Second Brain

Marketing Generalizations Overview

3 min read

Marketing Generalizations Overview

Summary

Chapter 8 synthesizes empirical findings into marketing generalizations — regularities in market response relationships that hold across multiple studies and conditions. Covers the criteria for a good generalization, three methods for discovering them (informal observation, literature review, meta-analysis), and the cautionary note about measurement error.

What is an Empirical Marketing Generalization?

Marketing Generalization

A marketing generalization is an approximate, quantitative summary of market response regularities that:

  1. Is based on repeated empirical evidence across multiple studies or datasets
  2. Has sufficient scope to be meaningful across conditions
  3. Is stated with precision (quantitative range, not just direction)
  4. Is parsimonious (simple formulation)
  5. Is useful for practitioners
  6. Has a theoretical link (explains the mechanism, not just the pattern)

The measurement-error caveat (Morrison & Silva-Risso 1995): Observed value = True Score + Error (Eq 8.11), so generalizations should ideally be based on latent true scores.

Three Methods for Discovering Generalizations

1. Informal Observation

An experienced researcher identifies patterns across studies they know well. Qualitative; subject to availability bias and selective recall.

2. Literature Review

Narrative review: synthesize all relevant studies at face value. Breaks down with as few as 7 studies (information processing limits).

Voting method: count positive-significant, negative-significant, and null results across studies. More systematic but misses effect-size information.

3. Meta-Analysis

Meta-Analysis

Statistical analysis of effect sizes (elasticities) across studies. Key features:

  • Dependent variable must be dimensionless (elasticities satisfy this)
  • Can estimate mean elasticity and its variance across studies
  • Can identify systematic moderators (brand size, product category, data frequency)

Marketing meta-analyses: Aaker & Carman (1982) on advertising; Assmus, Farley & Lehmann (1984) on advertising; Tellis (1988) on price; Sethuraman, Srinivasan & Kim (1999) on price cross-effects.

Caveat: model structures may generalize even when parameters do not.

Key Null Hypothesis Values for Marketing Models

Based on meta-analytic findings, recommended null hypothesis starting values (prior means) for future models:

ParameterRecommended NullSource
Short-run advertising elasticity0.10–0.22Assmus et al. (1984), Sethuraman & Tellis (1991)
Price own-elasticity−2.5Tellis (1988)
Advertising retention rate (monthly, raw)0.43–0.50Clarke (1976), Assmus et al. (1984)
Advertising retention rate (adjusted for aggregation)0.69–0.775Leone (1995)

Schultz-Wittink Decomposition

Primary vs. Selective Demand

A brand’s sales effect can be decomposed into:

  • Primary demand effect: brand advertising grows the total category (industry sales elasticity > 0)
  • Selective demand effect: brand advertising takes market share from rivals (market share elasticity > 0)

Table 8-2: six cases based on the combination of market share elasticity, primary demand elasticity, and rival sales cross-elasticity sign:

  • Case I: Primary demand only (Cereals: primary elasticity 0.029)
  • Case III: Competitive only (Cigars: market share 0.076, rival cross-elasticity −0.079)
  • Case IV: Both primary and primary sales (Deodorants)

The Measurement Error Warning

Because all elasticity estimates contain sampling error, the search for generalizations must account for the framework (Eq 8.11-8.12):

Meta-analytic means are biased if studies systematically use the same misspecified model. Good practice: weight studies by sample size and correct for known biases (temporal aggregation, endogeneity).

Graph View

Backlinks