Journal of Business Research 52 (2001) 277 ± 291

An empirical analysis of the determinants of category expenditure William P. Putsis Jr.a,*, Ravi Dharb a

London Business School, Sussex Place, Regent's Park, London NW1 4SA, UK b Yale School of Management, New Haven, CT, USA Received 1 July 1998

Abstract Previous research on the effectiveness of promotions by ``private label'' and ``national'' brands has focused primarily on its impact on relative market share. This research has documented the asymmetric nature of promotions between weaker (e.g., private labels) and stronger (e.g., national) brands [Blattberg RC, Wisniewski KJ. Price induced patterns of competition. Mark Sci 1989;8:291 ± 309, Fall; Allenby GM, Rossi PE. Quality perceptions and asymmetric switching between brands. Mark Sci 1991;10:185 ± 204, Summer]. This has led to suggestions that private label promotions make little sense since they are relatively ineffective in stealing share from national brands. However, recent research [Ailawadi K, Neslin SA. The effect of promotion on consumption: buying more and consuming it faster. J Mark Res 1998;35(3):390 ± 98, August; Chandon P, Wansink B. When and why does consumer stockpiling accelerate consumption volume? Unpublished manuscript, London Business School, 1999] suggests that promotions can result in category expansion in addition to brand switching Ð promotions may not simply be a zero-sum game. If promotions can indeed increase total category consumption they may not need to steal share in order to be profitable. Thus, understanding the determinants of category expenditure can be an important part of properly understanding the market interaction between private labels and national brands. This paper explores the determinants of total category expenditure. Building on previous research that has focused on a limited number of categories, this research attempts to produce more general results by using IRI data from 1991 and 1992 to estimate category-level expenditure equations for private label and branded products across 135 food product categories and 59 geographic markets. The results support the premise that national brand and private label promotion can have a significant effect on the level of category expenditure. Furthermore, the results suggest that there can be significant differences across markets and across categories. D 2001 Elsevier Science Inc. All rights reserved. Keywords: Promotion; Category expansion; Private labels

Competition between manufacturer ``national'' and retailer ``private label'' brands, a primary concern of marketing managers, has taken on a greater sense of urgency this decade. Indicative of an increase in competition, private label brands in US supermarkets reached an all-time high unit market share of 20.8% in the third quarter of 1997, according to IRI (BrandWeek, 11/24/97). Following the growth of private labels, the academic marketing literature has been quick to examine the major determinants of private label success. Early research focused on the causes underlying the variation in the market share of private label products across categories (Sethuraman, 1992; Sethuraman * Corresponding author. Tel.: +44-171-706-6733; fax: +44-171-7241145. E-mail addresses: [email protected] (W.P. Putsis Jr.), [email protected] (R. Dhar).

and Mittelstaedt, 1992; Hoch and Banerji, 1993; Quelch and Harding, 1996; Narasimhan and Wilcox, 1998). Sethuraman (1992), for example, identifies 12 marketplace factors as potential determinants of private label success. These factors include retail sales volume, average retail price, price differential between the private label and national brands, and the amount of retail private label versus national brand price promotion. Since the key differential competitive advantage of private labels in relation to national brands is their lower price, a number of studies have focused extensively on price effects using both across and within category data. The seminal work in this area shows that response to price promotion is asymmetric Ðwhen national brands are price promoted, they are more likely to draw sales from the lower quality private labels, whereas price promotions by low quality brands do not induce the same level of

0148-2963/01/$ ± see front matter D 2001 Elsevier Science Inc. All rights reserved. PII: S 0 1 4 8 - 2 9 6 3 ( 9 9 ) 0 0 1 0 3 - 4

278

W.P. Putsis Jr., R. Dhar / Journal of Business Research 52 (2001) 277±291

switching from national brands (Blattberg and Wisniewski, 1989; Allenby and Rossi, 1991). In a meta-analysis of national brand and store brand cross-promotional price elasticities, Sethuraman (1995) finds that the asymmetrical relationship is moderated by the market share of the national brand. One limitation of this research is that it fails to account for the ability of promotion to expand the category. Until recently, it was generally assumed that promotion is a zero-sum game (see, e.g., Blattberg and Neslin, 1990, Chap. 5), i.e., that it simply shifts consumption from one period to the next (purchase acceleration) or from one brand to another (brand switching). Recent evidence, however, suggests that sales promotion may be able to increase overall category demand by modifying the rate of consumption. For example, Ailawadi and Neslin (1998) employ a model that allows the usage rate to vary with household inventory. Using single-source scanner data for two product categories, they show that fewer stockouts and faster usage rates can indeed increase primary category demand. Specifically, they find that there is a 30% increase in yogurt consumption and an 11.5% increase in ketchup consumption due to increased stockpiling when an item is on feature or display. Consistent with behavioral and economic theory, Chandon and Wansink (1999) demonstrates that endogenous or promotion-driven purchasing leads to increased consumption for fruit juice and biscuits, but not for laundry detergent. Category expansion can have important managerial implications (Blattberg and Neslin, 1990; Ailawadi and Neslin, 1998). For example, if higher levels of promotion can increase the total size of the market, then analysis of market share tells only part of the story in determining the effectiveness of promotions. It is possible for market share to fall, but for total dollar sales to increase as the result of category expansion. Thus, an understanding of how promotions by store and national brands may potentially affect total category revenue is a crucial part of making optimal decisions regarding the marketing mix. Further, for retailers engaged in category management practices, category revenue generation is an important managerial consideration. However, despite the potential importance of the subject, the only studies directly addressing this subject are the two papers cited above (Ailawadi and Neslin, 1998; Chandon and Wansink, 1999). While providing valuable information on a limited number of categories, both papers suggest that there are likely to be significant differences across a broader set of categories. The objective of this paper is to conduct an exploratory empirical investigation into the structural factors that drive category expenditure. Conceptually, category expenditure may depend upon ability of the firms in the marketplace to influence demand by changing the price and promotion mix (demand side effects) and also the ability of the firms to raise market prices across geographic markets depending upon the nature of local

competition (supply side effects). Building on previous research that has demonstrated that individual market characteristics influence category demand and price elasticity (Blattberg et al., 1978; Hoch et al., 1995), and that the competitive structure in an industry influences price response (e.g., Weiss, 1989; Cotterill et al., 2000), we focus on differences that may exist across categories, markets, and between weaker versus stronger brands. We emphasize the fact that, as the first attempt at a broad analysis across markets and categories, this research is meant to be exploratory in nature. Consequently, we begin with a parsimonious framework for analyzing the impact of promotion on category expenditure. This framework leads to a descriptive empirical model, which allows us to examine 135 categories and 59 geographic markets. 1. The impact of promotion on category expansion Ð a parsimonious framework for analysis We begin by presenting a parsimonious framework that provides a vehicle for assessing the different factors influencing the ability of promotions to expand the category and use this framework below to develop hypotheses pertaining to the specific variables used in the empirical analysis. In the interest of parsimony, suppose that the market we are studying is a differentiated duopoly. Imagine that both products are competing in a specific geographic area, where i = {1, . . . , I } denotes the category and j = {1, . . . , J} denote the geographic area, while Pkij denotes the price charged at the retail level for product k (k = 1,2) in the i-th category and the j-th geographic market. Similarly, Qkij denotes the quantity sold of product k in category i and market j. In addition to price, each product is sold at the retail level with the support of temporary price reductions, denoted by gkij, and merchandising (feature and display) activity denoted by mkij. For now, we are not concerned with how these retail decisions are made, i.e., if the manufacturer or retailer initiated the decision, the amount of passthrough, competitive response, and the like. What we are concerned with here is the ultimate impact on total category size. Further, given the empirical focus of the paper, we formulate the model throughout using variable definitions consistent with the data used in the empirical analysis. For example, in the data used in the study, gkij is measured as the weighted average price reduction per unit. Consumer demand in the i-th category and the j-th market is a function of marketing mix and consumer demand variables such as income and population. Denote the set of market-level demand-shift variables as dj. Thus, we can represent demand for the k-th product as follows (Eq. (1): Qijk ˆ qijk …Pkij ; Plij ; mijk ; mijl ; gijk ; gijl ; dj †; where l 6ˆ k:

…1†

We can define total category expenditure in a given period (length of period considered and long run versus short-run effects will be discussed later) for category i,

W.P. Putsis Jr., R. Dhar / Journal of Business Research 52 (2001) 277±291

market j (EXPENDij) as the sum of the total expenditure on good one (EXPEND1ij) and good two (EXPEND2ij), respectively. Thus, EXPENDij simply equals the sum of the quantity of each good sold times its net price: EXPENDij  EXPENDij1 ‡ EXPENDij2  …P1ij

gij1 †Qij1 ‡ …P2ij

gij2 †Qij2 :

…2†

We can use this framework to derive specific hypotheses related to the ability of marketing mix variables to expand total category expenditure. To illustrate, imagine that a temporary price reduction for good 1 is instituted. What effect will that have on total category expenditure? Dropping the i and j subscripts for ease of presentation, the partial derivative of total category expenditure, EXPEND, with respect to g1 equals: @EXPEND=@g1 ˆ

Q1 ‡ …P1 g1 †@Q1 =@g1 ‡ …P2 g2 †@Q2 =@g1 :

…3†

In the case of linear demand (Zenor, 1994), Q1 ˆ a0 ‡ a1 P1 ‡ a2 P2 ‡ a3 g1 ‡ a4 g2 ‡ a5 m1 ‡ a6 m2 ‡ a7 d 1 Q2 ˆ b0 ‡ b1 P1 ‡ b2 P2 ‡ b3 g1 ‡ b4 g2 ‡ b5 m1 ‡ b6 m2 ‡b7 d2 ; …4† substituting Eq. (4) into Eq. (3), we can calculate, @EXPEND=@g1 ˆ

Q1 ‡ a3 …P1

g1 † ‡ b3 …P2

g2 †: …5†

Thus, we can break down the change in total category expenditure into the change in EXPEND1 due to the change in g1 (equal to Q1 + a3(P1 g1)) and the change in EXPEND2 due to the change in g1 (equal to b3(P2 g2)). In total, the change in EXPEND due to a unilateral change in g1 would depend upon the size of the (own and cross) promotional elasticities, relative prices and the magnitude of the temporary price reduction. Similar relationships can be derived for each of the variables considered in our analysis Ðwe present details on each hypothesis for all relevant variables in the next section.

2. Hypothesized effects and key variables This framework, combined with previous research on promotion and demand response, provides us with a convenient vehicle for examining the impact of promotions for ``weaker'' (e.g., private label) brands versus ``stronger'' (e.g., national) brands. Unfortunately, the limited amount of previous research on category expansion provides little insight into the development of a comprehensive theory regarding the determinants of category expenditure. Accordingly, we follow recent work by Hoch and Dhar (1997) and provide a brief conceptual

279

framework that guides the selection of consumer characteristics and competitive variables, building on the parsimonious model presented above. Our focus is on the variables that influence consumer demand elasticities (demand-side factors) and those that provide firms with an ability to raise price (supply-side factors). We address each in turn. 2.1. Demand-side factors Recent research has produced a great deal of evidence pertaining to marketing mix response for both price (e.g., Tellis, 1988; Hoch et al., 1995; Sethuraman, 1995; Narasimhan et al., 1996; Danaher and Brodie, 1998) and nonprice (e.g., Blattberg and Neslin, 1990) instruments. We use the previous research on ``own'' and ``cross'' instrument response to assist in guiding our hypotheses. 2.1.1. Regular (non-deal) price We begin by examining the impact of regular price changes by private labels and national brands on total category expenditure. The partial derivative of category expenditure (Eq. (2)) with respect to national brand price (P1) given demands in Eq. (4) is equal to ( Q1 + a1P1 + b1P2). Similarly, the partial derivative of expenditure (Eq. (2)) with respect to private label price (P2) is (Q2 + a2P1 + b2P2). From these expressions, both @EXPEND/@P1 and @EXPEND/@P2 are expected to be negative. To illustrate this analytically, note that a1 < 0 and b2 < 0 (demand is downward sloping in price; Tellis, 1988; Danaher and Brodie, 1998), and that b1 > 0, while a2 is expected to be zero or positive but close to zero (national brand can steal share from private labels by cutting price, but private labels have a difficult time stealing share from private labels; Blattberg and Wisniewski, 1989; Allenby and Rossi, 1991). Further, we expect that P1 > P2 (national brands are generally priced higher than private labels; Quelch and Harding, 1996; Cotterill et al., 1999; also see Table 2 below) and that a1 > b1 and b2 > a2 (own-price effects are greater than cross-price effects, Deaton and Muellbauer, 1980). Incorporating these expected relationships into the two partial derivatives @EXPEND/@P1 and @EXPEND/@P2 suggests that both partial derivatives are negative. Intuitively, we can think of these relationships as follows. For most if not all of the categories studied here, demand is elastic, i.e., the own-price demand elasticity is less than 1.0 (Tellis, 1988; Hoch et al., 1995; Danaher and Brodie, 1998). This is true for both private labels and national brands (Cotterill et al., 1999). Now, imagine that a private label price cut in enacted. Since demand is elastic, the decrease in private label price increases private label revenue. Since private labels cannot generally steal share from national brands (Blattberg and Wisniewski, 1989; Allenby and Rossi, 1991), there is likely to be little change in national brand revenue as a

280

W.P. Putsis Jr., R. Dhar / Journal of Business Research 52 (2001) 277±291

result of the private label price decrease. Thus, the total effect of a private label price cut is to increase total category revenue (hence a negative relationship). For national brands, a national brand price cut will increase total revenue for national brands (again, due to elastic demand). Further, this increase in revenue is expected to be greater than any resulting decrease in private label revenue since (a) the ``own-price'' effect of a national brand price cut is almost always greater than the ``crossprice'' effect on private label sales and (b) national brands are priced higher than private labels. Thus, we expect that price is negatively related to total category expenditure for both private labels and national brands. 2.1.2. Price promotion The rationale for assessing the impact of temporary price cuts is almost identical to that of regular price cuts. Similar to the rationale given above, we expect that @EXPEND/@g1 and @EXPEND/@g2 will be positive (see Eq. (5) for the partial derivative for national brands, with the private label expression analogous).1 Note that the expected sign for temporary price reductions is opposite that of regular price since g1 and g2 are expressed as percent price reductions, as opposed to regular price. Thus, we expect that temporary price reductions are positively related to total category expenditure for both private labels and national brands. 2.1.3. Non-price promotion In general, non-price promotion (display and feature activity) is expected, ceteris paribus, to produce an increase in the demand for a brand (see, e.g., Blattberg and Neslin, 1990). This should increase category expenditure since the increase in demand is not accompanied by an associated price reduction. In addition, merchandising (display and feature) activity is likely to draw additional attention towards the category, increase the likelihood of unplanned purchase, accelerate planned purchase and/or increase category substitution towards the promoted brand. Further, since (i) the average price charged by national brands in each category is higher and (ii) there is evidence that national brand feature and display have a greater impact on demand than private label display and feature activity (Cotterill et al., 2000), merchandising activity by national brands is likely to have a greater effect on category expenditure. Hence, we expect that both national brand and private label non-price promotions (display and feature) increase 1 Similar to the own (regular) price, the own-price promotion effect is greater than the cross-promotion effect (see, e.g., Blattberg and Neslin, 1990, pp. 351 ± 55). For example, Narasimhan et al. (1996) report an average own-price promotion demand elasticity (pure price cuts) across 108 categories of 2.3 (s = 0.54), while Sethuraman (1995) reports an average own-price promotion elasticity of 3.23 and an average cross price promotion elasticity of 0.54 (s = 0.30).

category expenditure, but that national brand promotions will have a larger effect on total category expenditure. To illustrate this mathematically, we note that @EXPEND/ @m1 = (P1 g1) a5 + (P2 g2) b5. Since both (P1 g1) and (P2 g2) are positive and since we generally expect that (i) (P1 g1) > (P2 g2), (ii) a5 > 0 and b5 < 0, and (iii) a5 > b5, the expression @EXPEND/@m1 should be positive. An analogous argument can be constructed for @EXPEND/@m2. One important potential caveat should be noted, however. If promotion by lower priced brands (e.g., private labels) stole share from higher priced (e.g., national brands) and promotion did not expand the category, private label promotion would result in a decrease in category expenditure. This provides a convenient empirical test of our proposition Ð if this were true, then we would observe a negative relationship between private label promotion and category expenditure, as opposed to the positive one hypothesized. 2.1.4. Market characteristics Previous research has suggested that price sensitivity is related to individual market characteristics (Blattberg et al., 1978; Hoch et al., 1995; Danaher and Brodie, 1998). One would expect this intuitively since considerable variation in consumer characteristics, demographics and preferences from one market to another is likely. This suggests that when conducting an analysis across markets, it is important to not only examine the impact of factors influencing individual demand elasticities, but also the underlying causes of cross-market differences in consumer response across the marketing mix. In general, we note that at a given price, the greater the size of the own-price response (parameters a1, b1), the lower the level of total category expenditure (since a1 and b1 are negative; see Eqs. (2) and (4)). Thus, any factor that makes a1 and/or b1 more negative (i.e., demand more elastic) serves to decrease overall category expenditure. Intuitively, this makes sense since firms facing elastic demands have a more difficult time raising price above marginal cost. Thus, any factor that increases (decreases) demand elasticity is expected to have a negative (positive) impact on total category expenditure. Hoch et al. (1995) find that the higher the percentage of elderly and the higher the percentage of consumers who are black and Hispanic in a geographic area, the lower the price elasticity. They do not offer a comprehensive analysis of why this is so, but there have been numerous explanations offered in the economics literature. For example, elderly consumers may find it more difficult to get around and comparison shop across stores or travel longer distances to access less expensive retailers. Further, older consumers may have established category knowledge and brand loyalties. In addition, much has been written about the under-representation of large, high volume retailers in ethnic urban neighbor-

W.P. Putsis Jr., R. Dhar / Journal of Business Research 52 (2001) 277±291

hoods Ða smaller choice set (number of available retailers) in these neighborhoods would be consistent with a lower price sensitivity. While these explanations may be speculative, Hoch et al. (1995) demonstrate empirically that older and more ``ethnic'' populations do have lower demand elasticities. Thus, as per the discussion above, we would expect that they also have higher levels of category expenditure. Finally, with respect to income, two results are possible. First, Blattberg et al. (1978) suggest that higher levels of income may be associated with a higher opportunity cost of time, suggesting lower price sensitivity and, thus, a positive relationship between income and category expenditure. Alternatively, as income rises, there is evidence (see, e.g., Deaton and Muellbauer, 1980) to suggest that less is spent on food to be prepared at home and more on meals away from home as income rises. Thus, it is possible that increases in income results in a reduction in the total amount spent on food items purchases in supermarkets (food product categories only are studied in our analysis). This would suggest a negative relationship between income and total category expenditure for food items. Consistent with these conflicting views, Hoch et al. (1995) find no significant relationship between income and price sensitivity. Consequently, we have no a priori prediction about the relationship between income and total category expenditure. 2.2. Supply-side factors Fortunately, we have a very rich and detailed literature in economics to guide hypotheses pertaining to supplyTable 1 Predicted impact of key determinants of category expenditure NB = National brand; PL = private label. Key determinants

Expected sign

Demand-side factors PL price NB price NB Price Reductiona PL Price Reductiona NB feature PL feature NB display PL display Local average income Higher ethnic percentage Age of local population

+ + + + + + ? + +

Supply-side factors Private label distribution Manufacturer CR4 Number of brands Supermarket ratio Local retailer CR4

+ + + + +

a Since Price Reduction is expressed as percent price reduction, the expected sign is the opposite of regular price.

281

side relationships. Previous research in the Industrial Organization (IO) literature (see, e.g., Weiss, 1989) has suggested that structural factors such as increased concentration can elevate market prices due to a higher level of market power available to firms in the market. For example, Cotterill et al. (2000) have demonstrated that both the national IO structure (manufacturer four-firm concentration ratio) and local market concentration (local retail four-firm concentration and percent of total market sales by supermarkets) can elevate the prices charged to consumers at the retail level. Further, Putsis (1997) has shown that, controlling for the number of manufacturers present, a higher number of brands competing in a category can result in increased market prices.2 All of this suggests that factors such as the number of brands present in a category and the level of national and local concentration can serve to increase the market power of brands competing in a given market, thereby increasing their ability to raise prices. Ceteris paribus, we expect that any factor that increases the ability of firms to raise price above marginal cost results in an increase in total category expenditure. Consequently, we hypothesize that higher levels of local market (retailer) concentration, national manufacturer concentration, and the number of brands competing in a category are all expected to result in a higher level of total category expenditure. We follow the conclusions drawn in Weiss (1989) and use the four-firm concentration ratio (CR4) at the manufacturer and retailer level. We also follow previous research addressing retailer concentration (Marion, 1979; Cotterill, 1986; Cotterill et al., 2000) and include the retail CR4. Finally, we suggest that increased private label distribution is indicative of increased attention given to a category by manufacturers and/or retailers (Hoch and Dhar, 1997), thereby suggesting a higher level of category expenditure. Further, increased availability of lower-priced private label products may attract additional buyers to the category. This measure is consistent with the category development index used in Raju (1992). Based on this discussion, Table 1 provides a summary of the key variables and their hypothesized impact on category expenditure. 2.3. Long-run versus short run effects The definition of expenditure given in Eq. (2) and discussed earlier was given for a ``period,'' but the

2

To illustrate, imagine identical two categories, each with three manufacturers and consumers are positioned equidistant in attribute space. The only difference between the two categories is that the manufacturers in category A have introduced three brands, each uniquely positioned in attribute space, while the manufacturers in category B have introduced 100 brands spanning attribute space. The 100 brands in category B make it more difficult for a new entrant to find a profitable entry position. Thus, the brand proliferation strategy creates an entry barrier (Schmalensee 1978).

282

W.P. Putsis Jr., R. Dhar / Journal of Business Research 52 (2001) 277±291

discussion did not specify the duration of the period being studied. Most of the previous research addressing promotion response (see Blattberg and Neslin, 1990; Dekimpe et al., 1998) and the limited work on category expenditure (Ailawadi and Neslin, 1998; Chandon and Wansink, 1999) have looked at the impact of promotions using weekly data. One problem with weekly data is that much of the promotion response may be due to consumer stockpiling (see Blattberg and Neslin, 1990, Chap. 5). Keane (1997) has demonstrated that even very large short-run increases in sales due to a promotion (e.g., 313% in his study) produce considerably smaller sustainable increases in sales (12% in his study). This is consistent with the findings of Dekimpe et al. (1998). Consequently, we focus on the long-term sustainable impact of promotion here by examining sales patterns over two years of data. We note that both short-run (e.g., weekly as defined in previous research) and long-run (e.g., over an entire year as defined here) effects are likely to be relevant managerially. Information on shortrun effects are important to managers (e.g., manufacturer product managers) in need of boosting short-run sales, while information on long-run effects are important to managers interested in achieving results that are sustainable. Consequently, this study should complement recent research examining short run effects (e.g., Chandon and Wansink, 1999).

In our empirical analysis below, we examine long-run sustainable changes in consumption, defining all measures for a calendar year. Thus, we address the question if promotional intensity (e.g., feature activity) increased by x percent in a given year, how much (if at all) would total category consumption increase by? Given that, for each category, this captures multiple interpurchase cycles, any observed increases in category expenditure should represent real, permanent increases, and not simply increases due to temporary consumer stockpiling. 3. Data and variable definitions The data used in this study are annual IRI market-level data on food products across 59 geographic markets and 135 categories from 1991 to 1992. For each market and category, we have standard IRI measures across price, unit and dollar volume sales, and merchandising (we will present an exact variable listing below). The available data include measures of feature and display activity, price promotions, unit and volume sales figures, price promotion, etc., for each of the 135 food categories and in the 59 geographic markets. We restricted our analysis to include only those markets and categories for which private label products have been introduced in the market by 1991. This left a final balanced panel of 7823 usable

Fig. 1. Variables used in the empirical analysis.

W.P. Putsis Jr., R. Dhar / Journal of Business Research 52 (2001) 277±291

observations varying both at the category and at the local market level. In order to address the supply-side influences discussed above, we also developed measures for supply-side characteristics at the market and category level. In particular, we developed measures for the brand-level four-firm concentration ratio (market share of the top four brands), the number of brands competing in the local market for each category directly from the IRI data. Further, we obtained measures of the four-firm local retail concentration ratio (market share of the top four retailers), and the shopping ``distribution'' (the ratio of supermarket to convenience store sales) of the local market from Progressive Grocer. Additional data was also obtained from Progressive Grocer on the demographic characteristics of the IRI geographic markets. This provided information on each market's population, income distribution, ethnicity (only information on the percent of the local market of Hispanic descent is available), and population. Consistent with previous research (e.g., Sethuraman and Mittelstaedt, 1992; Slade, 1995; Putsis and Dhar, 1998; Cotterill et al., 2000), composite branded and private label variables were created for the 135 product categories and 59 markets. In terms of specific variables used in the empirical analysis, two price variables are included in the analysis below Ð branded price (BRPRICE) and private label price (PLPRICE). Similarly, two price reduction variables were included (BRPRICEREDN and PLPRICEREDN). Four non-price promotion variables were available and included in the empirical specification Ðbranded display (BRDISPLAY) and feature (BRFEATURE) and private label display (PLDISPLAY) and feature (PLFEATURE). In addition, variables controlling for individual market differences on the demand side were constructed. Accordingly, income (INCOME), percent Hispanic (HISPANIC), and average age (AGE) represent demographic differences across markets. Further, since markets vary by size, we include a variable controlling for local population differences (POP). On the supply side, IO effects discussed earlier suggest that variables representing both national IO structure (manufacturer four-firm concentration ratio, MCR4) and local market concentration (local retail four-firm concentration, GROCCR4, and percent of total market sales by Supermarkets, SRATIO) have important implications for total category expenditure. Further, since brand proliferation has been linked to elevated equilibrium price in some industries, we include a measure of the number of national brands (NBRANDS) present in the individual category/market (see Putsis, 1997 for an explanation of this specification). Finally, note that since there are likely to be unit differences across categories and between private label and branded products, two variables reflecting average package size (BRUPERV and PLUPERV) were included in the analysis to control for package size differences. Similarly, since private label

283

distribution is likely to vary across categories and across markets, a variable controlling for this distribution (PLDISTN) was also included the analysis. A summary of the variables for the each market and the category is specified in Fig. 1 above. 4. Empirical analysis Given our objective of specifying a descriptive model of the determinants of category expenditure, choosing a parsimonious functional specification was of paramount importance. Thus, in specifying a functional form, we follow previous research in economics (e.g., Kelton and Weiss, 1989) and specify a generalized power function. This functional form is reasonably flexible and has been estimated in a variety of settings. Note that there are both similarities and differences between this specification and one derived from a traditional expenditure function approach, such as that of the linear expenditure system (see, e.g., Deaton and Muellbauer, 1980, Chap. 3). Following Kelton and Weiss (1989), since some variables are already defined in percentage terms, the log transform is taken only for those variables not already defined in percentage terms. This implies a generalized power function and a log-linear form, with the coefficient on each variable directly interpreted as an expenditure elasticity as follows: LNEXPENDITURE ˆ z0 ‡ z1 BRPRICE ‡ z2 PLPRICE ‡ z3 BRFEATURE ‡ z4 BRDISPLAY ‡ z5 BRPRICEREDN ‡ z6 PLFEATURE ‡ z7 PLDISPLAY ‡ z8 PLPRICEREDN ‡ z9 PLDISTN ‡ z10 MCR4 ‡ z12 NBRANDS ‡ z13 POP ‡ z14 INCOME ‡ z15 HISPANIC ‡ z16 AGE ‡ z17 SRATIO ‡ z18 GROCCR4 ‡ z19 BRUPERV ‡ z20 PLUPERV ‡ w: …6† We begin by estimating Eq. (6) for each of the 59 geographic markets and 135 categories. In conducting this empirical analysis, it was important to address endogeneity in estimation. In the presence of endogenous right hand side variables, estimating Eq. (6) directly by OLS violates an assumption in the classical linear regression model Ð the variables are not fixed in repeated samples. Consequently, following Cotterill et al. (2000), we estimated the reduced form price equations (price as a function of all endogenous variables in the system), retaining the fitted values. The fitted values were then used as regressors in the estimation of Eq. (6). In addition, we address the potential endogeneity of the trade promotion variables

284

W.P. Putsis Jr., R. Dhar / Journal of Business Research 52 (2001) 277±291

through the use of instrumental variables following Cotterill et al. (2000). The principle is similar to the approach taken by Berry et al. (1995). Specifically, each promotional vehicle for market i, category j, is expressed as a function of the promotional activity in each of the other j (j 6ˆ i) markets, using the fitted value as the instrument. Note that in order for this approach to eliminate the endogeneity bias, the equation errors for each promotion instrument have to be independent. This requires that display and feature decisions, for example, are made on a market by market (or chain by chain) basis. All estimations were performed using Limdep v. 6.0. This analysis enabled us to examine the impact of each variable on total category expenditure for individual markets and categories. However, conducting this analysis provided us with 194 (59 individual market plus 135 individual category results) separate sets of estimates. As argued by Hoch et al. (1995, p. 27), there is ``no a priori theoretical reasons why the coefficients should have the same magnitude across categories [or markets].'' All of this makes it difficult to arrive at general conclusions. Consequently, we also conducted a pooled analysis in order to obtain some degree of generalizability. Following the format used by Hoch et al. (1995), we first discuss the individual category and market results and then turn to the pooled results. Limitations of the approach taken are discussed in the Conclusions, Limitations and Future Research section below. 5. Results 5.1. Individual category analysis In an attempt to exploit some of the richness of the data, we began by running individual regressions for each of the 59

markets and 135 categories. Table 2 presents mean values for all variables used in the analysis. Note that in the regressions, all variables are expressed in either log or percentage terms, while the actual values of the variables are reported in Table 2 for ease of interpretation. For each variable, we also present the raw correlation between the variable and total expenditure for 1991 and 1992, respectively. Since presenting results for all of the 135 categories was impractical, we chose 10 that were representative of the larger set of categories in terms of overall fit and individual variable signs; Table 3a and b presents the results for these 10 categories (all equations were estimated with a constant term, but estimated constant was omitted from Tables 3 and 4 to avoid additional clutter). Standard errors are in parenthesis and all equations have 59 observations (one observation for each market). For each of the individual categories, the model performed quite well. The adjusted R2 ranged from a high of 0.660 in the tomato products category to a low of 0.173 in the spices category. This is very good given that we are conducting a cross-section analysis (see, e.g., Kennedy, 1985, p. 26). The signs for the individual coefficients are generally consistent across categories (and are broadly consistent with the pooled results presented below). For example, only 4 of the 40 non-price variables in Table 3a are negative (i.e., of the ``wrong'' sign). The results in Table 3a and b are generally consistent with the predictions made above. For example, on the demand side (Table 3a), all of the estimated price coefficients are negative, while virtually all of the temporary price reduction variables are statistically significant and positive. Non-price promotion variables are generally positive and significant. Age and ethnicity coefficients were generally estimated to be positive and significant. On the supply side (Table 3b), the coefficients on the structural variables (MCR4, NBRANDS, SRATIO, GROCCR4, PLDISTN)

Table 2 Mean values expressed in actual units Variable

Units

Mean value 1991

Branded price per unit Private label price per unit Average branded price reduction Average private label price reduction Percent sold on feature, branded Percent sold on display, branded Percent sold on feature, private label Percent sold on display, private label Population Household income Percent Hispanic Average age Private label distribution Manufacturer four-firm concentration ratio Number of brands Supermarket-to-convenience store sales Local grocery retain CR4

dollars dollars percent percent percent percent percent percent thousands dollars percent years percent percent integer ratio percent

1.59 1.28 19.46 19.13 6.21 11.23 5.63 10.84 2710 37,385 7.88 33.46 74.42 58.99 36.47 74.19 66.90

Correlation with 1991 expenditure 0.016 0.007 0.123 0.096 0.181 0.099 0.206 0.030 0.444 0.234 0.099 0.130 0.128 0.149 0.380 0.049 0.124

Mean value 1992 1.61 1.28 20.04 19.32 6.25 11.65 5.67 11.25 2667 39,383 8.12 33.30 75.36 58.14 37.86 72.78 68.34

Correlation with 1992 expenditure 0.021 0.011 0.138 0.111 0.190 0.106 0.220 0.036 0.434 0.264 0.096 0.046 0.126 0.156 0.382 0.102 0.150

Adjusted R2

BRUPERV

PLUPERV

GROCCR4

SRATIO

NBRANDS

MCR4

(b) Supply-side factors PLDISTN

AGE

HISPANIC

INCOME

POP

PLDISPLAY

PLFEATURE

BRDISPLAY

BRFEATURE

PLPRICEREDN

BRPRICEREDN

PLPRICE

(a) Demand-side factors BRPRICE

0.2755

0.0004 (0.0014) 0.0020 (0.0024) 0.2818 (0.0912) 0.0133 (0.0025) 0.0010 (0.0020) 1.8244 (0.2732) 1.0025 (0.3560)

1.7732 (0.4381) 1.300 (0.3027) 0.0083 (0.0063) 0.0065 (0.0031) 0.0386 (0.0113) 0.0082 (0.0029) 0.0127 (0.0038) 0.0024 (0.0012) 0.1409 (0.5730) 0.1078 (0.2504) 0.0108 (0.0192) 0.9662 (0.4163)

Bottled water

Table 3 Representative individual category results

0.2087

0.0008 (0.0009) 0.0071 (0.0019) 0.0623 (0.0913) 0.0009 (0.0036) 0.0033 (0.0016) 0.3677 (0.2889) 1.4612 (0.4131)

1.4201 (0.5935) 0.6398 (0.3855) 0.1111 (0.0200) 0.0222 (0.0108) 0.0203 (0.0063) 0.0200 (0.0037) 0.0069 (0.0023) 0.0032 (0.0017) 0.1011 (0.0505) 0.3585 (0.2416) 0.0282 (0.0150) 0.2369 (0.4103)

Coffee

0.4247

0.0023 (0.0008) 0.0012 (0.0012) 0.0890 (0.0403) 0.0039 (0.0015) 0.0015 (0.0007) 0.2743 (0.1449) 0.7521 (0.2489)

0.8066 (0.3442) 0.1920 (0.1440) 0.0055 (0.0022) 0.0002 (0.0011) 0.0016 (0.0018) 0.0023 (0.0011) 0.0007 (0.0011) 0.0020 (0.0007) 0.3782 (0.1836) 0.3803 (0.0899) 0.0211 (0.0066) 0.2210 (0.1505)

Canned fruit

0.4265

0.0049 (0.0012) 0.0267 (0.0069) 0.2842 (0.2729) 0.0136 (0.0046) 0.0129 (0.0033) 0.1173 (0.2950) 0.4085 (0.0793)

1.5652 (1.4380) 1.8132 (0.4677) 0.0864 (0.0226) 0.0424 (0.0102) 0.0398 (0.0123) 0.0283 (0.0054) 0.01864 (0.0046) 0.0100 (0.0026) 0.6262 (0.1485) 0.3946 (0.1041) 0.0594 (0.0165) 0.56395 (0.1344)

Salad dress

0.3096

0.0018 (0.0013) 0.0002 (0.0025) 0.1642 (0.1381) 0.0217 (0.0084) 0.0102 (0.0071) 0.1632 (1.916) 0.5528 (0.3460)

1.2416 (0.7997) 0.2991 (1.6506) 0.0143 (0.0258) 0.1134 (0.0419) 0.0021 (0.0052) 0.0033 (0.0078) 0.0442 (0.0169) 0.0057 (0.0021) 0.5898 (0.3792) 0.5454 (0.5073) 0.0497 (0.0477) 0.5881 (0.2944)

Oils

0.1725

0.0006 (0.0004) 0.0022 (0.0013) 1.102 (0.0194) 0.0080 (0.0016) 0.0001 (0.0007) 0.6561 (0.1338) 1.0637 (0.36701)

0.7351 (0.3119) 0.8460 (0.1729) 0.0019 (0.0024) 0.0004 (0.0073) 0.0082 (0.0031) 0.0024 (0.0010) 0.0010 (0.0010) 0.0041 (0.0009) 0.0013 (0.1941) 0.0917 (0.1076) 0.0307 (0.0067) 0.3807 (0.1785)

Spices

0.6599

0.0060 (0.0019) 0.0059 (0.0017) 0.1034 (0.0586) 0.0028 (0.0017) 0.0011 (0.0009) 0.1119 (0.2242) 1.2284 (0.2597)

0.3246 (0.3055) 0.6414 (0.2617) 0.0086 (0.0017) 0.0001 (0.0012) 0.0102 (0.0029) 0.0096 (0.0016) 0.0091 (0.0025) 0.0022 (0.0012) 0.4492 (0.2481) 0.2070 (0.1403) 0.0076 (0.0092) 0.0625 (0.1941)

Tomato products

0.4026

0.0342 (0.0126) 0.0886 (0.0323) 0.1230 (0.0915) 0.0151 (0.0128) 0.0449 (0.0183) 0.2918 (0.9173) 1.7640 (0.7544)

1.8700 (0.7808) 0.5564 (0.1930) 0.1154 (0.0426) 0.0544 (0.0225) 0.0343 (0.0148) 0.0390 (0.1521) 0.1602 (0.0584) 0.0467 (0.0175) 0.5015 (0.1841) 1.8504 (0.7980) 0.0212 (0.0169) 1.2074 (0.4441)

Vinegar

0.4297

0.0021 (0.0019) 0.0043 (0.0021) 0.0298 (0.0755) 0.0015 (0.0025) 0.0016 (0.0012) 0.1317 (0.3063) 0.6713 (0.6063)

0.4829 (0.5286) 0.2167 (0.3294) 0.0015 (0.0055) 0.0082 (0.0029) 0.0025 (0.0022) 0.0073 (0.0024) 0.0052 (0.0016) 0.0065 (0.0017) 0.5178 (0.4148) 0.1446 (0.1761) 0.0675 (0.0132) 0.4745 (0.3251)

Frozen juices

0.2429

0.0036 (0.0006) 0.0070 (0.0008) 0.0239 (0.0298) 0.0010 (0.0017) 0.0009 (0.0008) 0.1562 (0.1213) 1.1825 (0.2469)

1.3600 (1.3420) 0.5394 (0.1959) 0.0096 (0.00144) 0.0002 (0.0016) 0.0033 (0.0013) 0.0023 (0.0010) 0.0042 (0.0016) 0.0065 (0.0014) 0.5991 (0.2062) 0.1334 (0.1022) 0.0360 (0.0079) 0.8007 (0.2035)

Frozen vegetables

W.P. Putsis Jr., R. Dhar / Journal of Business Research 52 (2001) 277±291 285

Adjusted R2

BRUPERV

PLUPERV

NBRANDS

MCR4

Supply-side factors PLDISTN

PLPDISPLAY

PLFEATURE

BRDISPLAY

BRFEATURE

PLPRICEREDN

BRPRICEREDN

PLPRICE

Demand-side factors BRPRICE

0.6011

0.0093 (0.0015) 0.0155 (0.0027) 0.2890 (0.1119) 0.0396 (0.2744) 0.8972 (0.4492)

0.6016 (0.4345) 0.1445 (0.2877) 0.0432 (0.0081) 0.0037 (0.0032) 0.0072 (0.0032) 0.0017 (0.0035) 0.0054 (0.0015) 0.0101 (0.0015)

Cincinnati

Table 4 Representative individual market results

0.4905

0.0066 (0.0016) 0.0087 (0.0019) 0.6028 (0.1303) 1.1320 (0.3258) 0.8544 (0.4353)

1.222 (0.5514) 0.9254 (0.3693) 0.0311 (0.0087) 0.0025 (0.0054) 0.0144 (0.0056) 0.0270 (0.0040) 0.0004 (0.0003) 0.0041 (0.0020)

Dallas

0.4748

0.0192 (0.0078) 0.0403 (0.0137) 0.1838 (0.3910) 0.4765 (0.2498) 0.9920 (0.4915)

0.6957 (0.3827) 0.6054 (0.2829) 0.0192 (0.0072) 0.0763 (0.0380) 0.0438 (0.0095) 0.0628 (0.0183) 0.0316 (0.0139) 0.0093 (0.0040)

Kansas City

0.6475

0.0100 (0.0015) 0.0098 (0.0027) 0.0740 (0.1173) 1.3816 (0.2411) 0.0068 (0.3729)

0.3180 (0.4122) 0.1317 (0.0297) 0.0214 (0.0057) 0.0106 (0.0025) 0.0247 (0.0067) 0.0081 (0.0027) 0.01740 (0.0060) 0.0010 (0.0018)

Memphis

0.4602

0.0125 (0.0017) 0.0006 (0.0010) 0.3309 (0.1393) 0.4355 (0.2258) 1.1719 (1.2416)

0.1909 (0.1331) 0.2785 (0.0568) 0.0470 (0.0109) 0.0052 (0.0031) 0.0021 (0.0051) 0.0038 (0.0055) 0.0079 (0.0034) 0.0110 (0.0033)

Miami

0.4267

0.0027 (0.0030) 0.0013 (0.0035) 0.5318 (0.2201) 0.9691 (0.6660) 0.8593 (0.8444)

0.7365 (0.9926) 0.2350 (0.5600) 0.1032 (0.0246) 0.0088 (0.0090) 0.0231 (0.0094) 0.0380 (0.0099) 0.0195 (0.0075) 0.0044 (0.0040)

Milwaukee

0.3424

0.0089 (0.0030) 0.0079 (0.0043) 0.6664 (0.1958) 0.5880 (0.3571) 1.3796 (0.6610)

0.1152 (0.0674) 0.8292 (0.3718) 0.0743 (0.0132) 0.0016 (0.0046) 0.0255 (0.0074) 0.0321 (0.0061) 0.0047 (0.0032 0.0075 (0.0031)

New York

0.5662

0.0027 (0.0017) 0.0141 (0.0024) 0.3869 (0.1300) 0.5882 (0.2947) 0.4813 (0.4243)

0.4704 (0.4920) 0.1531 (0.0376) 0.0318 (0.0047) 0.0003 (0.0019) 0.0207 (0.0069) 0.0032 (0.0028) 0.0020 (0.0031) 0.0091 (0.0020)

Philadelphia

0.4845

0.0114 (0.0022) 0.0157 (0.0034) 0.6819 (0.1818) 0.9383 (0.2809) 0.2172 (0.5210)

0.0760 (0.6709) 0.1266 (0.0327) 0.0527 (0.0121) 0.0002 (0.0034) 0.0470 (0.0113) 0.0052 (0.0039) 0.0130 (0.0049) 0.0180 (0.0035)

Sacramento

0.4680

0.0122 (0.0021) 0.0013 (0.0035) 0.1985 (0.1795) 0.6358 (0.2959) 0.4680 (0.4113)

0.3891 (0.5201) 0.1408 (0.0387) 0.0232 (0.0085) 0.0046 (0.0066) 0.0041 (0.0113) 0.0317 (0.0071) 0.0073 (0.0046) 0.0080 (0.0041)

San Francisco

286 W.P. Putsis Jr., R. Dhar / Journal of Business Research 52 (2001) 277±291

W.P. Putsis Jr., R. Dhar / Journal of Business Research 52 (2001) 277±291

are also generally as predicted (Table 1) Ð only 7 out of the 50 estimated coefficients on these variables were of the ``wrong'' sign and statistically significant. There are significant differences in the magnitude of the coefficients on the non-price variables, as well as some variation in the estimated direction of the effect of price (regular and deal) on category expenditure across categories. This is likely due to idiosyncratic category differences (Narasimhan et al., 1996), which are reflected in the estimated expenditure elasticities. For example, feature advertising for items that can be stockpiled (e.g., bottled water, salad dressing and vinegar) seems to have a much larger impact on category expenditure than feature advertising for items that have a limited shelf life (e.g., coffee or spices). Further, expenditure in certain categories (e.g., bottled water) are very sensitive to price, while other categories (e.g., coffee or vinegar) are more sensitive to display or feature advertising. While we conjecture that cross-category differences in (a) the ability to stockpile items, (b) the susceptibility of an item to impulse purchase and (c) overall demand sensitivities will have a significant impact in these cross-category differences, we leave a more detailed examination of these conjectures for future research. Finally, we note that the results suggest that managers interested in strategically expanding a category's size will need to look carefully at the individual category. For example, our results suggest that managers with private label price discretion will only be able to expand the category by reducing private label price in the coffee and oils category. For many categories (bottled water, salad dressing, vinegar and frozen juices), temporary private label price reductions will actually decrease category expenditure. This is not true of non-price promotion. Non-price promotion (feature and display) will almost universally increase category expenditure. Thus, it appears that individual stores (or entire chains) wishing to use private label products as a vehicle to expand a category's overall size, would be much better off using display and feature activity than using price (Putsis and Dhar, 1999). Indeed, for a number of categories, private label non-price promotion expanded the category more than the equivalent promotional intensity on the part of national brands. 5.2. Individual market analysis Table 4 presents the individual market results for 10 representative markets. Similar to what was done with the individual category results, we have attempted to choose the individual markets so that they vary in size and by region of the country. Note that the individual market variables (INCOME, AGE, etc.) drop out of this analysis since the variation observed is cross-category. Each equation has 135 observations and standard errors are in parentheses. As discussed below, since the observed variation in the indivi-

287

dual market analysis is cross-category variation, the first difference form (1992 ± 1991, see Eq. (7) below) was employed here. Again, the model performs well. The individual market adjusted R2 ranged from a high of 0.65 in Memphis to a low of 0.34 in New York. The estimated non-price coefficients are generally consistent from market to market, while the estimated signs on the four price variables show considerably more volatility. Temporary national brand price reductions uniformly increases category expenditure (all 10 estimated BRPRICEREDN coefficients were positive and significant, consistent with the pooled results). However, the effectiveness of private label price reductions varied by market. Only 2 of the 10 private label variables (PLPRICEREDN) are statistically significant (all are positive in sign). With respect to the permanent price variables (BRPRICE and PLPRICE), private label and branded price reductions had a positive effect on category expenditure in Dallas, while only private label price reductions could increase total expenditure in Memphis, Miami, New York, Philadelphia, Sacramento and San Francisco. This highlights the importance of looking at individual markets and/or categories when performing an analysis of this type, consistent with prior research (Hoch et al., 1995). 5.3. Pooled analysis The primary focus of our investigation is on the individual category and market results. However, it is difficult to get a general sense of the pattern of results across 135 categories and 59 markets. Consequently, in order to get some sense of the general pattern of results across all categories and markets (not just the ones presented above), we conducted an analysis pooled across the all categories and markets. We present the pooled analysis and discuss the overall results in the context of both the pooled and the individual analysis below. In conducting the pooled analysis, a number of issues must be addressed. For example, since part of the variation in the data is across categories, estimation of Eq. (6) directly using a single year of data is inappropriate. A careful analysis of cross-category effects precludes analysis of price levels, as it is meaningless to compare the price of a pound of cheese to the price of a can of soup, for example. Accordingly, following Kelton and Weiss (1989), we estimated the first difference of the model when conducting the pooled analysis Ðin short, we take Eq. (6) for 1991 and subtract from Eq. (6) for 1992. This manipulation leaves us with the same functional form, except that each variable is measured as the difference between the value in 1992 and the value in 1991. Eq. (7), which was estimated on the pooled sample, will be referred to as the ``difference'' equation throughout. Note that the notation above implies that we have assumed the parameters z0 through z20 are constant from 1991 through

288

W.P. Putsis Jr., R. Dhar / Journal of Business Research 52 (2001) 277±291

1992. As long as the parameters z1 through z20 are the same in 1991 and 1992, the estimated parameters in the …LNEXPENDITURE92 LNEXPENDITURE91 † ˆ …z092 z091 † ‡ z1…BRPRICE92 BRPRICE91 † ‡ z2…PLPRICE92 PLPRICE91 † ‡ z3…BRFEATURE92 BRFEATURE91 † ‡ z4…BRDISPLAY92 BRDISPLAY91 † ‡ z5…BRPRICEREDN92 BRPRICEEDN91 † ‡ z6…PLFEATURE92 PLFEATURE91 † ‡ z7…PLDISPLAY92 PLDISPLAY91 † ‡ z8…PLPRICEREDN92 PLPRICEREDN91 † ‡ z9…PLDISTN92 PLDISTN91 † ‡ z10…MCR492 MCR491 † ‡ z12…NBRANDS92 NBRANDS91 † ‡ z13…POP92 POP91 † ‡ z14…INCOME92 INCOME92 † ‡ z15…HISPANIC92 HISPANIC91 † ‡ z16  …AGE92 AGE91 † ‡ z17…SRATIO92 SRATIO91 † ‡ z18…GROCCR492 GROCCR491 † ‡ z19…BRUPERV92 BRUPERV91 † ‡ z20…PLUPERV92 PLUPERV91 † ‡ …w92 w91 † …7† difference equation are identical to the parameters in the level equations for 1991 (in order to test the assumption that the parameters are equal across the two periods, we conducted a standard F-test using the 1991 and 1992 data separately and could not reject the null hypothesis of parameter equality at a = 0.01). Thus, the estimated coefficients can be directly interpreted as portrayed in Eq. (6), and we will do so throughout the rest of the section. This formulation has two attractive properties. First, estimating a first difference model controls for first-order fixed effects due to excluded local market and category variables in level regressions. Hausman and Taylor (1981) argue that excluded local market variables in panel data of this type can bias parameter estimates for level equations. They demonstrate that this bias can be avoided by specifying a set of city binary variables. These binary variables drop out of the model when one takes the first difference. This is also true for specifying a set of category binary variables in level regressions to control for excluded variables in individual categories. Thus, taking the first difference allows us to estimate Eq. (7) directly (as opposed to employing a fixed or random effects model on the level data). Second, taking the first difference provides a reasonable solution to perhaps the most difficult problem in private label cross-category analysis Ð quality measurement. Previous researchers have tried various ambitious attempts to include quality in cross category analysis. Hoch and Banerji (1993) use expert survey data, while Narasimhan and Wilcox (1998) use consumer survey data on category risk measures. While reasonable, these approaches have their

respective limitations. In our case, to the extent that quality is constant from 1991 to 1992, differencing eliminates the need to include a quality variable in the analysis Ð a (constant) level of quality drops out of the difference equation. Based upon this discussion and following Kelton and Weiss (1989) closely, the difference Eq. (7) was estimated using data pooled across categories and markets via two-stage least squares. Pooling strategies rest on a set of assumptions that should be tested and addressed in any empirical application. We follow the approach used by Hoch et al. (1995) here closely (the interested reader is referred to Hoch et al., 1995, who discuss the relevant pooling issues in this context in detail). We first test the pooling restriction that coefficients are equal across markets and categories using a likelihood ratio test, rejecting the null at the 0.01 level for both the market and category test. Given the differences observed across categories and markets, Hoch et al. (1995) suggest that it is possible to interpret the pooled results in the context of a random coefficients model, as we do here (see Hoch et al., 1995, p. 27). Specifically, if we denote the vector of parameters for category i as bi, then we can think of each of the i (i = 1, . . . , 135) category vectors, bi, distributed iid with mean E[bi ] = b, and Var( bi) = Vb . In order to characterize the central tendency among the categories, the objective is to make an inference about the mean of this random coefficient distribution (b). Table 5 presents the results of estimating Eq. (7) on the pooled data. Again, the model performed quite well. An

Table 5 Pooled expenditure equation results Ð Eq. (7) Dependent variable is DLNEXPENDITURE. Adjusted R2 = 0.282. All independent variables are expressed in difference form, as in Eq. (7). Variable

b

t-ratio

P-value

Demand-side factors BRPRICE PLPRICE BRPRICEREDN PLPRICEREDN BRFEATURE BRDISPLAY PLFEATURE PLDISPLAY POP INCOME HISPANIC AGE

0.1871 0.1028 0.0340 0.0111 0.0091 0.0024 0.0002 0.0003 0.3040 0.1620 0.0216 0.3313

2.03 3.36 22.26 1.42 12.46 23.18 4.56 9.53 3.15 3.48 6.36 4.20

0.043 0.001 0.000 0.156 0.000 0.000 0.000 0.000 0.002 0.001 0.000 0.000

Supply-side factors PLDISTN MCR4 NBRANDS SRATIO GROCCR4 PLUPERV BRUPERV Constant

0.0057 0.0082 0.1952 0.0004 0.0008 0.2802 0.0252 0.0094

28.82 25.51 11.71 0.547 2.26 5.59 0.030 ±

0.000 0.000 0.000 0.584 0.024 0.000 0.764 ±

W.P. Putsis Jr., R. Dhar / Journal of Business Research 52 (2001) 277±291

adjusted R2 of 0.282 is quite reasonable given that we are attempting to explain two levels of variation in cross-section (cross-category and cross-geographic market) using composite annual data, attempting to explain the variation in the changes of the variables from period to period (see, e.g., Kennedy, 1985, p. 26). 5.4. Overview of demand-side results Since all variables are expressed in either log or percentage form, the estimated coefficients represent expenditure elasticities Ð the percentage change in total category expenditure divided by a 1% change in each independent variable. In terms of managerially controllable variables, the results suggest that all forms of retail promotion and increased private label distribution can expand category expenditure. Consistent with our predictions, all four non-price promotions were estimated to have a positive and significant impact on category expenditure, supporting the notion that non-price promotions can increase total category expenditure. Also, as predicted, the results in Table 5 suggest that both display (0.0024 for branded versus 0.0003 for private labels) and feature (0.0091 for branded versus 0.0002 for private labels) effects were stronger for branded products (all presented comparisons are statistically different at a = 0.01.). Note, however, that the ``promotion expenditure elasticities'' are all quite small Ða 1% increase in feature or display activity resulted in considerably less than a 1% increase in category expenditure. This is consistent with the individual results presented above. Although these elasticities are small in magnitude, the aggregate expenditure effects can be quite large. In many of the larger categories, for example, a 10% increase in private label display activity implies a significant dollar increase in sales per category on an annual basis. With respect to price promotion, the estimated mean expenditure elasticities for branded permanent and temporary price reductions are 0.1871 and 0.0340, respectively, with both point estimates significant at a = 0.05. The analogous expenditure elasticities for private label permanent and temporary price reductions are 0.1028 and 0.0111, respectively, although only the regular price variable is statistically significant for private labels. Recall that the regular price and the price reduction variables should have opposite signs since the two price reduction variables are expressed as percent price reduction. Thus, these results suggest that permanent price reductions increase total category expenditure for both national brands and private labels as predicted, but only (temporary) price promotion by national brands increase category revenue. There is not a statistically significant relationship between private label price promotion and total category expenditure in the pooled sample. This highlights the care that must be taken when interpreting results from any cross-category analysis. The net

289

effect of a price change on category expenditure depends not only upon own and cross-price demand elasticities, but also upon price reaction functions between brands in the market. One brand's price cut affects not only its own revenue (through own-price effects) and the revenue of its rival (through cross-price effects), but has secondary effects on total category revenue due to the price response of rival firms. The category-by-category and market-by-market analysis presented above suggested that there are some categories and a number of markets for which private label promotion does indeed result in a significant increase in total category expenditure, but this does not show up in the pooled analysis. Further, there seems to be a fair amount of variation across categories (but not across markets) in the ability of price promotion to increase category expenditure. 5.5. Overview of supply-side results The supply side variables clearly play a significant role in category size. Variables capturing the manufacturer fourfirm concentration ratio, the number of brands competing in the market, and private label distribution all play significant roles in influencing total category expenditure. Each of these variables were of the predicted sign and highly significant. On the other hand, contrary to our predictions, local retail concentration had a negative estimated coefficient in the pooled sample (Table 5), which may be explained by the lower costs associated with distribution and purchasing that may be associated with higher local retail concentration (see Putsis, 1997). The pooled results on the supply side are broadly consistent with the individual category and market results Ð factors that enhance the ability of firms to raise price does indeed appear to increase overall category revenue. More specifically, we note that national brand price promotion has the most significant and consistent (cross-category and cross-market) impact on total category expenditure. If the goal is to increase category size, reducing national brand price is the easiest way to accomplish this objective. With respect to private label products, managers interested in expanding the category need look no further than increased distribution Ð a 1% increase in private label distribution increases overall category expenditure by 0.57%. 6. Discussion and managerial implications The objective of this paper was to conduct an exploratory empirical investigation into the structural factors that drive category expenditure, focusing on the impact of promotion. We suggested earlier that category expenditure may depend upon ability of the firms in the marketplace to influence demand by changing the price and promotion mix (demand side effects) and also the ability of the firms to raise market prices across geographic markets depending upon the nature of local competition (supply side effects).

290

W.P. Putsis Jr., R. Dhar / Journal of Business Research 52 (2001) 277±291

The results suggest that not only do both demand and supply side factors play a significant role in category expansion, but that there are significant differences across categories, across markets and across promotional instruments. Any general conclusion about the ability of promotion to generate sales response and produce additional category revenue must recognize that while such a statement will likely be right on average, but wrong for any individual market or category. Nonetheless, the most important implication is that if retail promotion can increase category expenditure, then sales promotion should no longer be viewed as a zero-sum game, with share as the barometer of a promotion's success. For managers of weaker brands that shy away from promotions since it is generally believed that they cannot steal share from stronger brands (Allenby and Rossi, 1991), it suggests that promotions still have the potential to be profitable. Overall, the following managerial implications can be derived from the discussion above. 1. Promotions can indeed increase category expenditure, although the impact of any promotion on expenditure will depend heavily upon the specific category, market and type of promotion. 2. Demand and supply side structural factors both play important roles in category expenditure. 3. The magnitude of the increase in category revenue varies significantly across markets and categories, but is relatively small on average. 4. Since demand elasticities and category characteristics are likely to influence the degree to which category expansion is feasible, the results suggest that brand managers should determine their appropriate response to competitor's marketing activities taking into account the effect on category size as well as market share. In categories where the category expenditure elasticities are high, competitive promotions by different players may not lead to ruinous competition but rather with an increase in demand and profitability. 5. Finally, it is generally assumed that retailers are more concerned with category volume and profit, whereas manufacturers are more concerned with market share. However, if retail promotions can have a significant impact on overall category revenue, retailers should consider favoring strategies that increase category size over those that simply increase share. 7. Conclusions, limitations and future research Any attempt to explain the growth of private labels in the US in recent years must be multi-faceted. The growing body of work addressing cross-sectional variation in private label growth (e.g., Sethuraman, 1992; Sethuraman and Mittelstaedt, 1992; Hoch and Banerji, 1993; Narasimhan and Wilcox, 1998) has enriched our understanding of why private labels flourish in some categories and under-perform in others.

In our study, we build on this research with a number of interesting findings. Both private label and branded promotions can indeed expand the ``size of the pie'' so that the nature of the private label Ð branded competition is not necessarily a zero-sum game. Both supply and demand side factors play a significant role in category development and expenditure. All forms of promotion can increase category expenditure, although there are significant differences across categories, markets and across promotional vehicles. Non-price promotion expands category expenditure for virtually every market and category, whereas the effectiveness of price promotion on category expenditure varies by market and category, as well as between private labels and national brands. Previous work on the effects of manufacturer concentration (e.g., Weiss, 1989) has been expanded to include the impact of local concentration. Finally, certain local market factors can also have important implications for total category expenditure. The variation observed across categories suggests that while a pooled analysis might provide estimates of the demand and reaction elasticities that are correct on average, they are likely to provide inaccurate estimates of the response for any specific category. Although a pooled analysis provides generalizability, the parameter estimates should be viewed as precisely that Ð general results that may not hold for specific categories. Understanding the interaction that occurs for any specific category requires intra-category analysis (Bresnahan, 1989). There are also a number of limitations of our analysis. For example, we note that the dependent variable used is composite in form (see, e.g., Farris et al., 1992) Ð EXPENDij depends upon both quantity and net price (see Eq. (7)). While this is not uncommon in a variety of settings in the marketing literature (e.g., market share and ROI analysis), and while expenditure functions have been analyzed frequently in economics (see Deaton and Muellbauer, 1980), it does present some difficult econometric issues. In the current applications, these are alleviated somewhat by the functional form specification and by the exact specification of the dependent variable. Further, in order to assess the impact of the use of a composite dependent variable, we performed some simple diagnostics (see Farris et al., 1992 for a detailed discussion), which suggested that its impact on the empirical results may not be a particular problem in this analysis. While not conclusive, given that our objective was to produce a descriptive model of the determinants of category expenditure, the results provide what we believe is a wellbalanced overview of the potential for category expansion across a wide set of markets and diverse categories. Since our analysis is intended to be exploratory and descriptive in nature, there are a number of important additional areas for future research. For example, the role of competitive interaction has been ignored above. Clearly, understanding how the pattern of interaction between firms impacts the ability of promotions to expand the category is

W.P. Putsis Jr., R. Dhar / Journal of Business Research 52 (2001) 277±291

important for assessing the broader managerial implications. Further, many of the cross-category differences that are observed are likely to be due to category-specific factors, such as the ability of consumers to stockpile the good, or how much of an impulse item the product is (Narasimhan et al., 1996). The analysis presented above is meant to be a starting point for understanding aggregate category effectÐ future research should attempt to understand why we observe differences across categories. This would be particularly important managerially. In addition, in our exploratory analysis, we took a very macro view. It would be interesting and worthwhile to distinguish between the impact of each variable on private label versus national brand expenditures. For example, additional insights may be gained by taking a more microapproach and estimating quantity and price equations for both private labels and national brands. These equations could then be aggregated to assess the impact on total expenditure. Differentiating between private labels and national brands in this fashion should provide additional insights into why these differences exist. Finally, research investigating the dynamic influence of competition between private label and national brands is needed. For instance, prolonged price promotion in a category may sensitize the market to price differences and help private labels in the long run. How this impacts the ability of promotions to expand the ``size of the pie'' is, at this point in time, an empirical question. We certainly encourage future research in these and related areas.

Acknowledgments The authors would like to thank the Food Marketing Policy Center at the University of Connecticut for use of the data. Support from the Yale School of Management Faculty Research Fund is gratefully acknowledged. Comments from Sridhar Balasubramanian, Subrata Sen, two anonymous reviewers and the associate editor improved earlier drafts substantially.

References Ailawadi K, Neslin SA. The effect of promotion on consumption: buying more and consuming it faster. J Mark Res (August)1998;35(3):390 ± 8. Allenby GM, Rossi PE. Quality perceptions and asymmetric switching between brands. Mark Sci 1991;10(Summer):185 ± 204. Berry S, Levinsohn J, Pakes A. Automobile prices in market equilibrium. Econometrica (July) 1995;63(4):841 ± 90. Blattberg RC, Buesing T, Peacock P, Sen S. Identifying the deal prone segment. J Mark Res (August) 1978;15(3):369 ± 77. Blattberg RC, Neslin SA. Sales Promotion: Concepts, Methods and Strategies. Englewood Cliffs, NJ: Prentice-Hall, 1990. Blattberg RC, Wisniewski KJ. Price induced patterns of competition. Mark Sci 1989;8(Fall):291 ± 309. Bresnahan TF. Empirical studies of industries with market power. In: Schmalensee R, Willig R, editors. Handbook of Industrial Organization New York: North-Holland, 1989 Chap. 17.

291

Chandon P, Wansink B. When and why does consumer stockpiling accelerate consumption volume? London Business School, 1999. (Unpublished manuscript). Cotterill RW. Market power in the retail food industry: evidence from Vermont. Rev Econ Stat 1986;68(August):379 ± 86. Cotterill RW, Putsis WP Jr, Dhar R. Assessing the competitive interaction between private labels and national brands. J Bus 2000;73(1):109 ± 37. Danaher PJ, Brodie RJ. Understanding the characteristics of price elasticities for frequently purchased package goods. University of Aukland, 1998. (Unpublished manuscript). Deaton A, Muellbauer J. Economic and Consumer Behavior. Cambridge: Cambridge University Press, 1980. Dekimpe MG, Hanssens DM, Silva-Risso JM. Long-run effects of price promotions in scanner markets. J Econom 1999;89(1 ± 2):269 ± 91. Farris PW, Parry ME, Ailawadi KL. Structural analysis of models with composite dependent variables. Mark Sci (Winter)1992;11(10):76 ± 94. Hausman JA, Taylor WE. Panel data and unobservable individual effects. Econometrica 1981;49:1377 ± 98. Hoch SJ, Banerji S. When do private labels succeed? Sloan Manage (Summer)1993;57 ± 67. Hoch SJ, Dhar SK. Why store brand penetration varies by retailer. Mark Sci 1997;16(3):208 ± 27. Hoch SJ, Kim B-D, Montgomery AL, Rossi PE. Determinants of store-level price elasticity. J Mark Res 1995;22(February):17 ± 29. Keane MP. Modeling heterogeneity and state dependence in consumer choice behavior. J Bus Econ Stat (July)1997;15(3):310 ± 27. Kelton C, Weiss L. Changes in concentration, change in cost, change in demand and change in price. In: Weiss L, editor. Concentration and Price. Cambridge, MA: MIT Press, 1989, pp. 41 ± 66. Kennedy P. A Guide to Econometrics (2nd edn.). Cambridge, MA: MIT Press, 1985. Marion B. The Food Retailing Industry. New York: Praeger, 1979. Narasimhan C, Neslin SA, Sen SK. Promotional elasticities and category characteristics. J Mark (April)1996;60(2):17 ± 30. Narasimhan C, Wilcox RT. Private-labels and the channel relationship: a cross-category analysis. J Bus (October)1998;7(4):573 ± 600. Putsis WP Jr. An empirical study of the effect of brand proliferation on private label Ð national brand pricing behavior. Rev Ind Organ (June) 1997;12(3):355 ± 71. Putsis WP Jr, Dhar R. The many faces of competition. Mark Lett (July) 1998;9(3):269 ± 84. Putsis WP Jr, Dhar R. Category Expenditure, promotion and competitive market interactions: can promotions really expand the Pie? London Business School, 1999. (Unpublished working paper). Quelch JA, Harding D. Brand versus private labels: fighting to win. Harv Bus (January ± February)1996;99 ± 109. Raju JS. The effect of price promotions on variability in product category sales. Mark Sci (Summer)1992;11(3):207 ± 20. Schmalensee R. Entry deterrence in the ready-to-eat breakfast cereal industry. Bell J Econ 1978;9:305 ± 27. Sethuraman R. The effect of marketplace factors on private label penetration in grocery products. Marketing Science Institute, October 1992. (Working Paper # 92-128). Sethuraman R. A meta-analysis of national brand and store brand cross promotional price elasticities. Mark Lett 1995;6(4):275 ± 86. Sethuraman R, Mittelstaedt J. Coupons and private labels: a cross-category analysis of grocery products. Psychol Mark (November/December)1992;9(6):487 ± 500. Slade ME. Product rivalry with multiple strategic weapons: an analysis of price and advertising competition. J Econ Manage Strategy (Fall)1995;4(3):445 ± 76. Tellis GJ. The price elasticity of selective demand: a meta analysis of econometric models of sales. J Mark Res 1988;25(November):331 ± 41. Weiss LW, editor. Concentration and Price. Cambridge, MA: MIT Press, 1989. Zenor MJ. The profit benefits of category management. J Mark Res (May)1994;31(2):202 ± 13.