in the box (27.8%), and the use of a brand character (38%). Television advertising of cobranded cereals is relatively common for brand movie characters (18.1%) and cartoon characters (8.6%), but less so for TV characters (3.2%) and sports characters (4.4%).

Table 2.

Breakfast Cereal Promotion Data Summary Statistics

Note: GRP, gross rating point.

To provide a sense of the type of cereal products that are advertised, we calculate the nutritional profile of the GRP-weighted average cereal and compare it with three other popular cereals (Table 3). Specifically, we use children's GRPs to create our hypothetical cereal product. Although most of the nutritional content is fairly similar, the weighted average sugar content of advertised cereals is higher than Cheerios and Corn Flakes and similar to Frosted Flakes. Cheerios also has higher fiber and protein than the average cereal advertised to children in our data set.Footnote ³

Table 3.

Nutrition Profile of GRP-Weighted Average Cereal versus Selected Cereal Products

4. Empirical Approach

In our analysis, we treat product advertising as a function of the brand characteristics. Specifically, we specify firm i’s advertising of brand j at week t as

(1) $$\begin{equation} Adv_{ijt}^g = {f^g}\left( {{X_{jt}},{\rm{\ }}firm{\rm{\ }}pric{e_{jt}},mkt{\rm{\ }}comp_{ - it}^g,{\rm{\ }}{{\rm{\theta }}_i},{{\rm{\theta }}_t},\bm{\beta },{\varepsilon _{ijt}}} \right), \end{equation}$$

where Adv is the number of advertising GRPs for three age groups (g): children, teens, or adults. Disaggregating by these age groups allows us to identify how firms advertise their products to specific market segments. The matrix X includes product nutritional characteristics, box characteristics, product age, and different forms of product cobranding. Box characteristics include whether there is a game on the box, an online game advertisement, a product coupon, a toy in the box, or a nutritional claim on the box and whether there is a brand character appearing on the box (e.g., Toucan Sam). Cobranding includes whether the box includes the image of a movie, cartoon, or television character or a celebrity/athlete. Firm price is the price paid for advertising calculated as the total expenditures for all breakfast cereal advertising divided by the total number of GRPs for all age groups for brand j at time t. The aggregate GRPs purchased by all other firms at time t for age group g are denoted by mkt comp. This is included to capture the competitive effects of other firm advertising. β is a vector of parameter estimates, ε is an independent and identically distributed error term, and f is a functional form that we assume to be linear and different for each age group.

In equation (1), we consider the product brand, including its characters and images, as given for any week in our data set. That is, we assume that prior to selecting advertising for the product, the firm has already determined its characteristics. Conceptually, this could be thought of as a hedonic advertising function where the “market-determined” product characteristics dictate the extent of advertising.

We recognize, however, that for certain cereal products, both firm price and cobranding variables may be endogenous to the advertising decision. First, there may be unobserved market characteristics that affect both advertising and the cobranding or pricing decisions. Failing to control for these would result in bias. Alternatively, a firm could potentially change its cobranding of a particular cereal after having made the decision to advertise the cereal or at the same time. Such reverse causality or simultaneity would also bias estimates of the relationship of product characteristics and television advertising.

To explicitly mitigate potential endogeneity because of omitted variables, we include firm fixed effects (θ _i ) to account for time invariant unobservables that might bias estimates of cobranding. To account for temporal factors that could lead to endogeneity, we include seasonal dummy variables, dummy variables for each quarter, and a week time trend (θ _t ). However, there could still be time-variant unobservable factors that bias our estimates to some degree. To examine this, we also estimate models where we include quarter dummy variables interacted with each firm effect.

By using weekly observations, we avoid simultaneity that might occur because of aggregation over longer time periods such as a month. There could still be simultaneity bias, however, even if reverse causality is not present. To account for this, we also estimate equation (1) with lagged values for box characteristics and product cobranding. This provides a stronger case that these other marketing characteristics were determined prior to the television advertising decision.Footnote ⁴ We do not include lagged nutritional values because they do not vary across the majority of cereals in our sample. In our analysis, we evaluate a 1 week and 1 month lag.

Nelson (Reference Nelson2008) and Nelson and Young (Reference Nelson and Young2008) use a similar approach to examine whether alcoholic beverage companies target youths with magazine advertisements. Specifically, they model beverage company demand for magazine advertising in terms of magazine readership characteristics. Overall, they find no evidence of targeting teens with advertising for alcoholic beverages. One concern with Nelson and Young's approach that we try to address is that they do not control for intermittent advertising, which is prominent in television advertising. That is, firms often engage in pulse television advertising, which is defined by intermittent advertising rather than a constant stream of advertising. The use of pulse advertising derives from the expectation that consumer response to advertising is a sigmoidal function with marginally decreasing returns to advertising (Feinberg, Reference Feinberg1992; Simon, Reference Simon1982). From an empirical standpoint, there are two different data-generating processes to consider with pulse advertising: the decision to advertise and how much to advertise.Footnote ⁵ Failing to control for the first process will bias estimates of the second. As a result, empirical results will tend to underpredict zero advertising and overpredict positive advertising values.

To this point, we explicitly define the decision of the firm to advertise as

(2) $$\begin{equation} {y_{ijt}} = g\left( {{X_{jt}},{\rm{\ }}market{\rm{\ }}pric{e_{ - it}},{\rm{\ }}{{\rm{\theta }}_i},{{\rm{\theta }}_t},\delta ,{\xi _{ijt}}} \right), \end{equation}$$

where y is a discrete variable that equals 1 if the firm advertises and 0 otherwise. We also include firm and time fixed effects and an error term ξ _ijt that is assumed to be correlated with the error term in equation (1)

In equation (2), we include the market price for a single unit of advertising, market price, which is calculated as the average price paid for a single unit of advertising by all other firms in the market, generally a 30-second spot. We exclude this variable in equation (1) and instead include the firm's own GRP price, which is the actual price a particular firm pays per GRP of advertising. By excluding the market unit price in equation (1), we assume that the market price for a unit of advertising affects the decision to advertise, but not the specific amount of advertising purchased by a firm. That is, if a firm observes high unit prices paid by other firms for advertising, it may choose to stay out of the advertising market. Similarly, low unit prices may encourage firms to advertise. If a firm decides to advertise, then it pays its own negotiated GRP price.

A relevant question is whether this exclusion restriction is appropriate. In particular, the market unit price may be similar to the GRP price for each firm. Looking at the summary statistics for each variable (Table 2), we can see that market price has an average price of $2,600 per unit and firm price has an average price of $1,930 per GRP. However, firm price exhibits far more variation with nearly four times the standard deviation and a much larger range. The correlation between the two variables is 0.085, suggesting that they are not highly related. Finally, we estimate equation (1) using ordinary least squares and include market price as an explanatory variable. The variable is not statistically significant for children, teen, or adult GRPs (results available on request). This suggests that the market price does not predict how many GRPs a firm purchases. We discuss how well market price predicts the decision to advertise in the Section 5.

There are several ways to estimate equations (1) and (2). If the error terms in each equation are uncorrelated, then we can ignore equation (2) and just estimate equation (1). Nelson (Reference Nelson2008) and Nelson and Young (Reference Nelson and Young2008) estimate an equivalent specification to equation (1) using both Poisson and negative binomial models, thereby ignoring the initial decision to advertise or not. If the error terms are correlated, however, failing to account for the first-stage decision to advertise will bias estimates of how much to advertise in the second stage. Consequently, we explicitly account for both processes.

6. Robustness

As previously discussed, our model estimates could be biased because of omitted variables, reverse causality, or simultaneity. Although fixed and time effects will help mitigate for omitted variables, we further estimate equation (1) using lagged box marketing and cobranding variables to attempt to control for simultaneity or reverse causality. We use lags of 1 week and 1 month but report only the former as the results for the latter are not meaningfully different (Table 6). As can be seen by comparing with the previous results, the primary findings do not change. In particular, the effect of sugar on children is still significant as are the same cobranding variables. With teenagers, many of the parameter estimates are significant at a low P value, with the exception of the price effect. The estimates for adult GRPs do not change significantly.

Table 6.

Zero-Inflated Negative Binomial Estimates of Advertising GRPs with Lagged Values

Notes: ***P < 0.01, **P < 0.05, *P < 0.1. Standard errors clustered by brand. GRP, gross rating point.

We also estimate equation (1) including quarter-specific dummy variables interacted with firm dummy variables to control for firm-specific intertemporal differences. The estimates (not reported) for children and teen GRPs do not change significantly; however, the model for adult GRPs does not converge.