What is a DMA?

A Designated Market Area is Nielsen's partition of the United States into 210 television markets, each a group of counties receiving the same broadcast signals. The definition is commercial rather than governmental — it exists because advertisers need to know who a station reaches — and that origin is exactly why it has become the default geo unit for media measurement.

The reason it works so well is alignment. Television and radio are bought by DMA, so a campaign switched on or off at DMA level is switched on or off along the same boundaries the media plan already uses. That removes the leakage problem that finer units create: within a DMA everyone gets the same television, and across DMAs they do not. A geo test at postcode level fighting a DMA-level buy is contaminated by construction.

They are extremely uneven, which is the main practical complication. New York covers over seven million households and the smallest markets cover fewer than fifty thousand, so a handful of large DMAs can dominate a national outcome and effectively shrink the sample far below 210. Stratifying by size before assignment, or weighting the analysis, is standard practice for that reason — and checking the concentration of volume across markets should precede any power calculation.

They are also a television construct being used for everything else. Digital campaigns can target far more precisely, e-commerce demand does not follow broadcast footprints, and a DMA can span state lines with different tax and shipping rules. Where a campaign is purely digital, a DMA may be needlessly coarse — the reason to keep using it is usually that other channels in the mix are not.

Outside the United States there is no direct equivalent and the substitutes are worse. The UK has ITV regions, most of Europe uses administrative regions such as NUTS, and elsewhere teams fall back on cities or states. None was designed around media footprints, so cross-boundary leakage is generally larger and matching markets on pre-period behaviour matters correspondingly more.

A national advertiser plans a DMA-level test and checks whether 210 markets really provide 210 units of information, by looking at how its own conversion volume is distributed across them.

The gap between 148 and 62 is what a raw unit count hides. Five markets carrying 31% of conversions means those five largely determine the national estimate, and the hundred smallest contribute 9% between them — they are units in name and contribute little information. The detectable effect moves from 5.4% to 8.3% once that is accounted for, which is the difference between a test that can resolve a typical campaign effect and one that cannot. Two responses follow. Stratify by size so both arms get a comparable mix of large markets, since a random split that put four of the top five in one arm would be badly unbalanced. And consider whether the smallest markets should be excluded entirely rather than adding noise — dropping the bottom 50 costs 3% of volume and removes a great deal of variance.