Advertising Restrictions and Competition in the Children's

To be considered for the Young Economist Award:

Advertising Restrictions and Competition in the Children’s Breakfast Cereal Industry

C. Robert Clark∗

March 14, 2005

Abstract

This paper takes advantage of the ban on advertising directed at children in the Canadian province of Quebec to examine the nature of advertising in the market for a particular children’s product, and to determine whether the advertising restriction differentially impacts certain varieties. Using a unique data set from the Print Mea- surement Bureau of Canada which surveys the purchasing behavior of some fifteen thousand Canadian households annually, as well as advertising and price data from AC Nielsen, I examine the children’s breakfast cereal market in Canada. If advertising in this market is informative, it should help to overcome perceived product differentiation, and so should lead to price competition and lower prices. If it is persuasive, it should generate the perception that there are fewer substitutes for promoted brands, and so should increase perceived differentiation and prices. I show that prices are higher in Quebec than in other Canadian regions, suggesting that the role of advertising in this market is to inform consumers about existence. If advertising is informative, a ban on advertising should increase the market shares of older, better-known brands and decrease the market shares newer and/or less well-known brands. This prediction is confirmed in the data: established brands have higher market share in Quebec than in Canadian regions where advertising is permitted, and the opposite is true for non-established brands.

JEL classiﬁcation: L1, L5, M3 Keywords: Advertising restrictions, Competition

∗Institute of Applied Economics, HEC Montréal, Universitéde Montréal, 3000 Côte-Sainte-Catherine, Montréal,QC, CANADA H3T 2A7 and CIRANO; [email protected]

1 1 Introduction

Advertising has come to represent a critical element of firm strategy. In the United States, in the year 2000, the average advertising to sales ratio for the two hundred largest advertising spending industries was close to 5% (AdAge.com). However, for a variety of products in an increasing number of jurisdictions, advertising is restricted. Cigarette advertising is restricted to varying degrees, in a number of countries including all OECD countries. Alcohol advertising is also often heavily regulated. Sweden, Greece and the Canadian province of Quebec have enacted legislation prohibiting at least some forms of advertising directed at children. Restrictions on advetising directed at children are also being considered in the U.S. where the American Psychological Association recently came out in favor of a ban on advertising directed at children under the age of eight, and in the European Union where a ban on junk food advertising to children is being considered.1 Reducing total consumption would seem to be the main purpose of these advertising bans.2 However, in addition to affecting the overall level of sales, an advertising ban may have a differential impact on specific firms, advantaging some relative to others. Whether or not there is a differential impact and which firms are advantaged or disadvantaged depends on the role that advertising plays. If advertising’s role is largely to inform new consumers about a product’s existence and characteristics (see Stigler (1961), Butters (1977), and Grossman and Shapiro (1984)) then older, better known brands will benefit. Newer products cannot use advertising to generate awareness and so it now becomes more difficult to inform potential customers of their existence.3 Word-of-mouth becomes a potentially more important avenue of information and so established products are likely to be harmed relatively less than are new products or those of which few consumers are aware. 1McKay, Betsy, ”Study Tries to Link Obesity in Children with Food Marketing”, http://online.wsj.com, January 27, 2005. 2The European Union junk food advertising ban, for instance, is being considering in order to fight rising levels of obesity (http://news.bbc.co.uk/2/hi/business/4190313.stm). 3 These predictions are supported by theoretical work by Clark (2003), whose static model of informative advertising and brand awareness predicts greater market shares for recognized brands in regions where advertising is permitted and the opposite for non-established brands, and by Doraszelski and Markovich (2002), whose dynamic model of awareness advertising predicts that if there are existing asymmetries amongst firms, then banning advertising may increase the market shares of firms with higher levels of awareness.

2 Alternatively, persuasion, and not information transmission, may be the more important function of advertising. Stigler and Becker (1977), and Becker and Murphy (1993) suggest that advertising affects the utility that consumers derive from consuming a product. Advertising creates prestige or image effects by acting as a complement to the consumption of the product. This type of advertising has also been referred to as goodwill advertising (Nerlove and Arrow (1962), Boyer (1974)). Advertising expenditures contribute to a stock of goodwill that depreciates over time. Without the ability to advertise, it is more difficult for brands to generate goodwill, and so if advertising is persuasive, then products with significant stocks of goodwill should benefit from the regulation (Doraszelski and Markovich (2003)). In addition to these effects on competition, the two theories of advertising yield different predictions of the impact of advertising on price. If advertising is informative, then it should help to overcome perceived product differentiation and so should lead to price competition and lower prices. On the other hand, if advertising is persuasive, then it generates the perception that there are fewer substitutes for promoted brands, and as a result increases perceived differentiation and prices. Therefore, if advertising is informative and it is banned, prices should be higher, while if it is persuasive and it is banned, prices should be lower. In this paper I take advantage of the ban on advertising directed at children in Quebec to examine the nature of advertising in the market for a particular children’s product, and to determine whether the advertising restrictions differentially impact certain varieties. The Quebec Consumer Protection Act has outlawed commercial advertising directed at persons under the age of thirteen. This legislation went into effect in April of 1980. From that time forward, advertising directed at children was prohibited under sections 248 and 249 of the Act. The Act lists three criteria that must be considered in order to determine whether an advertisement is directed at children: i) the nature and intended purpose of the goods advertised, ii) the manner of presenting the advertisement, and iii) the time and place it is shown. If products are intended exclusively for the use of children or have a marked appeal for children, then they cannot be advertised at all on children’s programmes (programmes for which children make up at least 15% of the audience), and can only be advertised on other programmes if they are treated so as not to appeal to the

3 needs of children.4 Using a unique data set from the Print Measurement Bureau of Canada (PMB) which surveys the purchasing behavior of some fifteen thousand Canadian households annually, as well as advertising and price data from AC Nielsen, I examine the children’s breakfast cereal market in Canada. I first show that prices for children’s brands are higher in Quebec than in other Canadian regions; this finding suggests that the role of advertising in this market is to inform consumers about existence. Given this, I then test the prediction that a ban on advertising increases the market shares of older, better-known brands (established brands) and decreases the market shares of newer and/or less well-known brands (non-established brands). This prediction is confirmed in the data: established brands have higher market share in Quebec than in Canadian regions where advertising is permitted, and the opposite is true for non-established brands. These results are consistent with those from studies of the effect of advertising restrictions on competition in different industries. Eckard (1991) studies the cigarette industry through an examination of the 1971 ban on television advertising in the United States. By comparing measures of competition in the cigarette industry in the ten years prior to the ban and in the ten years after, he finds that shares of leading brands were declining before the ban on advertising, but stable or increasing after its imposition. He also finds that the ban impedes the entry of new firms into the industry. Holak and Reddy (1986) report similar results. They find that the effect of past sales on current purchases is stronger after the ban on cigarette advertising. Sass and Saurman (1995) examine restrictions that vary by region. They analyse the malt beverages industry and find that advertising restrictions lead to increases in market concentration. The present study examines the reasons why brands are differentially affected by the advertising restrictions, whereas those mentioned above all focus on the effect of advertising restrictions on large market-share brands without modeling what it is about having a large market share that allows these brands to grow in the absence of advertising. Holak and Reddy do point out that there is some evidence that being an early entrant could be important. They show for some cigarette categories that purchase inertia for earlier entrants is greater after the ban. Sass and Saurman discuss what it is about large-

4There are some minor exceptions. In particular, in-store displays are exempt from the legislation.

4 share brands that allows them to enjoy greater market share in states where advertising is restricted: big-share brands are produced by large national brewers, while small-share brands are produced by local brewers. But since these two sets of brewers have very different marketing capabilities, it is not surprising that the local brewers are at a disadvantage where advertising is restricted. The national brewers are barely affected by state restrictions since their marketing campaigns tend to be national. So the local brewers are the only ones that are really hurt by the ban. In my study all brands are marketed and sold nationally and yet may still be differentially affected by the ban depending on whether or not they are established. The results presented here are also consistent with previous work on the breakfast cereal industry. Shum (2004) looks at this industry to determine whether advertising increases or decreases brand loyalty. Consumers that are loyal to a particular brand will perceive fewer substitutes for it. He shows that the breakfast cereal market is character- ized by considerable brand loyalty and that a brand’s advertising has a bigger impact on consumers that are not loyal to it– in other words advertising helps to overcome brand loyalty. This is in line with the predictions made in this paper. Without the ability to advertise, newer brands cannot generate awareness and steal customers from older and better-known brands. The remainder of the paper proceeds as follows. In the next section I characterize established and non-established brands and motivate the empirical framework. Section 3 outlines this framework–the data are presented and the empirical specification is outlined. In Section 4, I explain the results and consider some additional specifications. Finally, Section 5 concludes. All tables and figures are in the Appendix.

2 Motivation

The purpose of the paper is to test for a diﬀerential impact of the ban on the market shares of various children’s cereal brands. In this section, I describe in further detail the characteristics of brands that should and should not beneﬁt from the advertising restrictions. As suggested in the Introduction, an examination of the available price data for children’s breakfast cereals implies that advertising in this industry is primarily informative. Given this, economic theory predicts that brands of which consumers are

5 aware even in the absence of advertising should benefit if advertising is restricted. Without the ability to advertise, it is more difficult for newer and less recognizable brands to increase market share. In order to test for the differential impact of the ban it is necessary to somehow measure a brand’s level of recognition. However, this is not an easy task since it is difficult to measure consumer awareness of different brands. One possibility might be consumer surveys which ask respondents to assess the salience of a variety of brands. Such surveys exist in the United States: Harris Interactive interviews respondents and asks them to rate products according to their informed awareness. But I know of no such survey in Canada and so use an alternative approach to measure salience in this paper. The basic idea behind this approach is that I determine the factors that generate a high level of awareness for a product. What factors contribute to the awareness level of a brand? The amount of time a brand has been on the market is perhaps the most important determinant of the salience of a brand. However, just having been around for a long time is not sufficient for ensuring a high level of awareness. Firms must also communicate the existence of their brands to consumers. In the context of children’s breakfast cereals, the most important method of communication is through brand promotion. Children learn of the existence of brands through advertisements and in-store displays; they also learn about brands from their parents and so past promotional activity is important as well. Then, brands that have been around for a long time, and that have engaged in heavy promotional activity over the years should benefit if advertising is restricted since they should already be recognized by consumers. Meanwhile, it becomes more difficult for new brands or those that have not engaged in heavy promotional activity over the years to inform consumers of their existence. I refer to brands that have been around for a long time, and that have engaged in heavy promotional activity over the years as established. I label new brands or those that have not engaged in much promotion over the years non-established. I Section 3, below, I describe the method I use to classify the different children’s cereal brands into two groups–those that are established and those that are not. I examine the impact of the ban on advertising in Quebec on the market shares of established brands and non-established brands. I do not have ’before and after’ the ban data (this would require data on market shares, prices and advertising expenditures going back to the early 1970’s which are not available) and so I study market shares in

6 Quebec relative to those in the rest of Canada using data collected by the PMB of Canada beginning in the 1990’s. Controlling for prices and recent advertising expenditures, I test to see whether market shares of established brands are higher in Quebec than in other Canadian regions as a result of the ban and whether the opposite is true for non-established brands.

3 Empirical Framework

To test the prediction that market shares of established brands are higher where advertising is prohibited than where it is not, and that the opposite is true for non-established brands, I examine the relationship between shares of established brands and non-established brands in Quebec and the rest of Canada. To do so I construct variables which indicate whether a brand is established or not (E and NE) and whether the observation is from Quebec (Q). I then form an interaction term between the established indicator and the Quebec indicator, (E ∗ Q), and an interaction term between the non-established indicator and the Quebec indicator, (NE ∗ Q). A positive coeﬃcient on the E ∗ Q interaction term would indicate that the ban helps established brands, while a negative coeﬃcient on the NE ∗ Q interaction term would indicate that the ban hurts non-established brands.

3.1 Established Classiﬁcation:

I classify brands as established if they have been around for a long time and if they regularly engaged in heavy promotional activity over the years. A brand is said to have been around for a long time if it was introduced prior to 1980.5 As mentioned above, for children’s breakfast cereals, promotion is perhaps the most important means through which ﬁrms can communicate the existence of their brands. Children learn of the existence of brands through advertisements and in-store displays. They also learn from their parents through word-of-mouth communication which makes past promotional activity important as well. I do not have data on all of the types of promotion that ﬁrms have engaged in and so I capture a brand’s typical promotional activity by its past expenditure on national advertising. I use national advertising expenditure data available from AC Nielsen for children’s brands for the pre-sample years for which I have data

5Dates of introduction for child and adult/family brands were collected from Companies and their Brands and through direct contact with Kellogg’s, General Mills, Post and Quaker.

7 (1989-1996) and determine that for children’s brands advertising expenditure levels over this period are clustered in two distinct groups: (i) little or no advertising, and (ii) heavy promotional activity. I look for an obvious break between these two clusters, and it seems that heavy promotional activity starts at around $550 000 (all expenditure levels are in year 2000 dollars). Therefore, I classify a brand as having engaged in heavy promotion over the years if it spent at least $550 000 on advertising in over half of the eight pre- sample years.6 Based on this, I denote a brand established if, in addition to having been introduced prior to 1980, more than $550 000 have been spent promoting it nationally in at least five of the eight pre-sample years. Table 2 lists established and non-established brands. Although this classification method may seem somewhat ad hoc, I think that it captures the essence of what it means to be established. Nonetheless, in an effort to confirm that I have properly identified heavy promoters I also generate a classification based on the average total expenditure level for children’s brands that existed over the entire eight year pre-sample period. Brands that spent more than the average are considered established while those that spent less are not. This method yields exactly the same classification as the previous method. In Subsection 4.3, I test the robustness of the definition of established. I present results from another possible classification. Definition 2 examines how important heavy promotional activity is to being established by classifying a brand as established simply if it has been around for a long time. So brands are established if they have been around since at least 1980. The difference between the two definitions is the manner in which they classify older brands that have not been heavily promoted over the years. Definition 1 groups these brands with newer brands while Definition 2 groups them with older brands that have been heavily promoted. In fact, there is reason to think that older brands that have not been heavily promoted over the years are barely affected by a ban on advertising, since they never actually made much use of advertising as part of their overall marketing

6Between $400 000 and $550 000 there are very few observations while above and below these amounts there are relatively many. I have examined the sensitivity of my results to the selection of this cutoff. I consider cutoffs of $500 000 and $600 000. Relative to the $550 000 cutoff, one brand is re-classified as established rather than non-established if the cutoff is moved to $500 000, while two brands are re- classified as non-established rather than established if the cutoff is moved to $600 000. There is still a positive (negative) effect to being established (non-established) in Quebec relative to being established (non-established) in the rest of Canada.

8 strategies. In an attempt to isolate the different effects that are present, I also consider a third classification that sorts brands into three groups. Brands are either (1) new (introduced after 1980), (2) old and lightly promoted (introduced before 1980 but did not spend at least $550 000 in at least five of the eight pre-sample years), or (3) old and heavily promoted (introduced before 1980 and spent at least $550 000 in at least five of the eight pre-sample years).

3.2 The Quebec market:

For the method described above to be valid, consumption patterns in Quebec and the rest of Canada must be similar in the absence of advertising restrictions. In other words, the Quebec indicator, Q, must capture only the effect of the ban and not other factors which are specific to Quebec. In its Window on Quebec 2001, AC Nielsen reports that consumers in Quebec are similar to their counterparts in the rest of Canada in many regards. Most importantly, people do not seem to be more brand loyal in Quebec than in the rest of Canada on average. Using its Homescan Consumer Panel, AC Nielsen measures ongoing product acceptance and retention. It seems that Quebec consumers are no less likely to spread their purchases in a product category across brands. Were this not the case, then it could be argued that the superior performance of the established brands in Quebec is the result of greater brand loyalty on the part of Quebec consumers. In addition, the Window on Quebec 2001 reports that people in Quebec watch approximately the same amount of television (22.5 hours per week in Quebec vs. 21.5 hours per week in Canada) and spend approximately the same amount on food ($114.9 per week in Quebec vs. $112.09 per week in Canada) and on baked goods/cereals in particular ($13.42 per week in Quebec vs. $12.47 per week in Canada). Moreover Quebec consumers do not have a bias against pre- sweetened cereals. The PMB data show that in 1999, 25 % of cereal usage in Quebec was of pre-sweetened cereals, versus 26 % nationally. The PMB also reports that the fraction of households in Quebec with children under the age of twelve is not much different than in Canada as a whole (in 2000, 25.5% in Quebec vs. 27% nationally). I also examine whether there are any other factors specific to Quebec that could lead to higher market shares for established products and lower market shares for non-established products. The biggest difference between Quebec and the rest of Canada is in terms of language. Only 43 % of the population of Quebec can speak English (defined by

9 Statistics Canada as being able to carry on a conversation, 1996 Census). Given this, one possible interpretation of any result suggesting that established brands have bigger market share in Quebec is that people that speak only French have a preference for established brands. For instance some national advertising campaigns might be primarily designed for English programmes and so these advertisements would have little effect on people that speak only French. They would therefore be more likely to purchase brands that had been around for a long time because those are the ones they have heard of. This would be particularly problematic if established meant only old. However, I have defined established brands to be those that have been around for a long time and that advertise nationally regularly. Therefore, finding that established brands have bigger market share in Quebec implies that regular national advertising does have an effect on French speaking people. Moreover, although Statistics Canada reports that only 43 % of the population of Quebec can carry on a conversation in English, a much larger fraction understands enough English to make sense of breakfast cereal commercials. Children in Quebec begin learning English in Grade 3 (Grade 4 prior to 1999). Furthermore, if there is some Quebec-specific preference for established brands, it should be that market shares for established adult/family cereal brands are also higher in Quebec than in the rest of the country and market shares for non-established brands are lower. This does not seem to be the case. Figures 1-4 show average market shares in Quebec and in the rest of Canada for established and non-established children and adult/family cereal brands. Using the same definition of established (introduced prior to 1980 and spent at least $550 thousand on national advertising in over half of the eight pre-sample years), I find that there are eleven established adult/family brands. The Figures reveal that 5 of the 11 (45%) established brands have higher average market share in Quebec than they do in the rest of the country and that 16 of the 34 (47%) non-established brands have lower market share in Quebec. Contrast these percentages with those for children’s brands: 71% of established children’s brands have bigger market shares in Quebec than in the rest of the country and 60% of the non-established brands have smaller market shares in Quebec than in the rest of Canada.7 It is also the case that, despite the ban on advertising in Quebec, there is some advertising expenditure on children’s cereal brands in this province.8 Therefore one might also

7Price data for adult/family brands were not available and so regression analysis was not possible. 8Advertisements must be presented by agencies and/or advertisers to a review panel prior to being

10 be concerned that established brands have higher market shares in Quebec simply because they are more heavily promoted. One possibility is that more stringent French-language laws in Quebec have increased promotional costs in such a way as to encourage firms to promote their established brands more heavily. As discussed in further detail below, I use region-specific advertising expenditure data. As a result, I can control for the possibility that firms were discouraged from promoting their newer brands in Quebec.

3.3 Data:

In this Subsection, I describe the data used to perform the analysis. In order to determine the effect of the advertising restrictions on price and to test for the differential effect of the ban the following variables are necessary: market shares, prices, brand characteristics, and advertising expenditures. Market share data: The market shares are calculated using Category Reports from the Print Measurement Bureau. Each year the PMB surveys approximately 15 000 Canadian households. The PMB study design includes weighting to match Statistics Canada household and individual population data. There are two weighting processes: (i) design weights that compensate for over or under sampling, and (ii) projection to match the population estimates of Statistics Canada. Questionnaires are left behind by interviewers and contain questions about the household’s product usage. Respondents are asked to report on the brands of cold cereal used in their household in the previous year. PMB reports in its survey on the number of households that used a brand at least once. So a brand’s market share represents its fraction of cereal use as opposed to its fraction of cereal purchases. I am considering a brand’s share of the total breakfast food market, which consists of all cereal users (children’s brands and adult/family brands) and all individuals who claimed not to use any cereal.9 I use four years of PMB reports, 1998-2001. Each year’s PMB report provides results cleared to air. Additionally, some advertisements for children’s products may appear on programs that are directed at an adult audience. Nielsen Media Research examined advertisements for General Mills’ Nesquick and Lucky Charms for me and found that they appeared on such adult-directed shows as Chicago Hope. 9PMB asks respondants whether they did or did not purchase any cereal brands in the last six months. I assume that those who did not buy in the previous six months are those that did not buy in the entire previous year.

11 from a survey taken in the previous year (so my data cover 1997-2000). Usage is reported for six Canadian regions: Atlantic Canada, Quebec, Ontario, Manitoba/Saskatchewan, Alberta and British Columbia. For 1997, 55 cereals are reported on, out of which 15 are children’s brands. For 1998, 17 out of 57 are children’s brands, for 1999, 17 out of 67, and for 2000, 17 out of 74. My classification of breakfast cereals into children’s cereals and non-children’s cereals is motivated by the classifications of Nevo (2000) and Shum (2000).10 Table 2 lists children’s brands reported on each year along with their average market shares in Quebec and the rest of Canada. Figures 1-4 compare market shares for established and non-established children’s brands and established and non-established adult/family brands. Price data: The price data are collected by AC Nielsen.11 I have prices for each brand in each of the four years in each of the six regions. They are calculated using samples or aggregate census data for grocery supermarket banners by taking total revenue from boxes of all sizes divided by total quantity. This measure is not ideal as it does not represent a properly weighted average. I can construct a somewhat better measure of price for 1999 and 2000 since I have information on the average price per box of different sizes. For each brand I choose the box size closest to 400 grams and I calculate the average price per 100 grams. In Table 3, I list the average price of each brand in Quebec and in the rest of Canada using the more accurate 1999-2000 prices. Brand characteristic data: Brand characteristics come from the Nutrient Database for Standard Reference com- piled by the Agricultural Research Service of the United States Department of Agricul- ture. Nevo (2000) includes total calories, dietary fiber, and sodium as characteristics in his breakfast cereals study. I include fiber and sodium along with sugar rather than

10Certain Quaker Bagged Cereals might also be considered children’s brands. These brands are not included in the sample since PMB reports usage only in its 2001 survey and since price data were not available from AC Nielsen for these brands. Kellogg’s Cruncheroos was also not included because AC Nielsen did not have price data for this brand. However, the inclusion of Cruncheroos would almost certainly strengthen my results–it is a non-established brand whose average market share in Quebec is lower than its average market share in the rest of Canada. The same is true of three of the ﬁve Quaker bagged brands. 11These data were used for academic purposes only and the results obtained do not necessarily reﬂect the views of AC Nielsen.

12 calories. Sugar and fiber are measured in grams per 100 gram quantity, while sodium is measured in milligrams per 100 gram quantity. Brand characteristics do not vary over time or across regions. Advertising data: Advertising data come from the AC Nielsen Media Services annual estimates of Ad- vertising Expenditures in Canada. Annual advertising expenditures are reported for each of the six Canadian regions. Only advertising spending that originates in Canada is included; anything spilling over from the United States is omitted. The data are collected through the following sources: television, radio, daily newspapers, magazines, and out-of-home. Almost all of the advertising for ready-to-eat breakfast cereals is done through television. The following networks are included in AC Nielsen’s estimates: Télé-Métropole, Radio Canada, TélévisionQuatre Saisons, Canadian Broadcasting Corporation, Canadian Television, Baton Broadcasting Service-Ontario, Global, Maritimes Independent Television, The Sports Network, Musique Plus, MuchMusic, and Newsworld.

See Table 1 in the Appendix for summary statistics.

3.4 Empirical Speciﬁcation:

In addition to depending on whether or not it is established, a brand’s market share is influenced by a number of other factors. Market share depends on price, current and past advertising, and various brand characteristics. So I regress observed market share of each of the brands on price, advertising, lagged advertising, brand characteristics (sugar, fiber and sodium), the established indicator, the E ∗ Q indicator, and the NE ∗ Q indicator. More specifically, I estimate the following equation:

Sjrt = β0 +β1pjrt +β2ajrt +β3ajr(t−1) +β4xj +β5Ejt +β6 ∗Q∗Ej +β7 ∗Q∗NEj +εjrt (1)

where Sjrt is brand j’s share in region r in year t, pjrt and ajrt are brand j’s price and advertising expenditure in region r in year t respectively, ajr(t−1) is brand j’s advertising expenditure in region r in year (t − 1), and xj is a vector of observed characteristics for brand j. Year dummies are also included to control for any systematic shocks to demand.

The coeﬃcient β5 captures the eﬀect of being established. The primary parameters of

13 interest are β6 and β7, which are intended to capture the eﬀects of being established and non-established in Quebec relative to (being established and non-established in) the rest of Canada respectively. The omitted category is non-established outside of Quebec.

At this point, it is important to deal with the possibility that εjrt is correlated with certain explanatory variables. Failure to do so could result in biased and inconsistent estimates. Endogeneity bias may exist if unobserved (by the econometrician) determi- nants of a brand’s market share also affect its price or advertising expenditure. For instance, unobservable promotional activity, such as in-store displays and sponsorships of sporting/schooling events, could affect market shares directly but also influence prices and expenditures on traditional means of advertising. Systematic shocks to demand such as those generated by health warnings (for example announcements about the increasing obesity of children) can also affect market share as well as advertising and pricing strategies. In addition, brand equity might further compromise the exogeneity of the price and advertising expenditure variables. In an attempt to deal with this issue one might consider adding firm dummies to the model. There could be unobservables at the firm level that affect market share directly as well as advertising budgets. In other words, firms could have general marketing strategies that they apply to all of their brands. For instance, Kellogg’s may decide to pay for premium shelf space at the super market for its brands. This might increase market share directly and could be part of an overall promotional strategy that sees the firm spending more on television advertising as well. However, controlling for firm specific unobservables is not enough to resolve all endogeneity problems. While firm dummies should control for any unobserved heterogeneity stemming from the manufacturer that affects a brand’s market share and that influences its price and the amount spent promoting it, there is also brand-specific unobserved heterogeneity. To deal with this I adopt a brand fixed effects model. Including fixed effects means that the indicator for being established, Ej, the observed characteristics that make up xj, and firm dummies are dropped from the model since they do not vary from one market to another. However, I can still estimate my coefficients of interest since they do vary across markets. As noted in Nevo (2000), once brand effects are included, the error term is the unobserved region-year deviation from the overall mean valuation of the brand. Firms are assumed to be able to observe this deviation and account for it when choosing their marketing strategies. It seems likely that when setting their prices in each market (region-

14 year) firms take the deviation from the mean valuation into account. To deal with this, I adopt an instrumental variables approach proposed by Hausman. This method relies on the assumption that region-specific valuations of brands are independent across regions controlling for brand-specific intercepts. So prices of a brand in other regions are valid instruments. This is the case since prices of a particular brand in two regions are correlated owing to the common marginal cost of producing the brand but uncorrelated with the market-specific valuation of the brand. Therefore, as an instrument I use regional yearly average prices (not including the region being instrumented) in all four years.12 The assumption that valuations are independent across regions will not hold in a number of circumstances. For instance, if there is some national shock, then all regions will be affected. The inclusion of time dummies should help to correct this problem. Also if local advertising or in-store promotions are correlated across regions, then the inde- pendence assumption will be violated. Nevo points out that the larger the regions, the less likely it is that there will be correlation across borders. In my study regions are entire Canadian provinces (or multiple provinces in the cases of Manitoba/Saskatchewan and Atlantic Canada); it is therefore doubtful that in-store promotional strategies are coordinated across such vast distances. Although it is likely that advertising strategies are determined at the firm level or the brand level, there might be factors that vary over time that could affect both market share as well as the promotional strategy for a brand. To deal with this I add brand-year fixed effects to (1).

12A common set of instruments employed to deal with the endogeneity of price in discrete choice models (Nevo (2000), Rekkas(2001), Berry, Levinsohn and Pakes(1995)) includes the observed brand characteristics, the sums of the values of the same characteristics of other brands sold by that firm, and the sums of the values of the same characteristics of the brands offered by other firms. These variables are meant to proxy for the closeness of the competition in the market. I do not adopt this approach here since in my case there is almost no variation between markets in these instruments. This is because there is little variation over time and no variation across regions in the brands available for sale in my data and so this amounts to little more than adding brand dummies.

15 4 Empirical Results

4.1 Eﬀect of Advertising on Price

As mentioned in the Introduction, the two theories of advertising yield different predictions of the effect of advertising on price. If advertising is primarily informative, then it should help to overcome perceived product differentiation and so should lead to price competition and lower prices. On the other hand, if advertising is mostly persuasive, then it generates the perception that there are fewer substitutes for promoted brands, and as a result increases perceived differentiation and reduces price competition. In this subsection, I take advantage of these different predictions to determine whether advertising in the children’s breakfast cereal market is informative or persuasive. In Table 3, I present average prices for the different children’s brands in Quebec and in the rest of Canada. Average prices per one hundred grams in Quebec are higher for 13 of the 17 brands. This difference is confirmed using regression analysis. Following Milyo and Waldfogel (1999), who investigate the impact of relaxing liquor price advertising restrictions on prices, I regress the log of price on a Quebec indicator and time dummies. Results are reported in Table 4. The average price in Canada is .86 and the percentage increase in price if sold in Quebec is .025.

4.2 Diﬀerential Eﬀect of the Ban

In this subsection, I test for a differential impact on established and non-established brands. In Table 2, I list average market shares in Quebec and in the rest of Canada respectively. 5 of 7 established brands do better in Quebec than in the rest of Canada and 6 of 10 non-established brands do worse in Quebec than in the rest of Canada. Figures 1 and 2 graph average market shares for established and non-established brands respectively. The Quebec-Rest of Canada contrast is perhaps more obvious in these pictures. Further confirmation is provided by estimating equation (1). Results from the fixed effects estimation of (1) are reported in Table 5 under the heading B-FE. Mean market share in the data is .015. Being established in Quebec increases market share relative to being established in the rest of Canada by .0032, while being non-established in Quebec reduces market share relative to being non-established in the rest of Canada by .0014. Results from the fixed effects model with the instrument for price are reported in Table 3

16 under the heading B-FE-IV. The coefficients on E ∗ Q and NE ∗ Q are .0036 and -.0016 respectively. These coefficients are not statistically different from those arrived at without the use of an instrument. The same is true of the addition of brand-year fixed effects to equation (1) rather than brand effects. The results are reported in Table 5 under the headings B-Y-FE. Being established in Quebec increases market share relative to being established in the rest of Canada by .0033, while being non-established in Quebec reduces market share relative to being non-established in the rest of Canada by .0014. These results are not statistically different from those obtained using brand effects. These results indicate that the prediction that established and non-established brands are differentially affected by the ban are supported by the data. Established brands have higher market shares in Quebec where advertising is restricted and the opposite is true for non-established brands. Without the ability to advertise, new, less established brands are at a competitive disadvantage relative to established brands.

4.3 Additional speciﬁcations and sensitivity analysis

In this subsection I examine additional speciﬁcations to test the robustness of these results.

4.3.1 Deﬁnition of established:

I first test the robustness of my results to the way in which brands are classified as established or non-established. Definition 2 classifies a brand as established simply if it has been around for a long time–at least since 1980. The classification generated by this definition is quite different than the one generated by Definition 1. Results are reported in Table 6. In the fixed effects model, the coefficient on E ∗ Q is .0013 while the coefficient on NE ∗ Q is -.0016. So the effect of being established in Quebec according to this definition, is not as important as according to the first definition. This makes sense since, as discussed above, Definition 2 classifies old and lightly promoted brands as established. There is no reason that these brands should do better in Quebec than the rest of the country. To examine this issue more carefully I consider a third classification that sorts brands into three groups. Brands are (1) new (introduced after 1980), (2) old and lightly pro-

17 moted, or (3) old and heavily promoted. Results from this specification are reported in Table 7. Being old and heavily promoted and in Quebec increases market share by .0033. Meanwhile, being old and lightly promoted and in Quebec has no significant effect on market share and being new decreases market share by .0015.

4.3.2 Logit model

It is important to note that market shares calculated according to the specification in (1) are a function of own price and own advertising, but not prices and advertising levels of other brands. An alternative specification which allows brand j’s share in a particular market to depend on characteristics and choices of other brands follows Berry (1994) and employs as the dependent variable the difference between the log of each brand’s observed share and the log of the share of the outside alternative: ln(Sjrt) − ln(Sort). This specification is derived structurally from a discrete choice logit model of demand in which consumers compare the utility they would get from consuming each brand and purchase the brand that yields the highest utility. Utility is assumed to be a function of price and advertising and so a brand’s relative share is a function of advertising levels and prices of other brands in addition to its own. However, if advertising generates awareness, then it does not enter into the utility function but helps to determine a consumer’s choice set. The same is true of a brand’s classification as established or not. In such a situation, the logit specification can be considered only an approximation of actual behavior. Table

8 contains regression results from an estimation of (1) with ln(Sjrt) − ln(Sort) as the dependent variable. Fixed effect estimation shows that the coefficient on E ∗ Q is .17, while the coefficient on NE ∗ Q is -.4.

4.3.3 Price data

Given the manner in which average prices were calculated for the full four-year sample, I test the robustness of my results to the inclusion of more accurate price data. I run specification (1) using just the final two years of data since for these two years I have price data that represent a properly constructed weighted average. Using these data in a fixed effects model I find that increasing price by one dollar per hundred grams causes market share to fall by .0088. Being established and in Quebec increases market share by .0029. The coefficient on price is no longer significant once price is instrumented for.

18 The coefficient on E ∗ Q is .003. Both with and without the instrument for price, the coefficient on NE ∗ Q is not statistically significant, and it is not statistically different from that found with the four year data. These results are reported in Table 9. More accurate data mean less measurement error. However, this comes at a cost, since these data are available only for 1999 and 2000. This smaller sample means less variation and so is bad from an efficiency stand point.

4.3.4 Eﬀect of advertising in Quebec

In Table 10, I report results from the estimation of (1) with the inclusion of an interaction of advertising expenditure and the Quebec indicator. In light of the ban, a dollar spent on advertising in Quebec might have a different effect on market shares than a dollar spent on advertising in the rest of Canada. In particular, advertisements for children’s brands in Quebec are now often directed at adults. A similar advertisement might have a different impact on adults than on children. Results suggest that this variable has no significant effect on market shares but that, as a result of its inclusion, being established in Quebec is not quite as important economically (or statistically) relative to being established in the rest of the country. The coefficient on E ∗ Q falls to .002 from .0032 in a fixed effects estimation of (1). This result implies something about the role of advertising and the impact of being established in Quebec. Part of the power of being established in Quebec is clearly felt through the more important role parents play in brand choice in this market since children cannot be directly targeted with advertisements. In Quebec, parents are the target of ads for children’s brands and are more likely to be affected by ads for brands they rare familiar with from their youth. Therefore, once the different effect of advertising in Quebec is controlled for, one of the channels through which established brands do better in Quebec is no longer captured by E ∗ Q.

4.3.5 Regional eﬀects

The specification in (1) implies that the five regions that make up the rest of Canada are the same. However, there could be regional effects on market shares. To control for this I consider adding region indicators to (1), but this changes the interpretation of the coefficients of interest. The coefficient on E ∗ Q would represent the effect of

19 being established in Quebec relative to being established in the omitted region while the coefficient on NE ∗ Q would be the effect of being non-established in Quebec relative to being non-established in the omitted region. Rather than do this I estimate the following specification:

X Sjrt = β0+β1pjrt+β2ajrt+β3ajr(t−1)+β4xj +β5Ej + (βEr∗r∗Ej +βNEr∗r∗NEj)+εjrt r∈ROC (2) This specification allows for the comparison of the effects of being established and non- established in region r in the ROC (rest of Canada-Atlantic Canada, Ontario, Mani- toba/Saskatchewan, Alberta and British Columbia) relative to being established and non- established in Quebec respectively. Results are reported in Table 11. Being established in Ontario or British Columbia reduces market share relative to being established in Que- bec. In the other three regions, the effect is insignificant. Being non-established in any region except British Columbia increases market share relative to being non-established in Quebec.

5 Conclusions

In this paper I have examined the possibility that an unintended consequence of the ban on advertising directed at children in Quebec is to hinder competition. I show that prices are higher in Quebec because of the ban, which implies that advertising provides information and reduces perceived product diﬀerentiation. Therefore, older and better-known brands should have bigger market share in Quebec where advertising is prohibited than in regions where it is allowed, and the opposite should be true for non-established brands. Empirical analysis of the children’s breakfast cereal market supports these predictions. Established brands–those that have been around for a long time and that have advertised heavily and regularly over the years–have bigger market share in the province of Quebec where advertising is banned than they do in the rest of Canada. Non-established brands do better in the rest of Canada than they do in Quebec.

20 6 References

Ackerberg, D. (2002): “Empirically Distinguishing Between Informative and Prestige Ef- fects of Advertising,” Rand Journal of Economics, 33, 316-333.

——– (2003): “Advertising, Learning, and Consumer Choice in Experience Good Markets: A Structural Empirical Examination,” International Economic Review, 44, 1007-1040.

AC Nielsen Company of Canada (1995-2000): Market Track-Grocery Banner.

——– (2001): Window on Qu´ebec 2001.

AdAge website: www.adage.com.

Advertising Education Forum website: www.aeforum.org.

Anand, B., and Shacher, R. (2001): “Advertising, the Matchmaker,” Harvard Business School Working Paper 02-057.

Anderson, S., de Palma, A., and Thisse, J.-F. (1992): Discrete Choice Theory of Product Diﬀerentiation, The M.I.T. Press, Cambridge, MA.

Baltagi, B. (1995): Econometric Analysis of Panel Data, John Wiley & Sons, New York, NY.

Becker, G., and Murphy, K. (1993): “A Simple Theory of Advertising as a Good or Bad,” The Quarterly Journal of Economics, 108, 941-964.

Benham, L. (1972): “The Eﬀect of Advertising on the Price of Eyeglasses,” Journal of Law and Economics, 25, 242-262.

Berry, S. (1994): “Estimating Discrete-Choice Models of Product Diﬀerentiation,” The RAND Journal of Economics, 63, 841-890.

Berry, S., and Levinsohn, J., and Pakes, A. (1995): “Automobile Prices in Market Equi- librium,” Econometrica, 63, 841-890.

Boyer, K. (1974): “Informative and Goodwill Advertising,” The Review of Economics and Statistics, 56, 541-548.

21 Brown-Humes, C. (2001): “Creative Business: Advertising and Children,” Financial Times.

Butters, G. (1977): “Equilibrium Distributions of Sales and Advertising Prices,” The Review of Economic Studies, 44, 465-491.

Caron, A. and Martin, C. (1995) Les impacts de la réglementation québécoise de la pub- licitédestinéeaux enfants sur les ventes de temps et d’espace publicitaires, Laboratoire d’économie industrielle des médias et Groupe de recherche sur les jeunes et les médias, Universitéde Montréal,Montréal, PQ.

Clark, C.R. (2003): “A Model of Informative Advertising and Brand Awareness”, HEC Montreal, mimeo.

Doraszelski, D., and Markovich, S. (2003): “Goodwill and Awareness Advertising: Impli- cations for Industry Dynamics”, Stanford, mimeo.

Eckard, W. Jr. (1991): “Competition and the Cigarette TV Advertising Ban,” Economic Inquiry, 29, 119-133.

Gale Research Group: Companies and their Brands.

Goeree, M. (2002): “Advertising in the U.S. Personal Computer industry”, University of Virginia, mimeo.

Grossman, G., and Shapiro, C. (1984): “Informative Advertising with Diﬀerentiated Prod- ucts,” The Review of Economic Studies, 51, 63-81.

Harris Interactive (2001): Equitrend.

Holak, S., and Reddy, S. (1986): “Eﬀects of a Television and Radio Advertising Ban: A Study of the Cigarette Industry,” Journal of Marketing, 50, 219-227.

Jordan, A. (1996): The State of Children’s Television: An Examination of Quantity, Quality, and Industry Beliefs [Report No. 2], Annenberg Public Policy Center, University of Pennsylvania, Philadelphia, PA.

22 Kessides, I. (1986): “Advertising, Sunk Costs, and Barriers to Entry,” The Review of Economic and Statistics, 68, 84-95.

Milyo, J., and Waldfogel, J. (1999): “The Eﬀect of Price Advertising on Prices: Evidence in the Wake of 44 Liquormart,” The American Economic Review, 89, 1081-1096.

Nerlove, M., and Arrow, K.J. (1962): “Optimal Advertising Policy under Dynamic Con- ditions,” Economica, 29, 129-142.

Nevo, A. (2001): “Measuring Market Power in the Ready-to-Eat Cereal Industry,” Econo- metrica, 69, 307-342.

——– (2000): “A Practitioner’s Guide to Estimation of Random-Coeﬃcients Logit Models of Demand,” Journal of Economics and Management Strategy, 9, 513-548.

——– (2000): “Mergers with Diﬀerentiated Products: The Case of the Ready-to-Eat Cereal Industry,” RAND Journal of Economics, 31, 395-421.

Nielsen Media Services (1989-2001): Annual Summary of Advertising Expenditures in Canada, Nielsen Marketing Research, Markham, ON.

Petrin, A. (2002): “Quantifying the Beneﬁts of New Products: The Case of the Minivan”, The Journal of Political Economy, 110, 705-729.

Print Measurement Bureau (1997-2001): PMB Category Reports.

Rekkas, M. (2001): “Estimation of a Structural Model of Multiparty Elections: Canadian Evidence”, University of Toronto, mimeo.

Sass, T., and Saurman, D. (1995): “Advertising Restrictions and Concentration: The Case of Malt Beverages,” The Review of Economics and Statistics, 77, 66-81.

Shum, M. (2004): “Does Advertising Overcome Brand Loyalty? Evidence from the Break- fast Cereals Market”, Journal of Economics and Management Strategy, 13, 241-272.

Stigler, G. (1961): “The Economics of Information,” The Journal of Political Economy, 69, 213-225.

23 Stigler, G., and Becker, G. (1977): “De Gustibus Non Est Disputandum,” The American Economic Review, 67, 76-90.

United States Department of Agriculture USDA Nutrient Database for Standard Refer- ence, Release 14. Nutrient Data Laboratory.

Woodard, E. (1999): The 1999 State of Children’s Television Report: Programming for Children Over Broadcast and Cable Television [Report No. 28], Annenberg Public Policy Center, University of Pennsylvania, Philadelphia, PA.

Wooldridge, J. (2002): Econometric Analysis of Cross Section and Panel Data, The M.I.T. Press, Cambridge, MA.

24 7 Appendix:

Table 1: Summary statistics for children’s breakfast cereals. 1997-2000 unless otherwise indi- cated.

Min Max Mean St.Dev Share in market (%) .000067 .054 .015 .012 Share in market (%) ‘99,‘00 .000067 .0488 .0148 .012 Provincial Advertising ($ millions per year) 0 1.404 .085 .151 Provincial Advertising ‘99,‘00 0 .735 .067 .119 Price ($ per box) 2.06 5.59 3.87 .53 Price ($ per 100g in ‘99,‘00) .458 1.242 .87 .116 Sugar (grams per 100g) 35.47 53.9 41.31 4.37 Sodium (milligrams per 100g) 150 915 588.7 183.1 Fiber (grams per 100g) .7 5 2.51 1.09

25 Table 2:

avr market avr market in Brands Established Old share share PMB 98 Quebec ROC GM Cinn. Toast Crunch y n n 0.0029 0.01224 GM Count Chocula y n y 0.0008 0.00425 GM French Toast Crunch n n n 0.01288 0.0091 GM Golden Grahams y n y 0.01068 0.01154 GM Lucky Charms y y y 0.01189 0.01636 GM Nesquick n n n 0.00575 0.00517 GM Reese’s Puﬀs y n n 0.00322 0.00514 GM Trix y n n 0.00353 0.00552 Kellogg’s Cinn. Mini Buns y n n 0.00569 0.00436 Kellogg’s Corn Pops y y y 0.02891 0.02514 Kellogg’s Froot Loops y y y 0.03621 0.03661 Kellogg’s Frosted Flakes y y y 0.04272 0.03782 Post Alphabits y y y 0.02226 0.02189 Post Honeycombs y y y 0.03577 0.02194 Post Pebbles y n y 0.00233 0.00333 Post Sugar Crisp y y y 0.01749 0.01016 Quaker Cap’n Crunch y n y 0.01636 0.01585

26 Table 3: Average prices in 1999-2000 in Quebec and the rest of Canada Avr price per Avr price per 100g Brands 100g in Quebec Rest of Canada GM Cinnamon Toast Crunch $0.93 $0.88 GM Count Chocula $1.08 $1.03 GM French Toast Crunch $0.83 $0.82 GM Golden Grahams $0.93 $0.89 GM Lucky Charms (E) $0.91 $0.88 GM Nesquick $0.89 $0.88 GM Reese’s Puﬀs $0.94 $0.95 GM Trix $1.08 $0.93 Kellogg’s Cinnamon Mini Buns $0.96 $0.99 Kellogg’s Corn Pops (E) $0.92 $0.88 Kellogg’s Froot Loops (E) $0.89 $0.88 Kellogg’s Frosted Flakes (E) $0.81 $0.75 Post Alphabits (E) $0.68 $0.65 Post Honeycombs (E) $0.751 $0.746 Post Pebbles $0.83 $0.95 Post Sugar Crisp (E) $0.77 $0.73 Quaker Cap’n Crunch $0.87 $0.92

27 Table 4:

Results from estimation of ln(pjrt) = β0 + β1Q using only 1999 and 2000 data (more accurate price data). Number of observations: 204

Coef. .025∗ Quebec (.014) Year fixed effects included (*coefficient significant at 10% level; **coefficient significant at 5% level) (robust standard errors are reported in parentheses, observations are considered independent across brands, but not within brands)

28 Table 5: Results from estimation of (1) : Number of observations: 396

Dependent variable: Sjrt

B-FE B-Y-FE B-FE-IV

Coef. Coef. Coef. −.0021∗∗ −.0025∗∗ −.0089∗∗ Price (.0005) (.0005) (.0034) −.001 −.0038 −.001 Advertising (.0024) (.003) (.003) .001 .0025 −.0019 Lagged Advertising (.0021) (.0025) (.003) Established .0032∗∗ .0033∗∗ .0036∗∗ E ∗ Q (.0008) (.0007) (.001) −.0014∗∗ −.0014∗∗ −.0016∗∗ NE ∗ Q (.0006) (.0006) (.0008) Year fixed effects included included Brand fixed effects included included Brand-year fixed effects included (*coefficient significant at 10% level; **coefficient significant at 5% level)

29 Table 6: Results from estimation of (1) : Number of observations: 396 Different definitions of established: Definition 2

Dependent variable: Sjrt

B-FE B-Y-FE B-FE-IV

Coef. Coef. Coef. −.0019∗∗ −.0023∗∗ −.008∗∗ Price (.0005) (.0005) (.003) −.0005 −.003 −.0004 Advertising (.002) (.003) (.003) −.0015 .003 −.0009 Lagged Advert (.002) (.0025) (.003) Old .0013∗∗ .0013∗∗ .0012∗ Old∗Q (.0006) (.0006) (.0007) −.0016∗ −.0015∗ −.0014 New∗Q (.0009) (.0009) (.0011) Year fixed effects included included Brand fixed effects included included Brand year fixed effects included (*coefficient significant at 10% level; **coefficient significant at 5% level)

30 Table 7: Results from estimation of (1) : Number of observations: 396 Decomposition of Established

Dependent variable: Sjrt

B-FE B-Y-FE B-FE-IV

Coef. Coef. Coef. −.002∗∗ −.0025∗∗ −.009∗∗ Price (.0005) (.0005) (.003) −.001 −.004 −.001 Advertising (.002) (.003) (.003) −.001 −.002 −.002 Lagged Advert (.002) (.002) (.003) Old+lightly promoted Old+heavily promoted −.0013 −.0013 −.002∗ (Old+lightly promoted)∗Q (.0009) (.0008) (.0011) .0032∗∗ .0033∗∗ .0036∗∗ (Old+heavily promoted)*Q (.0008) (.0007) (.001) −.0015∗ −.0015∗ −.0013 New∗Q (.0009) (.0009) (.001) Year fixed effects included included Brand fixed effects included included Brand-year fixed effects included (*coefficient significant at 10% level; **coefficient significant at 5% level)

31 Table 8: Results from estimation of (1) : Number of observations: 396

Dependent variable: ln(Sjrt) − ln(Sort)

B-FE B-Y-FE B-FE-IV

Coef. Coef. Coef. −.24∗∗ −.28∗∗ −.97∗∗ Price (.06) (.06) (.4) −.48∗ −.56 −.48 Advertising (.29) (.35) (.34) .42∗ .42 .126 Lagged Advertising (.25) (.3) (.34) Established .16∗ .17∗ .206∗ E*Q (.09) (.09) (.11) −.4∗∗ −.4∗∗ −.43∗∗ NE*Q (.077) (.076) (.094) Year fixed effects included included Brand fixed effects included included Brand year fixed effects included (*coefficient significant at 10% level; **coefficient significant at 5% level)

32 Table 9: Results from estimation of (1) using only 1999 and 2000 data (more accurate price data). Number of observations: 204

Dependent variable: Sjrt

B-FE B-Y-FE B-FE-IV

Coef. Coef. Coef. −.0088∗∗ −.0089∗∗ −.011 Price (.04) (.04) (.009) .00008 .002 .00027 Advertising (.0045) (.005) (.005) −.0024 −.0007 −.0026 Lagged Advertising (.004) (.004) (.0041) Established .0029∗∗ .0029∗∗ .0030∗∗ E*Q (.0011) (.0011) (.0012) −.001 −.001 −.0009 NE*Q (.0009) (.0009) (.0009) Year fixed effects included included Brand fixed effects included included Brand-year fixed effects included (*coefficient significant at 10% level; **coefficient significant at 5% level)

33 Table 10: Results from estimation of (1) and allowing advertising expenditure to have a diﬀerent eﬀect in Quebec Number of observations: 396

Dependent variable: Sjrt

B-FE B-Y-FE B-FE-IV

Coef. Coef. Coef. −.0021∗∗ −.0025∗∗ −.0085∗∗ Price (.0005) (.0005) (.0033) −.0024 −.0056∗ −.0032 Advertising (.0025) (.003) (.0031) .0016 .0034 −.0006 Lagged Advertising (.0021) (.0025) (.003) .0049 .0057 .0076 Advertising *Q (.0036) (.0036) (.0046) Established .002∗ .0019∗ .0018 E ∗ Q (.0011) (.0011) (.0014) −.0016∗∗ −.0016∗∗ −.0019∗∗ NE ∗ Q (.006) (.006) (.008) Year fixed effects included included Brand fixed effects included included Brand-year fixed effects included (*coefficient significant at 10% level; **coefficient significant at 5% level)

34 Table 11: Results from estimation of (2) Number of observations: 396

Dependent variable: Sjrt

B-FE B-Y-FE B-FE-IV Coef. Coef. Coef.

−.0007 −.0011∗ .0062 Price (.0006) (.0007) (.0063) .002 −.0001 .0004 Advertising (.0022) (.003) (.003) .0053∗∗ .0072∗∗ .005∗∗ Lagged Advertising (.002) (.0024) (.0024) Established .0007 .00044 .0043 E*Atl (.001) (.001) (.0035) .0029∗∗ .0028∗∗ .0046∗∗ NE*Atl (.0008) (.0007) (.0018) −.0056∗∗ −.0058∗∗ −.0035 E*Ont (.001) (.001) (.0022) .0014∗ .0013∗ .0032∗ NE*Ont (.0008) (.0007) (.0019) −.0008 −.0008 −.0025 E*MS (.001) (.001) (.0019) .0018∗∗ .0018∗∗ .0007 NE*MS (.0007) (.0007) (.0013) .000003 .00013 −.0022 E*Alb (.001) (.001) (.0023) .0013∗ .0014∗ −.00009 NE*Alb (.0007) (.0007) (.0015) −.0055∗∗ −.0054∗∗ −.008∗∗ E*BC (.001) (.001) (.002) .00007 .0002 −.0027 NE*BC (.0008) (.0008) (.0027) Year fixed effects included included Brand fixed effects included included Brand-year fixed effects included (*coefficient significant at 10% level; **coefficient significant at 5% level)

Figure 1-Established Children's Brands

0,045 0,04 0,035 0,03

0,025 Quebec 0,02 ROC age Shares r 0,015 Ave 0,01 0,005 0

s s p e ts s rm k ri Pops la ombs n F phabi Cha d l yc y A k Cor te one ugar C . os P. S K . Froot Loops r . H Luc K F P. M P G K.

Figure 2-Non-Established Children's Brands

0,018 0,016 0,014 0,012 Quebec ares 0,01 h 0,008 ROC 0,006 verage S 0,004 A 0,002 0 GM C. T. GM Count GM F. T. GM GM GM GM Trix K. C. M. P. Q. Cap'n Crunch Chocula Crunch Golden Nesquick Reese's Buns Pebbles Crunch Grahams Puffs

Figure 3-Established Adult/Family Brands

0,1 0,09 0,08 0,07 0,06

Shares Quebec 0,05 ROC age r 0,04 0,03 Ave 0,02 0,01 0

os os es es es W ch i ri ran ran l K ak pi ia un eer l B heats B is ddi r h hee Fl in re ed S Al n W is pec h iz C . e Kr . C K ini c . S S est C .N Ra i K . v GM . Cor . M . R P H K K . ar K K poon S H GM S Q. P.

Figure 4-Non-Established Adult/Family Brands

0,045 0,04 0,035 0,03 ares h 0,025

S Quebec e g 0,02 ROC era 0,015 Av 0,01 0,005 0

s s s s t s e r s s t s o 1 x e sp m es es ix la ix e r es h e t an at n es a fe i e rio e ti io i u p gh l i b to k nc br ill a ch e a i W. ies ran an re r r a li ak ak s no sp fi a u n h br d . b Br F Li a ee b ee Ch eer Cr l l i Ri i ec M u d S dd w . u h Fi . he h l yl F F Cr ra h V Fl Cr d Fi h of O Cr in Br W h e re n o Q q Ch N C st G . Mus Kr it . n s d it hr ed 0% r S C . W ea ps an rn Ju t e w K an d ug es d e w L t in th r o K. K c t a n ai d . S st Sh Co ch a C. GM ed H F. a atm B Fa k Br on i ro ch a R d oa e 10 st . C K. . u n . re W. H t pl P. Q. un O gr O wi an N w al P lm Fr bo u ts P Q. ro GM ti s r o ci A lls S. G. ey Cr M A. F ul d B L ney Ri P. i B Nu . Sh S Ma t M M. GM l H. . o pe y . P S P. on . G G Bu . K S H ey P . H P es GM n K H . err P. on P . v M M ra . Al K nb P ar G K K. a H H l B P. Cr Q. . Al P. K