An Investigation into the Evolution of Marketing-Mix Effectiveness: An Empirics-First Approach

Abstract

This study investigates how the effectiveness of three core marketing-mix instruments—price, assortment, and distribution—evolves over time across brands, categories, and countries, using the empirics-first approach to generate relevant knowledge. The authors analyze household panel data covering over 16,000 brands in 85 consumer packaged goods categories for on average ten years across 34 countries in Asia, Europe, North America and South America. They adopt a rolling-window estimation approach to derive time-varying elasticities, followed by a moderator analysis to identify systematic drivers of change. Averaged across time, brand price elasticity is −.640, assortment elasticity is .311, and distribution elasticity is .213. However, time-invariant averages have limited meaning given that 98.5% of brands exhibit significant evolution in at least one instrument's elasticity. Substantial heterogeneity in trends emerges, moderated by brand, category, and country factors. The single most important predictor is the brand's baseline (mean) elasticity. The stronger the elasticity, the larger its downward trend. Brand power and whether the brand is a local brand emerge as other key factors. Consumer purchase frequency and share of wallet are stronger category predictors of elasticity trajectories than retailer factors. Findings underscore the need for dynamic, elasticity-based resource allocation in brand management.

A core responsibility of brand managers is to allocate scarce marketing resources to maximize firm performance (Sridhar et al. 2011). The guiding principle is straightforward: Marketing inputs with higher elasticities merit greater investment, all else being equal (Mantrala 2002; Saboo, Kumar, and Park 2016). This allocation is inherently dynamic, as the effectiveness of individual marketing instruments changes over time (Leeflang et al. 2009; Raman et al. 2012). Quantifying and understanding these time-varying effects is of paramount importance in an era of global change. The rapid spread of new technologies and the rise of e-commerce across the world (Rust 2020); rising incomes in emerging markets; changes in consumer preferences, economic conditions, and the global competitive landscape (Saboo, Kumar, and Park 2016); and the entry of no-frills retailers and private labels (Steenkamp, Benedict, and Geyskens 2014) all are radically transforming the business environment. In the context of marketplace shifts, general insights on how marketing effectiveness develops across brands, categories, and countries are immediately valuable to brand managers (Gordon, Goldfarb, and Li 2013). This article aims to provide such insights for three key marketing tools—price, assortment, and distribution.

While prior research has empirically examined the temporal dynamics of marketing-mix effectiveness, significant gaps remain. First, research on the factors influencing the evolution of marketing-mix effectiveness (discussed below) has almost exclusively focused on price. Second, even for price, our understanding of what drives variation in its effectiveness is limited. Several brand factors have been studied, but the empirical settings often involve a relatively small number of brands and product categories. As Hanssens (2015) points out, results in one product category may not apply to others, even within the same industry. Except for the product life cycle, little is known about how category characteristics affect these dynamics. Third, cross-national differences in how marketing elasticities evolve remain virtually unexplored. Empirical insights into cross-national variation and the role of country-level factors are relevant for international brand managers (Steenkamp 2017). This study addresses these three gaps by providing answers to the following two research questions: (1) How frequently, and in what directions, do price, assortment, and distribution elasticities evolve over time? and (2) Which brand-, category-, and country-level characteristics moderate systematic heterogeneity in these trajectories?

To this end, we adopt the empirics-first (EF) approach. Golder et al. (2023) have developed a set of criteria and associated questions to assess the suitability of the EF approach in a specific research context. In Table 1, we show that many of these questions can be answered affirmatively in our study. Importantly, the EF approach enables us to leverage the breadth of our cross-national, multicategory dataset to identify patterns and generate insights, while considering existing literature where relevant. Specifically, our findings are based on 16,000+ brands across 85 consumer packaged goods (CPG) categories, covering an average of ten years of household panel purchase data from 34 countries in Asia, Europe, North America, and South America.

Table 1.

Rationale for Adopting an Empirics-First Approach.

Aspect/Questions		Our Study
Real-World Relevance
	Is the phenomenon a determinant of enduring stakeholder significance?	The evolution in the effectiveness of marketing instruments is key to managerial decision-making.
	Does the research use new data on an important phenomenon?	Yes, the long time series (on average ten years) and the international coverage (34 countries) are new and allow for generalizations.
Literature
	Is the current literature thin, conflicting, unintuitive, and/or far afield from marketing?	Yes, few moderators of the evolution of price, and even more so for assortment and distribution have been studied (Table 2).
	Has the literature been consulted for potential inspiration, variables/factors to consider, and interpretation of the empirics?	Yes, where possible, the literature has been used.
Research Process
	Is the research agenda-free (i.e., does it begin and proceed without fixed ideas about its outcomes)?	Yes, we did not have fixed ideas about the outcome, even about the amount of change per se.
	Has the investigation incorporated hunches based on marketing expertise and experience?	Where appropriate, plausible hunches are offered.
	Has research been deepened with additional variables?	We purposefully cast a wide net by including 28 potential moderators of the evolution of marketing elasticities
	Has research been broadened (e.g., additional industries, organizations, categories, products)?	Broadened by examining 85 categories in 34 countries
Research Outcomes
	Is the phenomenon better understood empirically, conceptually, and/or theoretically?	The moderator analysis (Table 7) adds understanding to an underexplored issue of drivers of change in marketing effectiveness
	Is there advice for marketing stakeholders based on causal effects?	Yes, in the “Managerial Implications” section, we show how the optimal allocation of marketing efforts across instruments or countries changes over time, using our empirical findings.
	Has the research discovered empirical regularities?	Yes, these are summarized in the final section.
	Have the effect sizes received proper attention?	Yes, see Tables 7 and 8.
Robustness and Generalizability
	Was there a real possibility for falsifying the findings?	Yes, this is exemplified by a number of weak/null results for positioning and retail category factors
	Have both simpler models and more sophisticated models yielded consistent results?	Extended models (adding advertising and coupon usage in the U.S.) and more sophisticated models (Kalman filter model) yielded essentially the same estimates
	Are findings generalizable to other contexts, and could they potentially spark follow-up research?	Given the global scope and broad coverage, results from this study are likely to be generalizable to other contexts.
Presentation
	Can the results be incorporated into future meta-analyses?	Yes, the number of elasticities reported in this paper are 8–10 times larger than that of any previous primary study or meta-analysis
Theory
	Have the following questions been considered? Does the article offer plausible explanation(s) for the findings? Have multiple potential explanations been considered and evaluated? Have potential new theoretical relationships been proposed? Have conflicting theories been resolved?	1–3. Where applicable (e.g., for positioning, international brands, e-commerce, category concentration, market size), multiple plausible explanations have been offered and empirically evaluated. 3. We have proposed new theoretical relations (e.g., for own-instrument effectiveness and category buyer behavior). 4. Our empirical findings resolve certain conflicting explanations (e.g., for product life cycle, market size).
	Have new frameworks been proposed?	Our research framework (Figure 1) is new.

Notes: The “Aspects/Questions” column is taken from Golder et al. (2023, Table 4). We only include the questions that are features of our research. As stated by Golder et al. (p. 330): “It is unreasonable to expect EF articles to answer all questions in the affirmative.”

Our study follows a meta-analytic design, in line with McShane et al. (2024, Figure 3). We treat the estimated marketing effectiveness trajectories of thousands of brands as repeated observations of the same underlying phenomenon. This allows us to examine how these trajectories differ across brands and to assess which brand, category, and country characteristics moderate that variation (McShane et al. 2024, p. 5). Unlike traditional meta-analyses that combine results from different studies, we apply a consistent estimation approach across all brands. Beyond documenting broad empirical generalizations, revisiting the evolution of elasticities is particularly important because prior work (e.g., Datta et al. 2022) has reported limited temporal change, raising the question of whether meaningful dynamics exist at scale. Our unique multicountry, multicategory database enables us to detect subtle yet systematic shifts that smaller datasets cannot reveal.

Prior Research on Dynamic Marketing-Mix Effectiveness

Prior research on the time-varying effectiveness of price, assortment, and distribution has followed two main paths: analytical and empirical. Our study is empirical, but we briefly review relevant analytical work, as it underscores the value of understanding these dynamics for optimal resource allocation (Mantrala 2002; Sridhar et al. 2011). Dorfman and Steiner (1954) developed a normative model for allocating marketing expenditures based on market response, assuming constant elasticities and treating marketing costs as fixed costs. They demonstrated that a profit-maximizing firm should align distribution and assortment budgets with the revenue-productivity weighted elasticities of each instrument. Ingene and Parry (1995) extended this model to multinational firms balancing global standardization and local adaptation. Working within a static framework, Fischer et al. (2011) show that an internationally operating profit-maximizing firm should allocate budgets for any marketing instrument across countries based on the country-specific elasticity, as well as the country-specific profit contribution, sales base, and expected sales growth. Other models incorporated dynamics, competition, and uncertainty into allocation decisions (Mantrala 2002).

A core insight from normative models is that, all else equal, more budget should be allocated to marketing instruments with higher sales elasticities. However, most models assume constant effectiveness over time. Raman et al. (2012) extend the normative literature by proposing a profit-maximizing model for a monopolist that accounts for time-varying effectiveness and costs. They show that the optimal allocation ratio between two marketing instruments is proportional to their time-varying effectiveness in generating sales and inversely proportional to their time-varying costs. When elasticities vary over time, the allocation should also vary, highlighting the need for empirical insights into temporal changes in marketing effectiveness (Leeflang et al. 2009).

A second stream of research has focused on estimating the time-varying nature of price, assortment, and distribution elasticities. Table 2 provides an overview of key empirical studies.

Table 2.

Previous Empirical Research on the Evolution in Price, Assortment, and Distribution Elasticities.

	Time-Varying Marketing Effectiveness			Moderators of Evolution in Marketing Effectiveness			Data			Key Findings with Respect to Evolution
Study	Price	Assortment	Distribution	Brand Factors	Category Factors	Country Factors	No. Brands	No. Categories	Countries
Simon (1979)	X			Stage of brand life cycle			35	3(pharma, CPG)	Germany	The evolution of price sensitivity varies across brands. Price sensitivity decreases considerably from the introduction and growth stages to the maturity stage and increases during the decline stage
Tellis (1988)	X				Stage of product life cycle		367	Unknown(CPG, durables, pharma, others)	U.S., Europe, Australia, New Zealand	Meta-analysis: Price sensitivity is higher in the mature/decline stage than in the introduction/growth stage
Parker (1992)	X				Stage of the product life cycle			17(durables)	U.S.	No consistent pattern in the effect of stages of the product life cycle on price sensitivity
Mela, Gupta, and Lehmann (1997)	X			Advertising, price and nonprice promotion, brand loyalty		Recession	9	1(CPG)	U.S.	Price sensitivity increases over time but the trend is stronger for the nonloyal than for the loyal segment; advertising reduces the price sensitivity of the nonloyal segment and price promotions that of both segments; nonprice promotions decrease (increases) the price sensitivity of the loyal (nonloyal) segment; recession increases price sensitivity of loyal consumers
Jedidi, Mela, and Gupta (1999)	X			Advertising, price promotion, brand loyalty			4	1(CPG)	U.S.	Price promotions increase over-time price sensitivity; advertising and brand loyalty have no effect
Krishnamurthi and Papatla (2003)	X			Brand loyalty			16	4(CPG)	U.S.	Increases in brand loyalty reduce price sensitivity over time; the nature of the effect (concave vs. S-shaped) varies across categories and consumer segments
Bijmolt, Van Heerde, and Pieters (2005)	X				Stage of product life cycle		1,851	Unknown(CPG, durables)	U.S., Europe, Australia/ New Zealand	Meta-analysis: For relative (absolute) sales, price sensitivity does not change (increases); price sensitivity is lower in the mature/decline stage than in the introduction/growth stage
Ataman, Van Heerde, and Mela (2010)	X			Assortment, advertising, discounting, distribution			70	25(CPG)	France	Assortment and advertising (discounting) decrease (increases) price sensitivity
Gordon, Goldfarb, and Li (2013)	X				Average price elasticity, share of wallet; perishability, market concentration	GDP growth	121	19(CPG)	U.S.	On average, price sensitivity declines over time, but there is a large variation in the magnitude and direction of the trend across product categories. Price sensitivity in highly elastic categories moves with GDP growth whereas less elastic categories display the opposite pattern; share of wallet might explain this
Van Heerde et al. (2013)	X			Advertising		Business cycle	150	36(CPG)	U.K.	Price sensitivity increases over time and this trend is amplified in times of economic contraction
Datta et al. (2022)	X	X	X				1,600+	14(durables)	14 Indo-Pacific Rim countries	None of the marketing-mix instruments’ elasticities changed over time in a split sample (early vs. late) analysis
Sethuraman, Gázquez-Abad, and Martínez-López (2022)		X					1,996	Unknown(CPG, durables, service, other)	UnknownU.S., Europe, others	Meta-analysis; results refer to retail assortment elasticities, not brand assortment elasticities; assortment elasticity increases over time
This article	X	X	X	10 factors	11 factors	7 factors	16,000+	85(CPG)	34countries in Americas, Asia, Europe	Price (assortment and distribution) sensitivity increases (decrease) over time. Brand, category, and country factors have a substantial effect on this heterogeneity

Price has received the most attention by far, but the evidence is mixed. As for absolute sales, Bijmolt, Van Heerde, and Pieters (2005) and Van Heerde et al. (2013) report that price sensitivity increases over time, with the magnitude of annual change averaging −.04 to −.06. ¹ On the other hand, Datta et al. (2022) find no evidence for the time-varying effectiveness of price. For market share or choice, Bijmolt, Van Heerde, and Pieters (2005) report no change in price sensitivity, while Gordon, Goldfarb, and Li (2013) find that it declined on average by .035 per year. The latter study also reveals substantial variation across categories (e.g., price sensitivity fell by .16 per year for toilet tissue but rose by .30 for mayonnaise ² ). These mixed and sometimes contradictory findings underscore that existing studies, while valuable, cannot fully resolve whether temporal change is category-specific noise or reflects deeper structural patterns across markets. Our empirical generalizations build directly on this foundation by showing, at an unprecedented level of granularity, how cross-category and cross-country heterogeneity produce systematic patterns that prior studies could not detect. The simulation analyses introduced in the final section of the article use the analytical results of Fischer et al. (2011) and Raman et al. (2012) to translate these empirical patterns into actionable implications, illustrating how dynamic elasticities meaningfully alter optimal marketing-mix allocation.

Table 3.

Within-Variation Marketing-Mix Instruments.

Mix Variable	SD p25	SD p50	SD p75	CV p25	CV p50	CV p75
Price	.031	.149	1.080	.133	.205	.343
Assortment	1.017	2.030	4.035	.196	.293	.396
Distribution	.038	.058	.087	.206	.326	.509

Notes: Each row summarizes within-unit variability across brand–category–country units (i.e., brands). For each brand, we calculated its time-series standard deviation (SD), coefficient of variation (CV = SD divided by the unit mean), and median level. Percentiles (p10, p50, p90) report the distribution of SDs and CVs across all units, with the median (p50) reflecting the typical unit's variability.

Among the factors that influence changes in price elasticity, the product life cycle stage has attracted the most attention. On the one hand, as a category matures, consumers generally gain more knowledge about availability, prices, deals, and promotions, which can increase their price sensitivity. Additionally, early adopters tend to be less sensitive to price than later adopters, indicating that price elasticity is higher during the mature or decline stages than in the introduction or growth phases. On the other hand, strongly growing categories attract more new buyers with no established relationship or trust with a particular brand within the category. This means they are more open to switching if they find a better price. The meta-analytical results of Tellis (1988) indicate the former: Price sensitivity is higher in categories in the mature stage of the product life cycle. However, the (larger) meta-analysis of Bijmolt, Van Heerde, and Pieters (2005) finds the opposite.

Heavier brand advertising and broader assortments decrease price sensitivity over time, while frequent price promotions boost sensitivity to the regular (nonpromotional) price (Ataman, Van Heerde, and Mela 2010). Several studies suggest that the effects of these factors vary with consumer loyalty, although the results are again mixed. Krishnamurthi and Papatla (2003) find that brand loyalty reduces price sensitivity over time, whereas Jedidi, Mela, and Gupta (1999) find no such effect. Mela, Gupta, and Lehmann (1997) report that price sensitivity increases in both loyal and nonloyal segments, but more so among nonloyal consumers.

The state of the economy is the only macroeconomic factor that has received attention. Overall, price sensitivity rises when the economy is weak and decreases when it is strong, although there is variation across consumer segments (Mela, Gupta, and Lehmann 1997) and categories in function of category-level price elasticity and share of wallet (Gordon, Goldfarb, and Li 2013).

Compared with the extensive empirical attention devoted to temporal changes in price elasticity, the dynamics of assortment and distribution elasticities have received far less scrutiny (Table 2). Datta et al. (2022) find no evidence that the effectiveness of either marketing-mix instrument changed over time. Sethuraman, Gázquez-Abad, and Martínez-López (2022) focus on retail assortment size (as opposed to brand assortment) and consider both sales and attitudinal or behavioral outcomes. They find an increase in assortment elasticity of approximately .009 per year. ³

Study Framework

Our interest in the evolution of elasticities is motivated by unprecedented global changes. Overlapping technological, socioeconomic, and marketplace trends drive these shifts. While we do not aim to quantify the effects of these macro trends, we use them to contextualize changes in marketing-mix effectiveness. Specifically, these trends influence a range of brand-, category-, and country-level factors that shape the evolution of price, assortment, and distribution elasticities. Figure 1 presents a schematic overview of the conceptual framework guiding our empirical research.

Figure 1.

Research Framework.

Technological advances—such as digitization, big data, artificial intelligence (including generative AI), and the widespread adoption of mobile technology—are reshaping how firms engage consumers and manage their marketing mix (Rust 2020; Wichmann et al. 2022). One of the farthest-reaching consequences has been the rise in e-commerce, which has transformed pricing, assortment, and distribution dynamics across categories and markets.

At the same time, global socioeconomic conditions have shifted markedly. Rising incomes have brought many into the modern marketplace, especially in emerging economies that now serve as key consumer markets, production hubs, and laboratories for new brands. This growth has been accompanied by rising inequality (Rust 2020), creating a widening divide between affluent consumers and those with constrained means. Another notable economic trend is the return of inflation (Dekimpe and Van Heerde 2023), further affecting consumer sensitivity to price and access. Overlaying these trends is a broader transformation of the marketplace: Globalization expanded dramatically, with world trade rising from 50% to 60% of global GDP from 2002 to 2012. ⁴ In the mid-2010s, we witnessed a backlash, and globalization has stalled (Hu and Spence 2017).

These technological and socioeconomic trends reinforce and are reinforced by marketplace trends. Brand commoditization is said to be on the rise as technology spreads around the world, and some brands shift resources from long-term brand investments to activities to increase sales in the short term (Lodish and Mela 2007). This opens opportunities for no-frills private labels and hard discounters. At the same time, brands that continue to invest will be able to differentiate themselves from their competitors. Slowing economic growth in some parts of the world and the introduction of new brands, including from emerging markets (Kumar and Steenkamp 2013), mean that competitive intensity in many categories is increasing (Porter 1980). While the geographical reach of international brands is wider than ever before, the globalization backlash offers renewed opportunities for local brands (Steenkamp 2019).

Building on this macro-level backdrop, we now turn to a discussion of how these broader shifts manifest in brand-, category-, and country-level factors that, in turn, help explain the evolution in price, assortment, and distribution elasticities. We group these determinants into three domains—brand, category, and country. Consistent with the EF approach, we do not offer formal hypotheses. Rather, we succinctly review pertinent literature, supplemented by informed intuition, where appropriate. This practice is consistent with recommendations by Golder et al. (2023, p. 325).

Brand Factors

Brand-level factors capture how well individual brands are placed to respond to an evolving marketplace. Specifically, we consider three dimensions—brand positioning, brand reach, and brand clout—which may differentially shape a brand's ability to adapt to shifting consumer expectations and competitive pressures.

Brands with a distinct positioning carve out a clear and recognizable place in the changing market, making it easier for consumers to understand what the brand stands for and what to expect from it. This clarity can reduce price sensitivity over time (Krishnamurthi and Papatla 2003) and could arguably enhance the effectiveness of assortment and distribution efforts. However, a clearly positioned brand may become stuck in a niche or be less nimble in adjusting to a changing marketplace, leading to a decline in assortment and distribution effectiveness.

Brand reach across categories reflects decisions about brand architecture—specifically, the extent to which a brand offers purchase options across multiple categories. Umbrella branding can signal quality to consumers (Erdem 1998) and, given consumers’ strong preference for high quality (Zeithaml 1988), broader category presence enhances the effectiveness of marketing activities. However, brands that are present in more categories may lose distinctiveness relative to category-specific competitors, which could reduce marketing effectiveness over time.

We also consider the brand's geographical reach. Some brands are local, while others are sold in many countries. International brands have become symbols of global consumer culture, and with rising cultural globalization, their marketing efforts may elicit stronger consumer responses (Holt, Quelch, and Taylor 2004). However, international brands often sacrifice local responsiveness for global efficiency, weakening marketing-mix effectiveness in any given country (Steenkamp 2017). At the same time, consumers are rediscovering local brands, drawn by their authenticity and ties to local consumer culture (Steenkamp 2019). As the quality and assortment of local brands improve, consumer responsiveness to their marketing efforts may rise.

We consider four factors that reflect a brand's clout (Steenkamp, Benedict, and Geyskens 2014; Van Ewijk, Gijsbrechts, and Steenkamp 2022): market share, share growth, own marketing effectiveness, and competitive vulnerability. High-share brands may experience a decline in assortment and distribution effectiveness, having already achieved broad market coverage. Their market power also allows them to price toward the less elastic portion of demand (Gordon, Goldfarb, and Li 2013), resulting in reduced price sensitivity. Strongly growing brands are often in the brand growth stage of the brand life cycle (Simon 1979), which intuition would suggest is associated with heightened marketing effectiveness. Simon (1979) provides supporting evidence for price.

Brands typically differ in how effectively they deploy marketing instruments (Datta et al. 2022). But how does a brand's baseline effectiveness in wielding a particular instrument influence how that instrument's effectiveness evolves over time? If a brand is highly effective in using a particular instrument, there may be little room to grow, while there is ample room to decline in effectiveness. In market equilibrium, the effectiveness of incremental marketing efforts could be similar across brands. This suggests that high marketing effectiveness is associated with a downward trend in effectiveness. Some indirect evidence is provided by Sharp (2010), who documents a regression to the mean phenomenon in buyer behavior.

Finally, how does a brand's vulnerability to competitive moves affect the effectiveness of its own marketing instruments over time? If a brand is highly vulnerable to, say, competitive moves with a particular instrument, it may respond by intensifying its own use of that instrument. This can easily lead to a tit-for-tat dynamic where brands match each other's marketing actions, which will reduce the instrument's impact (Lal and Villas-Boas 1998).

Category Factors

Category-level factors reflect how broader marketplace shifts reshape competitive intensity and the perceived value of categories. We examine dimensions linked to commoditization and competitive threat, including e-commerce penetration, the presence of hard discounters and private labels, category concentration, promotional pressure, new product activity, and category growth (as proxy for the product life cycle). We also consider consumer behavior by examining category purchase frequency, penetration, and share of wallet.

E-commerce can commoditize brands, but it also offers opportunities for differentiation (Gielens and Steenkamp 2019). Thus, its effect on the trajectory of marketing instruments’ effectiveness is unclear. Categories characterized by a strong presence of hard discounters and private labels tend to experience intensified price competition, and constrained brand distribution and assortment expansion (Steenkamp and Sloot 2019). Accordingly, such categories might exhibit rising price sensitivity alongside declining assortment and distribution effectiveness.

Over time, sustained promotional pressure can contribute to category commoditization by training consumers to focus on price (Lodish and Mela 2007), leading to increased price sensitivity. At the same time, frequent promotions may boost category engagement, potentially enhancing the effectiveness of nonprice instruments. In contrast, ongoing innovation activity can counteract commoditization by shifting consumer focus toward differentiated benefits, thereby reducing reliance on price.

In concentrated brand markets, reduced competition and search costs (Pagoulatos and Sorensen 1986) can increase marketing effectiveness over time. Yet marketing effectiveness can decrease with concentration as fewer and more dominant brands tend to operate closer to the less elastic portion of the demand curve. Some tentative support can be found in Gordon, Goldfarb, and Li (2013), revealing a small correlation of −.11 between brand concentration and the evolution of price sensitivity. ⁵ We consider both retail and brand concentration.

Brands in categories that are in the growth stage of the product life cycle may temporarily experience a softening of competitive pressure as new buyers enter and total category demand expands. This growth opens the door for broader distribution and richer assortments aimed at both new and existing customers. Following Bijmolt, Van Heerde, and Pieters (2005) we might further expect that price sensitivity is higher in strongly growing categories.

Turning to the buyer side of the category, Gordon, Goldfarb, and Li's (2013) findings suggest that price sensitivity increases in categories that account for a higher share of wallet. The impact on the evolution in distribution and assortment elasticities is, however, unclear. In categories characterized by a high purchase frequency and a high level of penetration, consumer knowledge of prices and products is typically greater, which can stimulate price and assortment effectiveness.

Country Factors

Just as technological, socioeconomic, and competitive shifts reshape brands and categories, they also interact with underlying economic and cultural conditions to influence how marketing effectiveness evolves. Larger populations support broader product variety, resulting in more crowded product spaces and heightened price sensitivity (Desmet and Parente 2010). At the same time, the need to serve diverse preferences of a large population increases the relevance—and effectiveness—of assortment and distribution, making consumers more responsive to nonprice instruments. However, crowded product spaces can lead to consumer overload and reduced marketing effectiveness (Schwartz 2004). While one would expect that lower GDP per capita and higher GDP growth are associated with higher price sensitivity, empirical evidence reveals the opposite (Bahadir, Bharadwaj, and Srivastava 2015; Datta et al. 2022). In countries with high income inequality, many consumers remain marginal participants in the formal economy, reinforcing price sensitivity while limiting the effectiveness of assortment and distribution as differentiation tools (Kozlenkova et al. 2021).

Beyond economic conditions, cultural values also shape how marketing effectiveness evolves over time. In high-power-distance cultures, where prestige and wealth signal social position, consumers may become less price sensitive as higher prices convey status (Datta et al. 2022). In cultures with strong uncertainty avoidance, consumers tend to resist change and favor familiar choices. This can lead to declining responsiveness to new product introductions—lowering assortment elasticities—while increasing reliance on price and distribution as signals of quality and reliability (Steenkamp and Gielens 2003; Zeithaml 1988). Finally, in cultures that emphasize achievement and success (formerly labeled “masculinity”), price sensitivity may also decline, as the ability to afford higher-priced products is itself seen as a marker of success.

Data and Measures

Sample Composition

The global market research agencies Kantar, GfK, and IRI provided household panel data for 34 countries, tracking purchases in the CPG grocery sector. These countries span multiple regions, South/East Asia (China, India, Indonesia, Thailand, Taiwan, and Vietnam), the Middle East (Turkey), South America (Argentina, Bolivia, Brazil, Chile, Colombia, Ecuador, and Peru), North America (the Central American region, ⁶ Mexico, and the United States), Eastern Europe (Czech Republic, Hungary, Poland, Romania, Russia, and Slovakia), and Western Europe (Austria, Belgium, Germany, Denmark, Spain, France, Ireland, Italy, the Netherlands, Sweden, and the United Kingdom). This diverse sample allows us not only to monitor well-established Western markets but also to gain valuable insights into less-studied non-Western markets. Furthermore, the sample includes a wide range of economies. Web Appendix A provides an overview of the countries and data availability. Together, these countries are home to 4.5 billion people, representing more than 60% of the global CPG market.

On average, the panels encompass 10 years of data up to 2020, from a minimum of 5 years for Ireland and Vietnam to a maximum of 16 years for Germany and the United Kingdom. The data were provided at the individual consumer level, including stockkeeping unit (SKU) details and transaction dates. Since our focus is on analyzing brand performance across countries and product categories, the appropriate unit of analysis is the brand level. Therefore, we aggregated the data to the weekly brand level within each category and country. Throughout this discussion, any reference to “brands” pertains to brand-category-country combinations.

The dataset covers a wide range of CPGs, including food, beverages, household, and personal care products. Data are available for 85 product categories in total. These categories are comparable across countries, making them suitable for cross-country analysis. ⁷ However, not every category is tracked in each country. We included only those categories where at least two brands were consistently available in the market. This resulted in 63 categories per country on average for analysis. Web Appendix B provides an overview of the categories and data availability.

Within each available product category and country combination, we track national brands that have consistent data over time and represent a significant share of the market. To ensure a feasible set of brands for estimation, while also guaranteeing enough weekly observations with nonzero transactions and complete marketing-mix data, we select brands that account for at least 1% of the market, based on either volume or unit sales (see Datta et al. [2022] for a similar approach).

Additionally, brands must be available in the market with no more than a six-month interruption and must have been present in the market for at least three years. Combined, these criteria result in the selection of 16,642 brands. ⁸ On average, there are seven brands per category–country combination. The brands we retain exhibit a wide range of positioning and branding strategies. Some brands are present in almost every country (e.g., Coca-Cola, Gillette, Nescafé, Palmolive), others are available in a limited set of countries (e.g., Douwe Egberts in coffee, Tempo in paper towels), while others are local, being only available in a single country.

Variable Operationalization

Variables used to obtain time-varying elasticities

We use volume market share as our dependent variable, allowing for easy comparison across product categories and countries of varying sizes. Price is calculated by dividing revenue sales by volume (or unit) sales, resulting in a weighted average price that accounts for variations in SKU popularity within brands. This price variable reflects the actual shelf price, including any price discounts. Prior empirical research on price elasticities is overwhelmingly based on the actual price paid (Bijmolt, Van Heerde, and Pieters 2005). Assortment represents the brand's product range and is measured by the number of SKUs available in the market. To measure distribution, we assess all retailers in the dataset to determine if the brand is sold by each retailer on a monthly basis. For brands sold by a retailer, we consider the proportion of all the brand's SKUs carried by that retailer. We then calculate the brand's overall distribution as the weighted average of these proportions across all retailers, using the retailer's total grocery market share from the previous month as the weighting factor. ⁹

Table 3 shows meaningful within-brand variation for all three instruments. Median standard deviations indicate nontrivial temporal fluctuations in price, assortment, and distribution. The corresponding coefficients of variation confirm these fluctuations are substantial relative to unit means, ensuring sufficient within-brand variation for analysis. We compile competitive metrics for price, assortment, and distribution by averaging the data series for all competing brands within the category that hold at least a 1% market share.

To address missing observations, we linearly interpolate up to two weeks of missing intermediate data points in any given series. If a brand has more than two consecutive missing observations—often occurring with minor brands that may have low sales and can temporarily “disappear”—we select the time periods where the most consistent observations for the focal variables are available (see also Datta et al. 2022).

Variables used to assess heterogeneity in trends of time-varying elasticities

Table 4 details the operationalization of the variables used in this stage. Summary statistics and bivariate correlations are presented in Web Appendix C.

Table 4.

Operationalization of Variables in Step 3.

Factors	Definition	Source
Brand Factors
Price positioning (PricePosc,p,b)	Relative price of brand b in product category p in country c in relation to average price in product category p in country c, averaged across time.	Panel data
Assortment positioning (AssortPosc,p,b)	Relative number of SKUs offered by brand b in product category p in country c in relation to the average number offered in product category p in country c, averaged across time.	Panel data
Distribution positioning (DistriPosc,p,b)	Relative retailer presence for brand b in product category p in country c in relation to the average retailer presence in product category p in country c, averaged across time. Retailer presence is captured as the weighted presence at top retailers in country.	Panel data
Umbrella branding (CatReachc,p,b)	Number of product categories in which brand b is present in country c.	Panel data
International spread (IntReachc,p,b)	The number of countries in which brand b is present relative to total countries in the study for that particular product category; zero in case of local.	Panel data
Local (Localc,p,b)	Dummy variable that equals one if a brand is local in country c and not available elsewhere.	Panel data
Market power (Powerc,p,b)	Market share of brand b in product category p in country c, averaged over time.	Panel data
Brand growth (BrandGrowthc,p,b)	The evolution in the sales of brand b in product category p in country c. The evolution is the difference between the first and last year in the observation window.	Panel data
Category Factors
E-commerce share (Ecommc,p)	The percentage of sales in country c and product category p obtained through online channels over the observation window.	Euromonitor
Hard discounter share (HardDisc,p)	The percentage of sales in country c and product category p obtained through hard discounters over the observation window.	Euromonitor
Private label share (PLSharec,p)	The percentage of sales in country c and product category p obtained through private labels over the observation window.	Panel data
Retail concentration (RetailConcc,p)	The sum of the shares of the top three retailers in product category p in country c, averaged over the observation window.	Panel data
Brand concentration (BrandConcc,p)	The sum of the shares of the top three brands in product category p in country c, averaged over the observation window.	Panel data
Category promotion pressure (CatPromoc,p)	Total absolute volume sales sold on promotion for the total category, estimated based on promo algorithm, per month, averaged over the observation window.	Kantar, GfK
Category new product activity (CatInnoc,p)	Total number of new products for the top ten brands within the category per month.	Kantar, GfK
Category growth (CatGrowthc,p)	The evolution in the sales of product category p in country c. The evolution is the difference between the first and last year in the observation window.	Panel data
Category frequency (CatFreqc,p)	The average number of purchase events in the category per household, per month, averaged over the observation window.	Kantar, GfK
Category penetration (CatPenec,p)	Total number of unique households that bought in the category divided by the total number of unique households in the panel, per month, averaged over the observation window.	Kantar, GfK
Country Factors
Market size (MarketSizec)	Number of inhabitants in country c, averaged over the observation window. Expressed in the natural logarithm and averaged over the observation window.	Euromonitor
GDP per capita (GDPCapc)	GDP per capita in country c, averaged over the observation window.	Euromonitor
GDP growth (GDPGrowthc)	The evolution in GDP per capita in country c. The evolution is the difference between the first and last year in the observation window.	Euromonitor
Income inequality (Ginic)	The Gini index, where 0 represents perfect equality and 100 implies perfect inequality.	World Bank
Power distance (PDc)	Cultural dimension that captures the way people in a society relate to each other on a hierarchical scale.	Hofstede, Hofstede, and Minkov (2010)
Uncertainty avoidance (UAc)	Cultural dimension that reflects the degree to which a culture embraces uncertainty of the future.	Hofstede, Hofstede, and Minkov (2010)
Motivation toward achievement and success (Masc)	Cultural dimension that captures a preference in society for achievement, heroism, assertiveness, and material rewards for success.	Hofstede, Hofstede, and Minkov (2010)

Method

We proceed in three steps. First, to derive the time path for each of the marketing-mix elasticity for each brand we use a rolling-window approach, which we apply to every brand–category–country combination. Next, we estimate the trends in these time paths to gauge whether and how elasticities are evolving over time. In a third step, we regress the trends in elasticities on a broad set of brand, category, and country characteristics.

Step 1: Deriving the Time Path in Marketing-Mix Elasticities

Approach

Rolling-window estimation is a flexible and robust approach to estimate time-varying marketing-mix elasticities using a data sample of fixed size, which allows for direct comparison between estimates in different windows (Leeflang et al. 2000). It involves continuously updating the estimation model with the most recent data, allowing for the model parameters to vary over time. Rolling regression requires selecting a window size that is rolled over time in steps. This window size determines how many data points are selected for evaluating the coefficients. By systematically rolling the time window over successive periods, older data points are replaced with newer ones, enabling the capture of evolving relationships between variables. When rolled, the regression analysis is updated, along with the corresponding coefficient, and as such the “time path” of the estimated effects is revealed (Pauwels and Hanssens 2007). Rolling-window estimation has been used to glean the dynamic effect of broadcaster emotion on viewer tips (Lin, Yao, and Chen 2021), the evolving impact of product-harm crises (Cleeren, Van Heerde, and Dekimpe 2013), and business-model changes (Pauwels and Weiss 2008).

Rolling-window estimation provides a flexible and dynamic framework for addressing the Lucas critique in economic modeling, particularly when working with long time series. By continuously updating the model, allowing parameters to vary over time, and evaluating policy effects across periods, this approach helps account for evolving economic relationships and supports more reliable policy recommendations (Van Heerde, Dekimpe, and Putsis 2005).

Model specification

To derive the elasticities, we estimate the following model for each window and each brand-category-country combination:

$$\begin{array}{rl}\ln ({\mathrm{volms}}_{\mathrm{c},\mathrm{p},\mathrm{b},\mathrm{t}}^{\mathrm{k}})=& {\beta }_{0,\mathrm{c},\mathrm{p},\mathrm{b}}^{\mathrm{k}}+{\beta }_{1,\mathrm{c},\mathrm{p},\mathrm{b}}^{\mathrm{k}}\ln ({\mathrm{Price}}_{\mathrm{c},\mathrm{p},\mathrm{b},\mathrm{t}}^{\mathrm{k}})\\ & +{\beta }_{2,\mathrm{c},\mathrm{p},\mathrm{b}}^{\mathrm{k}}\ln ({\mathrm{Assort}}_{\mathrm{c},\mathrm{p},\mathrm{b},\mathrm{t}}^{\mathrm{k}})\\ & +{\beta }_{3,\mathrm{c},\mathrm{p},\mathrm{b}}^{\mathrm{k}}\ln ({\mathrm{Distri}}_{\mathrm{c},\mathrm{p},\mathrm{b},\mathrm{t}}^{\mathrm{k}})\\ & +{\beta }_{4,\mathrm{c},\mathrm{p},\mathrm{b}}^{\mathrm{k}}\ln ({\mathrm{C}\mathrm{\_}\mathrm{Price}}_{\mathrm{c},\mathrm{p},\mathrm{b},\mathrm{t}}^{\mathrm{k}})\\ & +{\beta }_{5,\mathrm{c},\mathrm{p},\mathrm{b}}^{\mathrm{k}}\ln ({\mathrm{C}\mathrm{\_}\mathrm{Assort}}_{\mathrm{c},\mathrm{p},\mathrm{b},\mathrm{t}}^{\mathrm{k}})\\ & +{\beta }_{6,\mathrm{c},\mathrm{p},\mathrm{b}}^{\mathrm{k}}\ln (\mathrm{C}\mathrm{\_}{\mathrm{Distri}}_{\mathrm{c},\mathrm{p},\mathrm{b},\mathrm{t}}^{\mathrm{k}})\\ & +{\beta }_{7,\mathrm{c},\mathrm{p}}^{\mathrm{k}}\ln (\mathrm{C}3\mathrm{\_}{\mathrm{Ret}}_{\mathrm{c},\mathrm{p},\mathrm{b},\mathrm{t}}^{\mathrm{k}})\\ & +{\beta }_{8,\mathrm{c},\mathrm{p},\mathrm{b}}^{\mathrm{k}}\ln (\mathrm{C}3\mathrm{\_}{\mathrm{Brand}}_{\mathrm{c},\mathrm{p},\mathrm{b},\mathrm{t}}^{\mathrm{k}})\\ & +{\beta }_{9,\mathrm{c},\mathrm{p},\mathrm{b}}^{\mathrm{k}}\ln ({\mathrm{Inflation}}_{\mathrm{c},\mathrm{t}}^{\mathrm{k}})\\ & +{\beta }_{10,\mathrm{c},\mathrm{p},\mathrm{b}}^{\mathrm{k}}\ln ({\mathrm{Trend}}_{\mathrm{t}}^{\mathrm{k}})+{\beta }_{11,\mathrm{c},\mathrm{p},\mathrm{b}}^{\mathrm{k}}\sin \left({\displaystyle \frac{2\pi \mathrm{t}}{52}}\right)\\ & +{\beta }_{12,\mathrm{c},\mathrm{p},\mathrm{b}}^{\mathrm{k}}\cos \left({\displaystyle \frac{2\pi \mathrm{t}}{52}}\right)\\ & +\sum _{\mathrm{i}=1}^6{\beta }_{12+\mathrm{i},\mathrm{c},\mathrm{p},\mathrm{b}}^{\mathrm{k}}{\mathrm{Control}\mathrm{\_}\mathrm{function}}_{\mathrm{i},\mathrm{c},\mathrm{p},\mathrm{b}}^{\mathrm{k}}\\ & +{\epsilon }_{\mathrm{c},\mathrm{p},\mathrm{b},\mathrm{t}}^{\mathrm{k}}.\end{array}$$

(1)

The subscripts c, p, b and t reflect country, product category, brand, and week. The superscript k expresses the window. The dependent variable ${\mathrm{volms}}_{\mathrm{c},\mathrm{p},\mathrm{b},\mathrm{t}}^{\mathrm{k}}$ represents the volume share of brand b within a specific product category p and country c in week t. The focal marketing-mix variables price, assortment, and distribution are denoted by ${\mathrm{Price}}_{\mathrm{c},\mathrm{p},\mathrm{b},\mathrm{t}}^{\mathrm{k}}$ , ${\mathrm{Assort}}_{\mathrm{c},\mathrm{p},\mathrm{b},\mathrm{t}}^{\mathrm{k}}$ , and ${\mathrm{Distri}}_{\mathrm{c},\mathrm{p},\mathrm{b},\mathrm{t}}^{\mathrm{k}}$ . We also include the same set of mix instruments for the combined competitors, i.e., ${\mathrm{C}\mathrm{\_}\mathrm{Price}}_{\mathrm{c},\mathrm{p},\mathrm{b},\mathrm{t}}^{\mathrm{k}}$ , ${\mathrm{C}\mathrm{\_}\mathrm{Assort}}_{\mathrm{c},\mathrm{p},\mathrm{b},\mathrm{t}}^{\mathrm{k}}$ , and ${\mathrm{C}\mathrm{\_}\mathrm{Distri}}_{\mathrm{c},\mathrm{p},\mathrm{b},\mathrm{t}}^{\mathrm{k}}$ . By taking the natural logarithm of the dependent variable and the marketing-mix variables, ${\beta }_{\mathrm{i},\mathrm{c},\mathrm{p},\mathrm{b}}^{\mathrm{k}}$ are elasticities.

We control for concentration in both brand and retailer markets by including $\mathrm{C}3{\mathrm{\_}\mathrm{Ret}}_{\mathrm{c},\mathrm{p},\mathrm{b},\mathrm{t}}^{\mathrm{k}}$ and ${\mathrm{C}3\mathrm{\_}\mathrm{Brand}}_{\mathrm{c},\mathrm{p},\mathrm{b},\mathrm{t}}^{\mathrm{k}}$ . We further control for inflation $({\mathrm{Inflation}}_{\mathrm{c},\mathrm{t}}^{\mathrm{k}})$ and for unobserved time-varying effects with a deterministic trend $({\mathrm{Trend}}_{\mathrm{t}}^{\mathrm{k}})$ . Two trigonometric terms, $\cos \left(\frac{2\pi \mathrm{t}}{52}\right)$ and $\sin \left(\frac{2\pi \mathrm{t}}{52}\right),$ are added to account in a parsimonious way for seasonal fluctuations (Gielens 2012). In addition, we include six control functions to correct for the potentially endogenous nature of the marketing mix. We offer more details about our identification procedure subsequently. Finally, ${\epsilon }_{\mathrm{c},\mathrm{p},\mathrm{b},\mathrm{t}}^{\mathrm{k}}$ is a white-noise residual. Lastly, our model is estimated at the brand–category–country level, allowing us to capture individual effects for each combination directly. Equation 1 thus relies exclusively on within-unit (over-time) variation, rendering additional fixed-effect corrections unnecessary.

The size of the time window k must be predetermined, and the window size may influence the ability to identify dynamic effects effectively. While wider windows provide more data points for estimation, narrower windows facilitate the observation of short-term dynamics approaching the end of a time series. The optimal window length strikes a balance between these competing factors, but it often relies on empirical experience rather than theoretical guidance. With 18 parameters to estimate, we use a window length of 130 weeks (approximately two and a half years), which yields about seven degrees of freedom per parameter. This strikes a balance between statistical robustness and a time horizon that aligns with the strategic planning cycles of most firms. Further, we shift the window each time with four weeks (or approximately one month), implying that each time the last four weeks of the 130-week window are dropped. Web Appendices A and B provide the number of window estimates per country and category, respectively.

Robustness checks

To evaluate the robustness of our modeling approach we check (1) the stationarity of the series, (2) different window lengths, (3) sensitivity to potentially confounding effects of price promotions and advertising, and (4) alternative estimation approaches.

Underlying stationarity

While rolling-window estimation does not strictly require stationary series, nonstationarity of variables can impact the reliability and interpretation of the results. Therefore, we assess stationarity both globally and locally. First, we conduct panel unit root tests on the full time series, which provide an overall view of stationarity but may obscure temporal variations. Second, we perform unit root tests within each rolling window to capture potential shifts in stationarity over time. The results indicate that the panel unit root test is passed for the full sample (p < .05). At the more granular window–brand–category–country level, over 95% of the rolling windows also pass the unit root test, suggesting that nonstationarity is not a concern.

Window length

In Web Appendix D, we report robustness checks related to our choice of window length. Specifically, we compare the results obtained using shorter windows of 90 observations (approximately five degrees of freedom per parameter) with those from longer windows of 150 (eight degrees per parameter) and 180 observations (ten degrees per parameter). As shown, the results remain highly robust across window lengths.

Sensitivity to omitted variables

Despite our best efforts, obtaining advertising data for our extensive set of brands proved impossible. Nevertheless, we obtained U.S. advertising expenditure data from the Kilts Center for Marketing. Additionally, while the price variable reflects the actual price paid and therefore accounts for price promotions, we were able to further refine the U.S. price measure by incorporating whether a purchase was made with a coupon and adjust accordingly. When both advertising and coupon usage variables are added to Equation 1, the results remain substantively unchanged. Summaries of the estimation results are provided in Web Appendix E. Our findings align with Datta et al. (2022), who showed for China and Hong Kong that not including advertising led to negligible bias in the estimates of price, assortment, and distribution elasticities.

Robustness to estimation approach

We reestimate the model using a Kalman filter framework. Unlike the rolling-window method, which estimates parameters over fixed-length segments, the Kalman filter continuously updates parameter estimates over time within a state-space structure, allowing for greater flexibility in capturing dynamic changes. However, this approach is computationally intensive, especially with large datasets involving nonlinearities and hierarchical structures. Given our data—spanning over 16,000 brands across three levels (brand, category, and country)—around 50% of the model specifications failed to reach convergence under the Kalman filter (Web Appendix F). However, for the subset of brands where convergence was achieved, the average elasticities using the Kalman filter correlated over .90 with those obtained from the rolling-window estimation. This suggests that the rolling-window approach provides reliable and computationally efficient estimates while maintaining consistency with more complex time-varying alternatives.

Endogeneity strategy

Manufacturers, like retailers, may anticipate how consumers react to their marketing mix and adjust their marketing-mix efforts accordingly. To account for possible endogeneity in these decisions, we use an instrumental-variable approach. Suitable instruments must meet stringent criteria, including being uncorrelated with the error term of the performance model, sufficiently correlated with the endogenous elements of independent variables, and uncorrelated with potentially omitted variables in the performance equation (Wooldridge 2010).

In the CPG industry, prices, assortment, and distribution coverage are influenced by the cost and markup structure, which is shaped by product nature, category, and retail environment. While costs are expected to correlate with marketing instruments, they should remain uncorrelated with demand and unobservable demand shocks. To mitigate this issue, Nevo (2001) proposes using marketing variables from similar but different markets as instrumental variables, under the premise that cost shocks causing exogenous variation in marketing variables in one market induce similar exogenous variations in others. Nevo's approach has been used in marketing by Dinner, Van Heerde, and Neslin (2014), Ma et al. (2011), and Van Heerde et al. (2013).

Nearby countries may share similar cost structures due to geographical proximity and other factors, but (consumer and wholesale) demand patterns for branded goods can diverge significantly. Regulatory disparities, cultural preferences, and local market dynamics contribute to variations in consumer behavior. Consequently, while costs may align across borders, demand remains a distinct and heterogeneous factor. In addition to varying consumer demand patterns, the set of retailers and their negotiation power may vary across countries. This variation means that brands must negotiate with different retail partners in each country, resulting in differences in the assortment offered and the intensity of distribution channels. These differences in distribution channels can lead to variations in product availability, shelf space, and promotional activities, further influencing consumer behavior. Even within the Eurozone, prices between neighboring countries such as Germany and Austria deviate because retailers substantially differentiate prices across borders, leveraging their market power to price discriminate between countries (Messner, Rumler, and Strasser 2023). However, despite this price differentiation, the underlying cost structures across nearby countries often exhibit similarities.

Given that we are working with national-level data covering a wide variety of product categories, we select information from geographically and/or culturally close national peer markets. For each peer market for a focal brand, we use values for the top three brands in the same category, excluding the focal brand if it is on the market, and calculate the average value. We follow a similar approach for competitor-mix variables. As such, we derive instruments that vary by brand, category, country, and time for the three focal and three competing marketing tools. We use the resulting instruments and all exogenous variables of the market share model to estimate the six instrumented variables for every brand–category–country combination. We draw on the parameters to compute the fitted residuals, which we subsequently include in Equation 1 as control functions $({\mathrm{Control}\mathrm{\_}\mathrm{function}}_{\mathrm{i},\mathrm{c},\mathrm{p},\mathrm{b}}^{\mathrm{k}})$ (Papies, Ebbes, and Van Heerde 2017).

Table 5 provides an overview of the potential endogeneity concerns, the corresponding instrumental-variable logic, and the control-function implementation used to correct for them.

Table 5.

Overview of Endogeneity Concerns and Identification Strategy.

Type of Endogeneity	Underlying Mechanism	Potential Bias	Instrument(s)	Why Instrument Is Valid in This Setting	Role of Control-Function Residual
1. Unobserved Demand Shocks (brand-, category-, or country-specific preference shifts not fully observed by the researcher)	Unobserved shocks simultaneously affect demand and firms’ marketing mix → regressors correlated with error	Upward/downward bias depending on shock direction	Peer-country average price, assortment, and distribution of top three brands	Demand shocks are local; cost shocks correlate across nearby/culturally close markets → instrument correlates with mix variables but not focal demand	Residual captures shock-driven endogenous variation; including it purges correlation with the error term
2. Simultaneity / Reverse Causality (firms adjust marketing in response to current or anticipated sales)	Demand → marketing (reverse causality) rather than solely marketing → demand	Elasticities reflect adjustment behavior, not causal effects	Same peer-country instruments	Peer-country mix reacts to their own markets, not focal brand performance; isolates supply-side variation	Residual isolates endogenous component tied to firms’ adjustments, removing simultaneity bias
3. Competitive Endogeneity (competitors change their mix in reaction to focal brand actions)	Competitor mix variables become functions of focal actions → cross-equation dependence	Bias in own and competitive elasticities	Peer-country competitor mix variables	Competitors in peer countries face similar cost structures but do not react to focal brand → provides exogenous competitor variation	Residual removes feedback-driven correlation between competitor actions and focal brand outcomes
4. Retailer/Channel Endogeneity (retailers adjust assortment/distribution in response to demand or negotiation outcomes)	Retailer-driven SKU listings, shelving, and distribution expand/contract when demand changes → endogenous nonprice mix	Overstated assortment/distribution elasticities	Peer-country assortment and distribution patterns of similar brands	Retailer negotiations and assortment norms are market-specific; peer-country values reflect cost/channel structure, not focal demand	Residual removes retailer-driven endogenous movement in distribution/assortment
5. Measurement Error in Marketing Variables (noise in SKU-level price, assortment count, and distribution recordings)	Classical attenuation or inconsistent estimates	Elasticities biased toward zero or unstable	Peer-country average marketing variables (top three brands)	Measurement error is idiosyncratic within country; peer-country averages carry signal but not focal noise	Residual isolates measurement-error component, uncovering the true exogenous variation

Notes: All instruments vary at the brand–category–country–time level because they are constructed from weekly marketing-mix series of the top three brands in each peer country, ensuring rich temporal and cross-sectional variation.

Step 2. Estimating the Trend in the Rolling-Window Elasticities Path

To capture marketing-mix elasticity trends, we estimate the following regression model separately for each of the three marketing elasticities—price (m = 1), distribution (m = 2), and assortment (m = 3):

$${\beta }_{\mathrm{m},\mathrm{c},\mathrm{p},\mathrm{b}}^{\mathrm{k}}={\alpha }_{0,\mathrm{m},\mathrm{c},\mathrm{p},\mathrm{b}}+{\alpha }_{1,\mathrm{m},\mathrm{c},\mathrm{p},\mathrm{b}}{\mathrm{Time}}^{\mathrm{k}}+{\mu }_{\mathrm{m},\mathrm{c},\mathrm{p},\mathrm{b}}^{\mathrm{k}}.$$

(2)

Time^k is a time-averaged linear trend and is captured by the start week of each wave k, thus α_0,m,c,p,b is the average elasticity over time for brand b in product category p and country c. The coefficient α_1,m,c,p,b reflects the trend in marketing-mix elasticity m for that brand.

To account for the uncertainty in the elasticity estimates, we estimate Equation 2 with weights equal to the inverse standard errors of the elasticity estimates (Van Heerde et al. 2013). We run this regression for each brand and thus have 16,642 trend estimates for each of the three marketing-mix elasticities.

Robustness

We explored alternative specifications beyond a linear trend, including a quadratic trend (time²) and a logarithmic specification (ln(time)) to account for potential decreasing returns to scale. However, neither alternative outperformed the linear specification in terms of model fit, as assessed by the Akaike information criterion, across the vast majority of cases. Given the superior parsimony and interpretability of the linear trend—especially in the context of our step 3 analyses—we proceeded with the linear specification.

Step 3. Estimating the Effects of Brand, Category, and Country Factors on Elasticity Trends

To quantify the moderating effects of brand, category, and country factors on elasticity trends, we estimate the following regression model for each marketing instrument m (m = 1 for price, m = 2 for assortment, m = 3 for distribution).

$$\begin{gathered} {\alpha }_{1,\mathrm{m},\mathrm{c},\mathrm{p},\mathrm{b}}={\gamma }_{0,\mathrm{m}}+{\gamma }_{1,\mathrm{m}}{\mathrm{PricePos}}_{\mathrm{c},\mathrm{p},\mathrm{b}}+{\gamma }_{2,\mathrm{m}}{\mathrm{AssortPos}}_{\mathrm{c},\mathrm{p},\mathrm{b}} \\ +{\gamma }_{3,\mathrm{m}}{\mathrm{DistriPos}}_{\mathrm{c},\mathrm{p},\mathrm{b}}+{\gamma }_{4,\mathrm{m}}{\mathrm{CatReach}}_{\mathrm{c},\mathrm{p},\mathrm{b}} \\ +{\gamma }_{5,\mathrm{m}}{\mathrm{IntReach}}_{\mathrm{c},\mathrm{p},\mathrm{b}}+{\gamma }_{6,\mathrm{m}}{\mathrm{Local}}_{\mathrm{c},\mathrm{p},\mathrm{b}}+{\gamma }_{7,\mathrm{m}}{\mathrm{Power}}_{\mathrm{c},\mathrm{p},\mathrm{b}} \\ +{\gamma }_{8,\mathrm{m}}{\mathrm{BrandGrowth}}_{\mathrm{c},\mathrm{p},\mathrm{b}}+{\gamma }_{9,\mathrm{m}}{\mathrm{OwnEff}}_{\mathrm{c},\mathrm{p},\mathrm{b}} \\ +{\gamma }_{10,\mathrm{m}}{\mathrm{CompVul}}_{\mathrm{c},\mathrm{p},\mathrm{b}}+{\gamma }_{11,\mathrm{m}}{\mathrm{Ecom}}_{\mathrm{c},\mathrm{p}} \\ +{\gamma }_{12,\mathrm{m}}{\mathrm{HardDis}}_{\mathrm{c},\mathrm{p}}+{\gamma }_{13,\mathrm{m}}{\mathrm{PLShare}}_{\mathrm{c},\mathrm{p}} \\ +{\gamma }_{14,\mathrm{m}}{\mathrm{RetConc}}_{\mathrm{c},\mathrm{p}}+{\gamma }_{15,\mathrm{m}}{\mathrm{BrandConc}}_{\mathrm{c},\mathrm{p}} \\ +{\gamma }_{16,\mathrm{m}}{\mathrm{CatPromo}}_{\mathrm{c},\mathrm{p}}+{\gamma }_{17,\mathrm{m}}{\mathrm{CatInno}}_{\mathrm{c},\mathrm{p}} \\ +{\gamma }_{18,\mathrm{m}}{\mathrm{CatGrowth}}_{\mathrm{c},\mathrm{p}}+{\gamma }_{19,\mathrm{m}}{\mathrm{CatFreq}}_{\mathrm{c},\mathrm{p}} \\ +{\gamma }_{20,\mathrm{m}}{\mathrm{CatPene}}_{\mathrm{c},\mathrm{p}}+{\gamma }_{21,\mathrm{m}}{\mathrm{CatSOW}}_{\mathrm{c},\mathrm{p}} \\ +{\gamma }_{22,\mathrm{m}}{\mathrm{MarketSize}}_{\mathrm{c}}+{\gamma }_{23,\mathrm{m}}{\mathrm{GDPCap}}_{\mathrm{c}} \\ +{\gamma }_{24,\mathrm{m}}{\mathrm{GDPGrow}}_{\mathrm{c}}+{\gamma }_{25,\mathrm{m}}{\mathrm{GINI}}_{\mathrm{c}}+{\gamma }_{26,\mathrm{m}}{\mathrm{PD}}_{\mathrm{c}} \\ +{\gamma }_{27,\mathrm{m}}{\mathrm{UA}}_{\mathrm{c}}+{\gamma }_{28,\mathrm{m}}{\mathrm{MAS}}_{\mathrm{c}}+{\delta }_{\mathrm{m},\mathrm{p}}^{\mathrm{″}}+{\delta }_{\mathrm{m},\mathrm{c},\mathrm{p},\mathrm{b}}, \end{gathered}$$

( 3)

where the variables are defined in Table 4. We use a random effect at the category $({{\delta }^{″}}_{\mathrm{m},\mathrm{p}})$ level. As in step 2, we use a weighted regression to control for uncertainty in the trend estimates.

Results

Univariate Results

Before discussing elasticity trends, we provide insight in the baseline or mean elasticities, α_0,m,c,p,b, as obtained through Equation 2 (Table 6). The mean own elasticities have the expected sign: negative for price and positive for assortment and distribution (all ps < .001). The own elasticities are significant at p < .05 for 80%–90% of the brands. Likewise, the mean competitive elasticities have the expected sign, positive for price and negative for assortment and distribution (all ps < .001). The mean own elasticities align with Datta et al. (2022), and the mean competitive price elasticity matches the .26 reported in a meta-analysis by Auer and Papies (2020). Thus, our results have face validity. Moreover, this converging evidence provides further confirmation that our estimates are robust to omitted variable bias.

Table 6.

Mean and Median Own and Competitive Elasticities.

	Price		Assortment		Distribution
	Mean (Median)	Added Z	Mean (Median)	Added Z	Mean (Median)	Added Z
Own elasticity	−.640 (−.471)	−177.5	.311 (.276)	202.34	.213 (.164)	127.12
Competitive elasticity	.274 (.280)	109.65	−.178 (−.154)	−62.69	−.115 (−.102)	−46.94

Note: Significance of the means is tested using the method of added Zs (Rosenthal 1991). All means are significant at p < .001.

Moving to our focal interest, viz. the trends, α_1,m,c,p,b, we notice that for each marketing-mix instrument, for about 75%–80% of the brands, the trend is significant at p < .05. ¹⁰ Combined across the three instruments, marketing effectiveness remains constant for only 1.61% of the brands. This confirms Raman et al.’s (2012) argument that change, rather than stability, is the rule. The trends are virtually uncorrelated across marketing-mix instruments, the largest correlation in absolute magnitude being between price and distribution (r = −.057). Looking at individual brands, all three elasticities become significantly stronger for only 6.1% of the brands, while all three become significantly weaker for 7.1% of the brands.

Table 7 breaks down the annual change for each marketing-mix element. Across the more than 16,000 brands, changes in price, assortment, and distribution elasticities exhibit pronounced heterogeneity. Price elasticities show the widest dispersion, with some brands becoming substantially more or less price sensitive over time. Personal care brands trend more positively in price elasticity than brands in food and beverages or household care, while there is relatively little variation across these three product types in trends for assortment and distribution elasticities. Geographic variation is more pronounced: Asia exhibits the broadest range of changes across all three instruments, particularly in price. North America combines a positive shift in average price elasticity with a negative shift in average distribution elasticity, and Eastern Europe shows the sharpest average decline in assortment elasticity. The lack of uniformity across these dimensions underscores the need to examine the factors driving these divergent paths.

Table 7.

Annual Changes by Marketing-Mix Instrument.

	Mean	p10	p25	Median	p75	p90
A: Price
Overall	.001	−.643	−.220	.001	.231	.633
Region
Asia	.026	−1.080	−.367	.029	.479	1.103
Eastern Europe	−.016	−.757	−.315	−.015	.274	.698
Latin America	−.009	−.412	−.146	−.012	.128	.409
North America	.046	−.246	−.086	.032	.157	.336
Western Europe	−.002	−.554	−.209	.006	.220	.533
Product Type
Food and beverages	−.006	−.706	−.257	−.006	.247	.669
Household care	−.010	−.646	−.228	.002	.231	.616
Personal care	.017	−.492	−.159	.009	.206	.568
B: Assortment
Overall	−.009	−.298	−.133	−.005	.119	.280
Region
Asia	−.006	−.390	−.173	−.003	.154	.369
Eastern Europe	−.020	−.338	−.152	−.013	.127	.291
Latin America	−.002	−.285	−.132	.001	.128	.283
North America	.000	−.224	−.101	−.003	.099	.232
Western Europe	−.008	−.250	−.112	−.006	.099	.235
Product Type
Food and beverages	−.007	−.303	−.132	−.004	.123	.289
Household care	−.009	−.303	−.138	−.003	.124	.290
Personal care	−.010	−.283	−.133	−.007	.110	.253
C: Distribution
Overall	.000	−.258	−.109	−.002	.106	.256
Region
Asia	.018	−.325	−.137	.007	.150	.401
Eastern Europe	−.002	−.239	−.108	−.006	.102	.235
Latin America	.001	−.261	−.108	.001	.113	.252
North America	−.013	−.241	−.115	−.007	.078	.215
Western Europe	−.005	−.239	−.099	−.003	.092	.224
Product Type
Food and beverages	−.007	−.303	−.132	−.004	.123	.289
Household care	−.009	−.303	−.138	−.003	.124	.290
Personal care	−.010	−.283	−.133	−.007	.110	.253

Building on these patterns of heterogeneity, Figure 2 plots the mean elasticity against the annual change in elasticity. The figure reinforces the cross-national variation in trends. It also illustrates that differences in the evolution of marketing-mix effectiveness are not solely a function of the baseline elasticity within a market. In several cases, countries with relatively high baseline elasticities experience further increases, whereas others show declines, and the reverse is true for low-baseline markets. Thus, the baseline provides an initial reference point, but the trajectory of consumer response in each country is shaped by additional market- and context-specific factors.

Figure 2.

Average and Annual Change in Elasticities.

Drivers of Heterogeneity in Evolution in Marketing-Mix Elasticities

Our univariate analysis revealed considerable heterogeneity in the evolution of marketing-mix elasticities across brands. We now turn to our third-stage regression, exploring the potential impact of various brand, category, and country drivers. Table 8 presents the regression coefficients, associated p-value, and their effect size. The effect size is calculated as the effect of the predictor if it is one standard deviation above its mean value. ¹¹

Table 8.

Regression of Annual Change in Marketing-Mix Elasticities on Predictors.

	Price			Assortment			Distribution
	γ	p-Value	Effect Size	γ	p-Value	Effect Size	γ	p-Value	Effect Size
Intercept	1.647	.005		.085	.678		1.077	<.001
Brand factors
Price positioning (γ1)	.072	.018	.047	.010	.193	.007	−.018	.010	−.011
Assortment positioning (γ2)	−.004	.873	−.003	−.028	.001	−.020	.018	.010	.013
Distribution positioning (γ3)	−.042	.235	−.017	−.046	<.001	−.018	.009	.338	.004
Category reach (γ4)	−.018	.004	−.048	−.002	.282	−.006	.006	.001	.016
International reach (γ5)	−.208	.024	−.047	.031	.320	.007	−.068	.010	−.015
Local brand (γ6)	−.101	.005	−.101	−.053	<.001	−.053	−.040	<.001	−.040
Brand power (γ7)	−.830	<.001	−.131	.276	<.001	.044	−.288	<.001	−.046
Brand growth (γ8)	−.495	<.001	−.072	.190	<.001	.028	.100	<.001	.015
Own effectiveness (γ9)	−.115	<.001	−.215	−.101	<.001	−.189	−.101	<.001	−.189
Competitive vulnerability (γ10)	.019	<.001	.061	−.003	.009	−.009	−.007	<.001	−.022
Category Factors
E-commerce share (γ11)	−.005	.324	−.020	−.002	.255	−.008	.010	<.001	.044
Discounter share (γ12)	.002	.208	.020	.003	<.001	.032	.002	.002	.016
Private label share (γ13)	.361	.659	.021	−.117	.675	−.007	.027	.924	.002
Retailer concentration (γ14)	.040	.725	.006	.012	.762	.002	−.253	<.001	−.040
Brand concentration (γ15)	.329	.005	.059	−.199	<.001	−.036	−.107	.003	−.019
Promotion pressure (γ16)	−.835	<.001	−.085	.428	<.001	.043	.005	.916	.001
Innovativeness (γ17)	.001	.064	.033	<.001	.103	.010	−.000	.006	−.015
Category growth (γ18)	−.214	.001	−.047	.005	.823	.001	.076	<.001	.017
Purchase frequency (γ19)	.007	.001	.059	−.004	<.001	−.035	.005	<.001	.044
Penetration (γ20)	−.348	<.001	−.090	−.001	.962	.000	−.028	.296	−.007
Share of wallet (γ21)	−2.087	.003	−.057	1.100	<.001	.030	−2.090	<.001	−.057
Country Factors
Market size (log) (γ22)	.027	.060	.037	.002	.630	.003	−.012	.005	−.016
GDP per capita (log) (γ23)	−.093	.003	−.087	−.011	.349	−.010	−.050	<.001	−.047
GDP growth (γ24)	−.203	.008	−.064	−.110	<.001	−.035	−.003	.894	−.001
Income inequality (γ25)	.000	.965	.001	.001	.178	.010	−.003	<.001	−.024
Power distance (γ26)	−.002	.029	−.052	.000	.928	−.001	−.002	<.001	−.035
Uncertainty avoidance (γ27)	−.001	.323	−.020	−.000	.157	−.010	.001	.006	.017
MAS (γ28)	.000	.919	−.002	−.001	.003	−.016	−.002	<.001	−.043

Notes: Effect size is calculated as the parameter estimate multiplied by one standard deviation of the predictor. MAS = motivation toward achievement and success.

Brand factors

Several brand-level characteristics significantly shape how marketing-mix elasticities evolve. Brands that emphasize a premium price positioning become less price sensitive over time (γ_{1, price} = .072, p = .018), which is consistent with Krishnamurthi and Papatla (2003). Premium brands also experience a decline in distribution effectiveness (γ_{1, distribution} = −.018, p = .010), indicating diminishing marginal returns to availability and suggesting that premium brands might be stuck in a niche. Brands with a larger assortment relative to their competitors experience a downward trend in assortment effectiveness (γ_{2, assortment} = −.028, p = .001). Interestingly, while brands with larger assortments see diminishing returns to further assortment expansion, they benefit from higher distribution effectiveness (γ₂, _distribution = .018, p = .010), suggesting that broad assortments strengthen the sales impact of shelf presence.

Brand reach attributes reveal further patterns. Brands with broader category reach (umbrella brands) see the effectiveness of their marketing activities increase, at least for price (γ_{4, price} = −.018, p = .004) and distribution (γ_{4, distribution} = .006, p = .001), which is aligned with the idea that umbrella branding signals quality (Erdem 1998). Brands with broader international reach become more price sensitive (γ_{5, price} = −.208, p = .024) but see a decline in distribution effectiveness (γ_{5, distribution} = −.068, p = .010).

Our study framework posed two opposing forces: (1) cultural globalization, which should boost the marketing effectiveness of global brands, and (2) lower local responsiveness, which has the opposite effect. It appears that both factors may be operating, albeit on different instruments. Since distribution is still mostly a local affair, as Lidl's initial centrally managed expansion and hit-and-miss performance in the United States illustrates—the decline in distribution effectiveness is particularly interesting. Local brands exhibit significant trends across all three instruments, but in distinct directions. They become more price sensitive over time (γ_{6, price} = −.101, p = .005) while experiencing declines in assortment (γ_{6, assortment} = −.053, p < .001) and distribution (γ_{6, distribution} = −.040, p = .000) effectiveness. Local brands are experiencing a comeback in various local markets (Steenkamp 2019), but the decline in assortment and distribution elasticity suggests they still have some way to go before they can compete on nonprice instruments.

The impact of brand power is also multifaceted. High-share brands experience a growing price sensitivity (γ_{7, price} = −.830, p < .001). Our finding does not support the conjecture that larger brands have more market power to price toward the less elastic portion of demand, likely because we control for price position. High-share brands further benefit from increased assortment effectiveness (γ_{7, assortment} = .276, p < .001), while seeing a decline in distribution effectiveness (γ_{7, distribution} = −.288, p < .001). This divergence may reflect saturation effects in access channels, alongside consumer expectations for breadth among premium brands. Growing brands also become more price sensitive over time (γ_{8, price} = −.495, p < .001), a finding consistent with Simon (1979). They also benefit from improved assortment (γ_{8, assortment} = .190, p < .001) and distribution (γ_{8, distribution} = .100, p < .001) effectiveness.

More elastic brands tend to become less elastic over time (γ_{9, price} = −.115, p < .001; γ_{9, assortment} = −.101, p < .001; γ_{9, distribution} = −.101, p < .001). This is consistent with the idea that in the longer run, brands’ marketing effectiveness becomes more similar as markets move more toward equilibrium. Brands that are more vulnerable to competitor marketing moves experience a decline in their own marketing effectiveness. We find this for all three instruments (γ_{10, price} = .019, p < .001; γ_{10, assortment} = −.003, p = .009; γ_{10, distribution} = −.007, p < .001), which is in line with the tit-for-tat argument (Lal and Villas-Boas 1998).

Category factors

E-commerce increases distribution effectiveness (γ_{11, distribution} = .010, p < .001), reflecting the expanding role of omnichannel access in reaching consumers. In categories with a strong discounter presence, brands see greater assortment (γ_{12, assortment} = .003, p < .001) and distribution effectiveness (γ_12, distribution = .002, p = .002), perhaps due to the value contrast and higher footfall. Private label share, in contrast, has no significant effect.

Retailer concentration significantly reduces distribution effectiveness (γ_{14, distribution} = −.253, p < .001). Brand concentration reduces the sensitivity to all three instruments (γ_{15, price} = .329, p = .005; γ_{15, assortment} = −.199, p < .001; γ_{15, distribution} = −.107, p = .003), suggesting that in concentrated markets, firms operate closer to the less elastic portion of demand (Gordon, Goldfarb, and Li 2013).

The price sensitivity of brands in heavily promoted categories exhibits an upward trend (γ_{16, price} = −.835, p < .001), which is aligned with Jedidi, Mela, and Gupta (1999). However, promotions also draw more consumers to the category, which offers opportunities for assortment expansion to cater to their needs (γ_{16, assortment} = .428, p < .001). High innovation intensity in the category reduces brands’ price sensitivity (γ_{17, price} = .001, p = .064), as brands offer new, nonprice benefits. We further find that distribution sensitivity declines (γ_{17, distribution} = −.000, p = .006). This is unexpected because new SKUs should benefit from wider distribution. It is possible that new SKUs merely replace old ones.

Categories that are in the growth stage of the product life cycle experience an increase in distribution effectiveness (γ_{18, distribution} = .076, p < .001) and in price sensitivity (γ_{18, price} = −.214, p = .001). Strongly growing categories attract more new buyers who appear to be more price sensitive, offering prospects for increased distribution. This latter finding is consistent with Bijmolt, Van Heerde, and Pieters (2005).

Purchase frequency decreases price sensitivity (γ_{19, price} = .007, p = .001) and assortment effectiveness (γ_{19, assortment} = −.004, p < .001) and increases distribution effectiveness (γ_{19, distribution} = .005, p < .001) over time. Similarly, category penetration heightens price sensitivity (γ_{20, price} = −.348, p < .001). Share of wallet exhibits divergent effects: It increases price sensitivity (γ₂₁, price = −2.087, p = .003), which is consistent with Gordon, Goldfarb, and Li (2013). It also reduces distribution effectiveness (γ₂₁, distribution = −2.090, p < .001) and enhances assortment effectiveness (γ_{21, assortment} = 1.100, p < .001).

Country factors

Market size decreases price elasticity (γ_{22, price} = .027, p = .060) and reduces distribution effectiveness (γ_{22, distribution} = −.012, p = .005) over time. These findings are consistent with the idea that large markets have more crowded product spaces, which reduces marketing effectiveness (Schwartz 2004). GDP per capita (γ_{23, price} = −.093, p = .003) and GDP growth (γ_{24, price} = −.203, p = .008) are associated with increasing, rather than decreasing, price sensitivity. This goes against budget constraint theory but is aligned with empirical evidence (Bahadir, Bharadwaj, and Srivastava 2015; Datta et al. 2022). Higher GDP per capita is also associated with declining distribution effectiveness (γ_{23, distribution} = −.050, p < .001), and higher GDP growth is associated with diminishing assortment effectiveness (γ_{24, assortment} = −.110, p < .001). Income inequality dampens distribution effectiveness (γ_{25, distribution} = −.003, p = .000), which aligns with Kozlenkova et al. (2021).

Power distance increases price sensitivity (γ_{26, price} = −.002, p = .029) and reduces distribution effectiveness (γ_{26, distribution} = −.002, p < .001), which is in line with Datta et al. (2022). We suggested that in countries high in uncertainty avoidance, assortment elasticities might decline, as new products are inherently riskier and may increasingly rely on price and distribution as quality cues. We find directional evidence for all three effects, but only the effect on distribution effectiveness is statistically significant (γ_{27, distribution} = .001, p = .006). Finally, a country's degree of motivation toward achievement and success is negatively associated with the evolution in assortment (γ_{28, assortment} = −.001, p = .003) and distribution (γ_{28, distribution} = −.002, p < .001) elasticity.

Discussion

Drawing on a massive, consistently collected dataset spanning 34 countries, we apply a rolling-window modeling framework to track how price, assortment, and distribution elasticities evolve over time for more than 16,000 brands in over 80 CPG categories in four continents over 5–15 years. Roughly equal numbers of brands experience increases and decreases in effectiveness for any given marketing-mix instrument. Yet, when averaged across brands, the magnitude of annual change appears minimal, potentially creating the false impression of stability. These averages conceal substantial heterogeneity in the direction and magnitude of change. To uncover underlying patterns, we examine brand-, category-, and country-level predictors of elasticity trajectories. Table 9 summarizes the results of this analysis. We rank the significant predictors for each marketing instrument by the magnitude of their effect and derive implications for marketing theory and practice, which we discuss next.

Table 9.

Ranking of Predictors by Effect Size.

#	Effect on the Trend in Price Elasticity	Effect on the Trend in Assortment Elasticity	Effect on the Trend in Distribution Elasticity
1	Own price elasticity (−)	Own assortment elasticity (−)	Own distribution elasticity (−)
2	Brand power (+)	Local brand (−)	Share of wallet (−)
3	Local brand (+)	Brand power (+)	GDP/capita (−)
4	Penetration (−)	Promotion pressure (+)	Brand power (−)
5	GDP/capita (−)	Brand concentration (−)	E-commerce share (+)
			Purchase frequency (+)
6	Promotion pressure (−)	Purchase frequency (−)	MAS (−)
		GDP growth (−)
7	Brand growth (−)	Discounter share (+)	Local brand (−)
			Retail concentration (−)
8	GDP growth (−)	Share of wallet (+)	Power distance (−)
9	Competitive price vulnerability (−)	Brand growth (+)	Income equality (−)
10	Brand concentration (−)	Assortment positioning (−)	Competitive distribution vulnerability (−)
	Purchase frequency (−)
11	Share of wallet (+)	Distribution positioning (−)	Brand concentration (−)
12	Power distance (+)	MAS (−)	Category growth (+)
			Uncertainty avoidance (+)
13	Category reach (+)	Competitive assortment vulnerability (−)	Category reach (+)
			Discounter share (+)
			Market size (−)
14	Price positioning (−)		International reach (−)
	International reach (+)		Brand growth (+)
	Category growth (+)		Category innovativeness (−)
15	Market size (−)		Assortment positioning (+)
16	Category innovativeness (−)		Price positioning (−)

Notes: For each marketing-mix instrument, the predictors are ranked from strongest to weakest based on their effect size, provided that the associated p-value is below .10. In parentheses is whether the predictor is associated with an increase (+) or a decrease (–) in the trend of the marketing-mix instrument in question. Brand-level predictors are regular font, category-level predictors are underlined, country-level predictors are in italics. MAS = motivation toward achievement and success.

Implications for Marketing Theory

Our results reveal several broader conceptual themes: strong convergence in elasticity trajectories, the systematic role of competitive vulnerability, the limited role of one of branding's core elements (viz., brand positioning), the important role of local brands, the centrality of category shopping behaviors, and the puzzling effects of country wealth and growth. Each of these factors is of great relevance to marketing theory. Our results expose gaps in current theory and provide a platform for future work.

First, the strongest predictor of change is the brand's own baseline elasticity. For all three instruments, high average responsiveness predicts the steepest declines over time—consistent with a regression-to-the-mean pattern observed by Sharp (2010) and the idea that markets might move toward greater equilibrium over time. Yet, the underlying mechanisms for this strongest and most consistent effect are currently speculative and require stronger theorizing.

Second, high sensitivity to competitor actions consistently erodes the effectiveness of the corresponding own instrument over time, suggesting that competitive “tit-for-tat” dynamics not only neutralize short-term gains but may gradually degrade a brand's long-term capacity to move the market. However, its relative effect size is notably smaller than that of a brand's own clout and category-level buyer-behavior factors. This indicates that while competitive vulnerability matters, the trajectory of marketing effectiveness is shaped more strongly by intrinsic brand strength and the consumption patterns embedded in the category. Future research could explore conditions in which competitive vulnerability plays a more determinant role.

Third, positioning emerges as a weak predictor of change. One interpretation is that positioning stabilizes effectiveness, dampening shifts over time. Another is more provocative: The influence of positioning on the trajectory of marketing effectiveness may be overstated. Given the prominent role of positioning in branding theory, this warrants further research.

Fourth, local brands experience greater changes in responsiveness than international brands, with a notable shift from assortment and distribution to price sensitivity. This may indicate commoditization or quality parity with global competitors, both of which intensify price-based competition. This challenges the assumption that local authenticity protects against price erosion. An unanswered question is how international versus local brands will fare in an era of retreating globalization.

Fifth, retailer factors appear less “disruptive” than expected. Given their prominent place in retail strategy debates, it is striking that they exert a rather modest influence on changes in marketing effectiveness. While e-commerce and hard discounter formats may disrupt channel structures and margins, their role in reshaping long-term responsiveness appears weaker than widely assumed—suggesting that elasticity evolution is more grounded in brand- and category-level forces. Might retailer factors exert influence indirectly via brand factors?

Sixth, the way consumers shop within a category, rather than where they shop, emerges as a powerful driver of changes in marketing-mix responsiveness. Penetration is a top-five factor in shaping the change in price sensitivity. High purchase frequency generally strengthens distribution effectiveness but comes at the expense of price and assortment responsiveness. High share of wallet has the opposite effects. These patterns suggest that ingrained shopping habits and budget priorities—not broad retailer trends like e-commerce penetration or discount formats—play a central role in shaping how marketing effectiveness evolves over time.

Seventh, we find that in wealthy countries and those with strongly growing incomes, price sensitivity is increasing over time. This dynamic pattern aligns with and extends the observation of Datta et al. (2022), but it goes against standard economic theory. Why that is the case is unclear. Do consumers in richer environments rely less on price as quality indicator compared with consumers in poorer markets (Van Ewijk, Gijsbrechts, and Steenkamp 2022)? Might brands have more relevance in poorer countries (Fischer, Völckner, and Sattler 2010)?

These themes all suggest promising research directions. Given that the themes cast a wide net, which is in the spirit of EF research, future research will need to employ a variety of research designs. Experimental and quasi-experimental designs could isolate the mechanisms behind convergence in elasticity trajectories, such as equilibrium forces, consumer learning, or competitive adaptation. A combination of survey research to measure brand positioning and panel data can be used to explore the role of brand positioning and local brands. Field experiments might be employed to study in-depth natural experiments and exogenous shocks offer opportunities to study how competitive responses or retailer format changes reshape the evolution of elasticities. Cross-format retail analyses could deepen understanding of how channel structures influence the long-term effectiveness of assortment and distribution, providing a bridge between elasticity generalizations and emerging omnichannel theory.

Managerial Implications

Allocation aids

Our results show that 98.5% of brands experience a change in marketing effectiveness, and this change is often substantial. This underscores the limitations of static allocation rules, which risk misaligning resources (Raman et al. 2012).

Local market allocation

Local brand managers should regularly reassess the balance of spending across mix instruments to reflect the evolving elasticity patterns of their brand. To dynamically allocate budget across marketing instruments, the manager can turn to the analytical model of Raman et al. (2012). We illustrate the profit-maximizing allocation of a given budget to assortment and distribution. The purpose of our simple example is to demonstrate that the changes in marketing elasticities identified in our study have managerial relevance. Our illustration concerns a leading beverage brand in the United States, where we compare the optimal allocation for 2020 versus 2015. We assume equal unit costs for assortment and distribution. Of course, the brand manager has this cost information and can add it to the calculation in a straightforward way. The results are reported in Table 10. We can see the optimal allocation ratio changes noticeably. This brand benefits from a strengthening in both assortment and distribution elasticity, but the former increases considerably more. As a result, the percentage of the profit-maximizing budget allocated to assortment increases from 19.9% in 2015 to 44.3% in 2020.

Table 10.

Allocation of Assortment Versus Distribution Budget.

Year	${\mathbf{E}\mathbf{l}\mathbf{a}\mathbf{s}\mathbf{t}}_{\mathbf{t}}^{\mathbf{a}\mathbf{s}\mathbf{s}\mathbf{o}\mathbf{r}\mathbf{t}\mathbf{m}\mathbf{e}\mathbf{n}\mathbf{t}}$	${\mathbf{E}\mathbf{l}\mathbf{a}\mathbf{s}\mathbf{t}}_{\mathbf{t}}^{\mathbf{d}\mathbf{i}\mathbf{s}\mathbf{t}\mathbf{r}\mathbf{i}\mathbf{b}\mathbf{u}\mathbf{t}\mathbf{i}\mathbf{o}\mathbf{n}}$	${\mathbf{a}\mathbf{l}\mathbf{l}\mathbf{o}\mathbf{c}}_{\mathbf{t}}^{*}$ (% to Assortment)
2015	.021	.103	19.9%
2020	.056	.124	44.3%

Notes: Allocation is given by ${\mathrm{alloc}}_{\mathrm{t}}*=\frac{{\mathrm{Elast}}_{\mathrm{t}}^{\mathrm{assortment}}\times {\mathrm{Cost}}_{\mathrm{t}}^{\mathrm{distribution}}}{{\mathrm{Elast}}_{\mathrm{t}}^{\mathrm{distribution}}\times {\mathrm{Cost}}_{\mathrm{t}}^{\mathrm{assortment}}}$ , where ${\mathrm{Elast}}_{\mathrm{t}}^{\mathrm{assortment}}$ and ${\mathrm{Elast}}_{\mathrm{t}}^{\mathrm{distribution}}$ are the assortment and distribution elasticities in planning period t (e.g., year t), and ${\mathrm{Cost}}_{\mathrm{t}}^{\mathrm{assortment}}$ and ${\mathrm{Cost}}_{\mathrm{t}}^{\mathrm{distribution}}$ are the unit costs of assortment and distribution. Since we do not have cost data (although obviously, the brand manager would have them), for the purposes of this illustration we assume these to be the same across instruments.

Cross-market allocation

Our results also inform global brand managers about their dynamic cross-national budget allocation. Building on the same beverage brand, we illustrate this for price elasticity to determine the allocation of a brand's promotion budget across countries. We use Fischer et al.’s (2011) optimal allocation rule. To keep the example tractable, we do this analysis for ten important CPG countries across four continents. We compare two allocation strategies: no-change assumption—based solely on baseline price elasticities—versus change-adjusted allocation, incorporating the observed three-year change in price elasticity. Table 11 shows the results.

Table 11.

Allocation of Promotion Budget Across Countries.

Country	Sales (Millions)	Expected Growth Sales	Baseline Price Elasticity	Annual Change in Price Elasticity	“Optimal” Allocation Assuming Static Elasticity	“Optimal” Allocation Accounting for Change
Argentina	94.14	.993	−.298	−.079	.5%	.8%
Chile	151.55	.991	−.131	−.155	.3%	1.4%
China	100.13	1.024	−2.351	−.568	4.2%	4%
France	704.57	.993	−.125	.067	1.5%	.0%
Germany	1,686.32	.994	−.742	.027	21.6%
Italy	120.40	1.000	−.731	−.098	1.5%
Poland	136.01	.908	−5.837	−1.941	12.5%	22.0%
Spain	240.12	.999	−.244	−.065	1.0%	1.6%
U.K.	946.34	.985	−1.905	.166	30.9%	20.0%
U.S.	4,462.75	.999	−.333	−.031	25.8%	29.1%

Notes: The last column shows the optimal allocation after three years. The “optimal” allocation is based on the decision rule developed in Fischer et al. (2011): % allocated to country c for brand $\mathrm{b}=\frac{{\mathrm{sales}}_{\mathrm{cb}}\times {\hat{\alpha }}_{\mathrm{cbt}}^1\times {\mathrm{margin}}_{\mathrm{cb}}\times {\mathrm{growth}}_{\mathrm{cb}}}{{\sum }_{\mathrm{j}\in 10}{\mathrm{sales}}_{\mathrm{jb}}\times {\hat{\alpha }}_{\mathrm{jbt}}^1\times {\mathrm{margin}}_{\mathrm{jb}}\times {\mathrm{growth}}_{\mathrm{jb}}}\times 100\text{\%},$ where sales_cb is equal to a brand's average monthly sales in country c in the last year, ${\hat{\alpha }}_{\mathrm{cbt}}^1$ is brand b's predicted price elasticity in country c in year t, margin_cb is brand b's profit margin in country c, and growth_cb is brand b's expected (unit) growth rate. Since we do not have cost or profit-margin data (although obviously, the brand manager would have both), for the purposes of this illustration and following Datta et al. (2022), we assume these to be the same across countries.

Poland's already high baseline price sensitivity (−5.837) becomes even more pronounced over time (−1.941), nearly doubling its optimal allocation from 12.5% to 22.0% when the dynamics in price effectiveness are considered. The percentage allocated to the United States and China also grow. The United Kingdom's price elasticity becomes less elastic (+.166), reducing its optimal allocation by one third. Germany also sees its share decline. The managerial takeaway is clear: Dynamic, elasticity-informed allocation channels resources to markets where marketing actions will deliver the largest incremental impact—and prevents overinvestment in markets where responsiveness is eroding.

Archetype recommendations

Because the direction and magnitude of change vary widely across instruments, brands, categories, and countries, managers need to translate broad patterns into context-specific spending decisions. To illustrate, we use our effect size estimates to project annual changes in price, assortment, and distribution effectiveness for four managerially important brand profiles (Table 12). ¹² We provide the baseline change as benchmark.

Table 12.

Predicted Annual Change in Marketing-Mix Elasticities for Four Managerially Relevant Brand Scenarios.

Scenario	Price Trend Change	Assortment Trend Change	Distribution Trend Change
Premium global brand	−.094	.090	.049
Brand with high clout	−.574	−.032	−.1
Brands in staple category	−.125	.042	.113
Brands in commoditized categories	−.191	.103	.093
Baselinea	−.094	.077	.076

^a

Baseline is evaluated at the mean of all independent variables.

Premium global brands—with a high price premium and broad international coverage—show an appreciable increase in overall marketing-mix effectiveness over time. Global brand managers can make a case to maintain or even expand spending on all three levers. This is welcome news for companies such as P&G, Mars, Unilever, and Nestlé: Even in an era of slowing globalization, we see no evidence of a general decline in their ability to influence consumer behavior. High-clout brands—with high market share, strong brand growth, and high own-instrument effectiveness—experience erosion in assortment and distribution responsiveness, while their price effectiveness increases. The falling assortment and distribution elasticities tilt the optimal allocation ratio toward price-related spending. Staple category brands, in markets with high purchase frequency and penetration, benefit from substantial increases in price and nonprice assortment and distribution effectiveness. This pattern is also observed in commoditized category brands, which are characterized by high private label share, heavy promotion pressure, and low innovativeness.

Emerging versus developed markets

The distinction between emerging and developed markets is of considerable managerial importance (Khanna and Palepu 2010; Sheth 2011). Managers are interested in knowing whether there are substantial differences in the moderators of annual change in price, assortment, and distribution effectiveness. To explore this issue, we conducted an additional analysis in which we reestimated Equation 3 separately for emerging and developed markets, allowing us to compare the trajectories and drivers of annual changes across the two groups. Three major takeaways emerge.

First, the convergence effect of own instrument effectiveness is two to three times larger in emerging markets than in developed markets, indicating faster convergence toward the mean in less mature markets. Second, for assortment and distribution, patterns are broadly similar across emerging and developed economies (see Web Appendix I), but that is not the case for price elasticity (Table 13). In emerging markets, premium pricing, high retail concentration, high purchase frequency, and high GDP per capita reduce price sensitivity over time; in developed markets, these same factors increase it.

Table 13.

Regression of Annual Change in Marketing-Mix Elasticities on Predictors.

	Emerging Markets			Developed Markets
	γ	p-Value	Effect Size	γ	p-Value	Effect Size
Intercept	−1.490	.2004		3.816	<.001
Brand Factors
Price positioning (γ1)	.217	<.001	.130	−.087	.002	−.060
Assortment positioning (γ2)	.016	.689	.012	.011	.666	.007
Distribution positioning (γ3)	−.030	.619	−.012	−.022	.495	−.009
Category reach (γ4)	−.029	.003	−.081	.000	.943	−.001
International reach (γ5)	−.171	.307	−.035	−.175	.028	−.042
Local brand (γ6)	−.142	.038	−.068	−.036	.263	−.018
Brand power (γ7)	−1.186	<.001	−.172	−.560	<.001	−.096
Brand growth (γ8)	−.154	.265	−.026	−.910	<.001	−.122
Own effectiveness (γ9)	−.159	<.001	−.335	−.045	<.001	−.126
Competitive vulnerability (γ10)	.026	<.001	.105	.010	.001	.036
Category Factors
E-commerce share (γ11)	−.010	.246	−.045	−.014	.005	−.056
Discounter share (γ12)	.012	.002	.084	.003	.070	.032
Private label share (γ13)	.599	.668	.042	−.228	.671	−.015
Retailer concentration (γ14)	.546	.010	.099	−.275	.042	−.038
Concentration brand (γ15)	.508	.015	.091	.197	.092	.035
Promotion pressure (γ16)	−1.401	<.001	−.162	.064	.689	.006
Innovativeness (γ17)	.002	.131	.065	.005	<.001	.079
Category growth (γ18)	−.164	.122	−.041	−.153	.072	−.030
Purchase frequency (γ19)	.011	.001	.117	−.000	.004	−.019
Penetration (γ20)	−.249	.100	−.071	−.056	.536	−.013
Share of wallet (γ21)	−2.930	.010	−.100	−6.924	<.001	−.128
Country Factors
Market size (log) (γ22)	.006	.859	.009	−.078	.007	−.094
GDP per capita (log) (γ23)	.219	.003	.129	−.133	.007	−.073
GDP growth (γ24)	−.146	.341	−.045	−.898	<.001	−.122
Income inequality (γ25)	.006	.502	.037	.023	.007	.095
Power distance (γ26)	.002	.697	.019	−.001	.210	−.027
Uncertainty avoidance (γ27)	−.010	.001	−.214	.000	.665	.009
MAS (γ28)	−.002	.599	−.024	−.001	.053	−.028

Notes: Effect size is calculated as the parameter estimate multiplied by one standard deviation of the predictor. MAS = motivation toward achievement and success.

Third, certain conditions amplify the trend in price sensitivity only in emerging markets. Umbrella branding and heavy category promotion raise price sensitivity over time substantially in emerging markets but have a negligible effect in developed markets. Similarly, high cultural uncertainty avoidance increases price sensitivity sharply in emerging markets but not in developed markets—perhaps reflecting the greater perceived financial risk of high prices (Kaplan, Szybillo, and Jacoby 1974) in less affluent, risk-averse contexts. Brand concentration has a stronger dampening effect on the trend in price sensitivity in emerging than in developed markets. However, brands in the growth stage of the brand life cycle in developed markets exhibit strongly growing price sensitivity. This is aligned with Simon (1979).

These results confirm that there are managerially meaningful differences between developed and emerging markets, at least for price. Market maturity effects appear to manifest not through static structural indicators alone, but through their interaction with competitive, cultural, and category conditions.

Limitations

A first limitation concerns our sample-selection criteria. By focusing on brands with at least 1% market share and consistent multiyear presence, we necessarily underrepresent small or niche brands; this may bias mean elasticities downward and dampen the observable variance in their evolution. In addition, broader structural changes in categories, such as brand proliferation, SKU rationalization, or shifts in differentiation, may influence elasticity trajectories in ways our large-scale design cannot fully isolate. These issues reflect the inherent trade-off between breadth and depth in empirical generalizations research.

Second, estimating all parameters simultaneously using, for example, a Bayesian dynamic linear model (Ataman, Mela, and Van Heerde 2008) has statistical advantages, but the size of our dataset made this infeasible. However, this is a powerful alternative for smaller data sets.

Third, our moderator analysis is only feasible with this number of brands if we use the same trend specification. We used the linear trend, which is parsimonious, robust, and easily interpretable. However, as is to be expected in large-scale empirical research, the same trend specification may not work equally well for all brands. Future research, studying a small number of brands (see also Parker 1992; Simon 1979) can go deep by allowing a unique change specification for each brand instrument. Linear splines are one option; higher-order polynomials are another. An important outcome of such a study is to see whether commonalities can be identified across brands.

Fourth, we aggregate across consumers. Although this is common practice in this kind of studies (Van Heerde et al. 2013), future research could investigate whether change trajectories differ between segments, using, for example, latent class models (Mela, Gupta, and Lehmann 1997). But these latent segments are likely to be category-specific, so an in-depth analysis should focus on a few categories at best and assess whether common patterns emerge.

Fifth, although we do include retailer variables in our research framework, the analysis could be deepened by looking at differences between different types of retail formats, such as hypermarkets, supermarkets, discounters, and online. Due to data sparseness, such an analysis requires a country with a very large panel.

Sixth, despite our best efforts, we could not secure advertising data for our set of countries. Although the effect of advertising is typically much smaller than the effect of other marketing-mix instruments (Ataman, Van Heerde, and Mela 2010; Shapiro, Hitsch, and Tuchman 2021), it remains among the most visible and most debated marketing instruments. We show for the United States that not including advertising resulted in minimal bias in the price, assortment, and distribution elasticities, a result that replicated and confirmed a similar analysis conducted by Datta et al. (2022) for other countries (China and Hong Kong) and another industry (consumer durables). Nevertheless, future research could estimate the evolution in brand advertising elasticities for a smaller set of countries.

Seventh, our study focuses on long-term trends. There are other features of elasticity trajectories that are of interest, such as variability, cyclicality, and structural breaks. This requires an additional analytical layer—each requiring its own modeling framework, interpretation, and heterogeneity analysis. Future research could explore these important issues.

Finally, future research could also go beyond the “traditional” set of instruments and study the effects and the temporal dynamics of different marketing tools that are aligned with rapidly transforming environments such as promotions targeting microsegments, and speedy deliveries from fulfillment centers that take advantage of modern digital and mobile technologies.

Supplemental Material

sj-pdf-1-jmx-10.1177_00222429251412261 - Supplemental material for An Investigation into the Evolution of Marketing-Mix Effectiveness: An Empirics-First Approach

Supplemental material, sj-pdf-1-jmx-10.1177_00222429251412261 for An Investigation into the Evolution of Marketing-Mix Effectiveness: An Empirics-First Approach by Katrijn Gielens and Jan-Benedict E.M. Steenkamp in Journal of Marketing