Guides · Replication and generalization

Empirical generalizations in marketing: how to build on them

An empirical generalization is a regularity that recurs across documented conditions and can be stated simply, often as a number such as an average elasticity. Marketing's generalizations are collected in special issues, meta-analyses and an MSI volume. A cumulative study builds on one by testing it under conditions nobody has checked, finding where it breaks down, or updating it with newer data, and reports the conditions rather than a single average.

What counts as an empirical generalization

The standard marketing definition, from a 1995 Marketing Science special issue, describes a pattern that repeats across different circumstances and can be stated simply in mathematical, graphic or symbolic form. The definition does not require the pattern to be causal, universal or governed by identical parameters everywhere, and its author meant it to exclude purely verbal statements such as y increases with x unless they are quantified.^[1]

Advertising researchers later sharpened the idea: a generalization is a result that holds over a documented range of conditions, or varies in a known way across them, much as the boiling point of water shifts predictably with air pressure.^[2]

On this view a generalization is judged by its empirical content, meaning the range of circumstances under which the same result is known to hold, and not by whether a theory already explains it. Lack of theory limits its usefulness but not its validity within the conditions covered.^[3]

Where established generalizations are collected

• The MSI volume Empirical Generalizations about Marketing Impact, first published in 2009 with one-page summaries from about 65 authors; its 2015 second edition held about 120 generalizations.^[4]
• Meta-analyses that quantify marketing-mix effects, for example an average price elasticity of about -2.62 across 1,851 estimates from 81 studies, and average sales call and advertising elasticities of 0.31 and 0.12.^[5][4]
• A 2009 Journal of Advertising Research special issue from a Wharton conference, which selected 23 advertising generalizations and surveyed 80 academics and practitioners on how important and how certain each one was.^[6]
• A 2013 follow-up section from a second conference, which lists each proposed law with its documented conditions and possible boundaries, such as brand users recalling brands about 1.7 times more than non-users across more than 600 surveys and 210 categories.^[2]
• The 1995 Marketing Science special issue from the first Wharton conference, which set out to list marketing's generalizations; examples discussed include the adoption curve of new technologies and the negative binomial distribution of purchase occasions.^[1]

Finding the gap: conditions nobody has checked

Most established generalizations are thin on conditions. The organizers of the advertising conference concluded that many of their laws lacked knowledge of where they do and do not hold, and participants named missing conditions such as competitive context, product and service categories, media types, emerging markets and the economic climate.^[6]

They pointed researchers to three kinds of gap: generalizations that are important but rest on few studies or lack independent confirmation, important and well-supported ones missing key conditions, and important areas covered by few generalizations at all.^[6]

A JAMS editorial on the topic sets a similar agenda: sharpen existing generalizations by establishing moderators that push an effect above or below its average, extend them into digital marketing, where the knowledge base is described as minimal, and cover outcomes for other stakeholders such as employees and investors.^[4]

Meta-analyses often name the studies still needed. The price elasticity meta-analysis found that sales elasticities had grown in magnitude over four decades while share and choice elasticities stayed fairly constant, and it closed by identifying which new elasticity studies would be most valuable.^[5]

Three ways to build on a generalization

• Extend it to a new domain or market. One study fitted the negative binomial model of purchase frequency to leisure-time physical activity with data from three countries and three years, and it described the behavior well; another found the brand loyalty patterns known from Western packaged goods, including double jeopardy, in two Chinese cities over five years, alongside a local shift from exclusive to multi-brand loyalty.^[7][8]
• Find its boundary. A test of the double jeopardy line across 37 categories and 40 quarters found that a straight line fits under limited conditions but breaks down when the full range of brands is included, where the Dirichlet model or an exponential curve fits the data better.^[9]
• Update or revise it with newer data. The price elasticity meta-analysis revisited an earlier synthesis with far more estimates and found, among other things, that correcting for price endogeneity makes elasticities larger while modeling consumer heterogeneity has no significant effect on them.^[5]
• Differentiated replication serves all three purposes at once: it tests whether the result holds under markedly different conditions, extends its practical scope, and searches for the exceptions where it does not hold.^[10]

Design for many sets of data

Pursuing generalizations changes study design. Instead of one data set analyzed for significance or best fit, the aim is many deliberately different data sets and a check on whether a simple model found earlier holds again in each. One attitudinal regularity was tested this way across about 100 data sets, 9 product categories and two countries.^[3]

Varying several conditions together is acceptable for this purpose, because the aim is to show the result survives the differences. Planned systematic replications can also reveal small but consistent deviations; one covering double jeopardy across 34 US categories found no systematic bias in the earlier model while setting clearer limits on its irregularities.^[10]

When only one large data set is available, divide it into subsets that represent different conditions of interest, such as media, categories, seasons, countries or brand sizes, rather than reporting a single pooled result.^[2]

Replication can be built into a single paper. The physical activity study replicated its analysis across activity intensities, countries and years, while acknowledging that inconsistent survey measures limited how widely it could test the model.^[7]

Report conditions, not just an average

Advertising researchers have criticized marketing studies for presenting a single finding, or even a meta-analytic average, as a generalization. They cite a correlation pooled from 324 studies covering very different conditions, which says little about where it applies, and they also warn against leaning on significance tests in one small data set.^[2]

Their standard is to document as many conditions as possible under which a law holds and which conditions alter it, ideally with results repeated by different researchers using transparent data and analyses, and with simple analyses preferred over complex ones.^[2]

Quantified magnitudes make generalizations usable. Knowing only the direction of an effect cannot tell a manager whether an action will pay off, so generalizations are most useful stated as magnitudes such as elasticities, a unit-free measure that is comparable across studies and can feed the financial return metrics senior management asks for.^[4]

Be aware that much of the work above, including the advertising conference series and the double jeopardy and physical activity studies, comes from researchers at or working with the Ehrenberg-Bass Institute. The meta-analytic generalizations come from other groups, and generalizations that lack confirmation by independent researchers are named as a research priority, so independent tests are themselves a contribution.^{[6][2][7][9][5]}

Where theory comes in

A generalization can come before any theory that explains it, or be predicted by one, and the interplay between the two is what leads to higher-level theory.^[1]

Looking for the regularity first often works better in practice: if the theorizing later proves wrong, the regularity remains, and knowing it narrows the range of theories worth considering. The Dirichlet model of buying behavior is presented as a case where many separate regularities were later subsumed in one theory, though it still cannot say why one brand is bigger than another.^[3]

A 2022 Journal of Marketing article makes the broader case for such empirics-first research: work that starts from a real-world phenomenon, analyzes data and produces valid insight without necessarily testing theory, and that can serve as a stepping-stone to theory.^[11]

Consumer research is moving toward the same position, with a 2024 review arguing that routine replication, including replication planned alongside meta-analysis, should be an accepted prerequisite for claiming an empirical generalization.^[12]

The evidence behind this

• Empirical generalizations in marketing: what the research shows
• NBD-Dirichlet model: what the research shows
• Double jeopardy: what the research shows
• Laws of buyer behaviour: what repeat-buying research shows

References

1. Frank M. Bass (1995). Empirical Generalizations and Marketing Science: A Personal View. Marketing Science.
2. Yoram (Jerry) Wind, Byron Sharp, Karen Nelson-Field (2013). Empirical Generalizations: New Laws for Digital Marketing — How Advertising Research Must Change. Journal of Advertising Research.
3. A. S. C. Ehrenberg (1995). Empirical Generalisations, Theory, and Method. Marketing Science.
4. Dominique M. Hanssens (2018). The value of empirical generalizations in marketing. Journal of the Academy of Marketing Science.
5. Tammo H. A. Bijmolt, Harald J. van Heerde, Rik G. M. Pieters (2005). New Empirical Generalizations on the Determinants of Price Elasticity. Journal of Marketing Research.
6. Yoram Wind, Byron Sharp (2009). Advertising Empirical Generalizations: Implications for Research and Action. Journal of Advertising Research.
7. Amy L. Wilson, Byron Sharp, Cathy Nguyen, Svetlana Bogomolova (2017). Expanding Marketing Empirical Generalisations to Health Behaviours: Physical Activity Is Not So Different From Buying Behaviour, After-All. Australasian Marketing Journal.
8. Mark D. Uncles, Chao Wang, Simon Kwok (2010). A Temporal Analysis of Behavioural Brand Loyalty among Urban Chinese Consumers. Journal of Marketing Management.
9. Cullen Habel, Larry Lockshin (2013). Realizing the value of extensive replication: A theoretically robust portrayal of double jeopardy. Journal of Business Research.
10. R. Murray Lindsay, Andrew S. C. Ehrenberg (1993). The Design of Replicated Studies. The American Statistician.
11. Peter N. Golder, Marnik G. Dekimpe, Jake T. An, Harald J. van Heerde, Darren S. U. Kim, Joseph W. Alba (2022). Learning from Data: An Empirics-First Approach to Relevant Knowledge Generation. Journal of Marketing.
12. Oleg Urminsky, Berkeley J Dietvorst (2024). Taking the Full Measure: Integrating Replication into Research Practice to Assess Generalizability. Journal of Consumer Research.

Bibliographic details only. Follow the DOI to read each study at its publisher.