Case Study: Market Concentration Metrics

The executive question: how concentrated is market voice across industries?

This case uses the same parquet file behind the Industry Ad Spend Explorer, but the metric question is different. The studio asks how advertising budgets moved through the Covid shock. Here the question is cross-sectional: which source-defined markets have concentrated advertising voice, and how sensitive is that answer to the definition of "market" and "firm"?

The economic intuition comes from industrial organization. Market power is not observed directly in a descriptive ad-spend table. Instead, the analyst chooses a market boundary, chooses a firm boundary, converts dollars into shares, and summarizes the share distribution. That is why concentration metrics are excellent teaching examples for a visualization chapter: the chart is only as honest as the metric definition underneath it.

The case borrows reference thresholds from the DOJ/FTC 2023 Merger Guidelines, which describe HHI above 1,000 as concentrated and HHI above 1,800 as highly concentrated. Those lines are used here as visual benchmarks, not as legal conclusions. The same guidelines also emphasize that market shares and concentration are most probative when calculated in a properly identified market; a broad ad-spend industry is a useful screen, not a final relevant market.

The metric contract

The source table has two hierarchies that matter. On the firm side it includes PARENT, SUBSIDIARY, ADVERTISER, BRAND, and PRODUCT. On the market side it includes Industry_Group, INDUSTRY, MAJOR, CATEGORY, SUBCATEGORY, and MICROCATEGORY. The headline firm unit in this case is an owner proxy: use PARENT when it is known, and use ADVERTISER when PARENT is the pooled PARENT UNKNOWN bucket. That rule avoids treating many unrelated unknown-parent advertisers as one giant firm.

The default market proxy is INDUSTRY, not the broader Industry_Group. That gives 61 source-defined industries, versus 25 broad groups, 292 major buckets, 928 categories, and 1,695 subcategories. The share variable is total positive dollar_spent from January 2018 through December 2022. The data contains 2.98 million positive spend rows over 60 months and about $369B in measured spend. The metrics describe share of advertising voice, not share of sales, prescriptions, patients, or units.

The HHI formula squares shares before summing them. That is the reason a market with one 70% firm looks very different from a market with four roughly equal leaders, even if both have visible top-four concentration. The effective-entities transformation, 10,000 / HHI, gives a useful intuition: household soaps, cleansers, and polishes has an HHI of about 4,992, equivalent to roughly 2.0 equal-sized firms.

Market definition changes the answer

The broadest market definition is too coarse for the headline. Industry_Group pools many distinct product and service markets, so it makes the economy look mostly fragmented: only 2 of 25 broad groups are highly concentrated, and only 11.3% of spend lies in markets with HHI above 1,000. At INDUSTRY, the median HHI rises to about 1,018 and 36.0% of spend lies in concentrated markets. At MAJOR, CATEGORY, and SUBCATEGORY, concentration becomes the normal pattern because the buckets approach product categories rather than industries.

This is why the rest of the case uses INDUSTRY as the default. It is narrow enough to avoid the worst pooling of the 25 broad groups, but still broad enough to support a cross-industry comparison without making thousands of tiny categories carry the main story. The narrower levels become sensitivity checks and targeted drilldowns, especially for regulated categories such as prescription drugs.

Concentration is rare, but top-four voice is common

Figure 3 ranks substantial INDUSTRY markets by owner-proxy HHI. A spend floor is used only for display so tiny source buckets do not dominate the rank chart. The most concentrated large markets include household soaps, cleansers, and polishes; discount department and variety stores; cigarettes and tobacco; department stores; and men's toiletries and skin care. Several are not monopoly-like by CR1, but they have extremely high CR4. Discount department and variety stores is the clearest example: Walmart's leader share is 35.3%, while the top four owners account for 99.2% of measured voice.

Figure 4 shows why CR1, CR4, and HHI should be read together. Household soaps is a dominant-leader case: CR1 is 69.5% and CR4 is 89.2%. Discount department and variety stores is a top-four case: the leader share is lower, but CR4 is 99.2%. Medicines and proprietary remedies is the opposite: it has very large spend, but broad ownership and a low HHI at the INDUSTRY level.

The FDA-style surveillance implication is practical rather than legal. If a regulated claims team wants broad coverage of a concentrated prescription category, a small set of owners may cover a large share of public-facing spend. If the same team wants broad coverage of a fragmented category, owner-level sampling is a long-tail problem. For prescription drugs, even the INDUSTRY level can still be too coarse; the same logic should be repeated by condition, product class, channel, and claim type.

The FDA-relevant drilldown is narrower than industry

The pharmaceuticals example shows why market definition cannot be delegated to a field name. Holding Industry_Group = Pharmaceuticals, the INDUSTRY level is not concentrated: medicines and proprietary remedies has HHI around 664, and pharmaceutical houses has HHI around 717. But the prescription subcategories are a different story. Diabetes/endocrine medications, arthritis medications, respiratory disorder medications, hematology/immunology medications, and heart disorder medications all cross the high-concentration HHI benchmark in this ad-spend proxy.

The firm boundary is not a detail

The largest methodological risk is entity definition. The same underlying rows can be summarized by parent owner, advertiser, or brand. Those definitions are not interchangeable. Parent-level aggregation is closest to a firm-level IO question because it groups multiple brands and advertiser labels under common ownership. Advertiser and brand definitions are useful for marketing execution, but they split corporate families and usually make concentration look lower.

The biggest example is household soaps, cleansers, and polishes: owner-proxy HHI is about 4,992, while advertiser-level HHI is about 1,008 and brand-level HHI is about 353. That is not a contradiction. It is the measurement hierarchy doing exactly what it is supposed to do. If the question is corporate market power or coordinated control of advertising strategy, split brands are too granular. If the question is creative execution or consumer-facing brand clutter, brand-level concentration may be the right object.

Thresholds should be shown as sensitivity, not hidden as cleaning

The user-facing problem with this dataset is the micro-entity tail. The median owner-by-INDUSTRY entity spends only about $4k over the full five-year window, while the 99th percentile spends about $8.4M. It is reasonable to ask whether tiny positive spend entities should be removed before computing concentration. But the choice is not free: dropping the tail and renormalizing retained entities mechanically raises HHI.

The main metric therefore keeps all positive entity spend. Figure 7 treats thresholds as a sensitivity check. The $1M cutoff barely changes household soaps or medicines and proprietary remedies because each still retains 99.6% of spend. It changes miscellaneous services and amusements more because the cutoff retains only 87.0% of spend. In that setting, the higher HHI is partly a definition artifact.

The rule for practice is simple: use all positive spend as the default denominator, then report threshold sensitivity with retained spend. If the HHI ranking is stable and retained spend is high, the conclusion is robust to de minimis noise. If the cutoff drops a large share of spend, treat the recomputed metric as an upper-bound stress test, not a replacement for the main estimate.

Concentration moved through the shock, but not in one direction

The 2018 to 2022 comparison is descriptive, not causal. Still, it shows a useful pattern: concentration can rise even as category spend falls. Household soaps became much more concentrated from 2018 to 2022 while measured spend fell. Department stores and miscellaneous merchandise also became more concentrated. Gasoline, lubricants, and fuels moved the other way: HHI fell sharply as spend fell, so decline alone does not imply consolidation.

For surveillance, this kind of chart is a triage tool. A category with rising concentration may deserve a different sampling plan, especially if the leader changes, if spend shifts toward television or digital, or if the industry is already in a regulated claims environment. But the chart does not say why concentration changed. It could reflect entry, exit, media reallocation, product launches, missing ownership detail, or demand shocks.

What the FDA perspective adds

FDA does not need this analysis to decide whether an ad is misleading. That requires claim-level evidence, labeling, context, and applicable rules. The Office of Prescription Drug Promotion focuses on whether prescription drug information is communicated truthfully and with balance. FDA's direct-to-consumer TV and radio rule guidance also emphasizes that major risk statements must be presented in a clear, conspicuous, and neutral manner.

Concentration metrics add a different layer: they help an analyst decide where the public message environment is structurally dominated. In a concentrated voice market, a few sponsors can shape a large share of consumer exposure. In a fragmented market, risk comes less from one owner's dominance and more from coverage, sampling, and heterogeneous claims.

The pharmaceutical finding is therefore two-sided. At the broad group and INDUSTRY levels, pharmaceutical ad spend is not highly concentrated by HHI. At prescription subcategory level, several high-spend markets are highly concentrated. That does not settle a regulatory question, but it tells the analyst where a claim-level review plan should become more targeted: therapeutic area, condition, prescription status, DTC versus professional promotion, media group, and possibly drug class. The visualization lesson is that the first chart should expose that next question rather than bury it.

Method notes and robustness

The calculations use the following choices:

1. Keep all rows with positive dollar_spent through December 2022.
2. Define the headline firm unit as PARENT unless the parent is PARENT UNKNOWN, in which case use ADVERTISER.
3. Aggregate to owner-by-INDUSTRY total spend over 2018-2022.
4. Convert each owner's spend into an industry share.
5. Compute CR1, CR4, CR8, HHI, and effective entities from that share vector.
6. Recompute market-definition sensitivity across Industry_Group, INDUSTRY, MAJOR, CATEGORY, and SUBCATEGORY.
7. Recompute minimum-spend sensitivity at owner-spend cutoffs of $10k, $100k, and $1M, always reporting retained spend.

The robustness checks support the main descriptive pattern. The highest-concentration substantial industries remain high under minimum-spend thresholds because the retained spend share is near complete. Long-tail industries remain low-HHI even after aggressive cutoffs, but those cutoffs drop more denominator and should be read as stress tests. Market-definition sensitivity is larger than threshold sensitivity: moving from broad groups to narrower source categories changes the substantive answer more than dropping de minimis spend does.