Reference stations · 4 live

Controls

How We Know · Every number live from the database

Live Monitor

–

listener-count readings

–

plays analyzed

–

unique tracks measured

Lifetime totals from the measurement history currently retained in Airplay Live · updating live

Two Halves, One Answer

It could be noise. Noise would not repeat.

The Big 80s · … tracks · plays split at random

Split-half agreement (r)

The Big 80s0.39

Dance UK0.10

Box UK0.02

r = 0.39 across … tracks

Each dot is one track scored twice, from two random halves of its own plays with no shared data between them. The halves agree far above chance, and the agreement matches what the variance model predicted before we measured it.

Box UK is the built-in control: that catalog carries almost no repeatable track-to-track difference in listener hold, and the meter reports the zero instead of inventing a signal.

Off the Clock

The hour a track is given should not decide its score.

Schedule bias · before and after the ambient correction

Listeners tune out faster at some hours, so a raw drop count partly tracks the schedule instead of the track. After the ambient hour is subtracted, that link is gone: time of day now accounts for 0.068% of the difference between plays.

When It Says No

A meter that finds a difference everywhere finds nothing.

Genre pairs across four stations · Appeal difference ±95 % CI

Pairs that should differ do: Eurodance against Hip-Hop, Pop against Soul. Pairs that should not, do not: Pop against Disco, Pop against New Wave both cross zero.

One Ballad Every Hour

A test where every radio person already knows the answer.

A deliberate off-format experiment · dance stream, summer 2025 · frozen archive

One slow ballad closed every hour on a dance stream. The group was defined by the schedule, not by the meter, and the meter was told nothing about it. Measured raw, with no smoothing, the ballads land about 8 points below the rest of the catalog on average (34 ballads with five or more measured plays, out of 87 aired), at about twenty concurrent listeners.

Their slot is not to blame: ambient turnover at the end of the hour measures the same as the rest of the day. The in-format genres are indistinguishable from each other on this scale; only the schedule-defined outsiders drop.

What It Does Not Measure

Honest numbers include the ones that limit us.

What we measure is the station's own online stream, read as a census, not a sample: every listener on the stream, about every five seconds. In the US, about one in seven hours of AM/FM listening now comes through a stream (Edison Share of Ear: 14% of AM/FM listening time, Q1 2026), and Nielsen already credits that stream inside a station's ratings, as one line with the over-the-air signal. The stream audience is the station's same format core, two thirds of streamers also listen over the air, but it is a skewed slice of it: more male, more upscale, more at-work and at-home than the full broadcast audience.

So read the score as a fast leading indicator from a station's most engaged listeners, not as a stand-in for callout or panel ratings, not as a read on the in-car audience, and not as a claim about listeners who never touch the stream. Where those matter, weight it.

Three more things this system does not do. It reads only what airs: it can say nothing about a track the station is not playing. It measures decisions, not feelings: a listener who stays counts the same whether the track moved them or simply did not move them out. And it has not measured burn: rotation on our reference stations never gets heavy enough. The most-played music track airs about 13 times a week, and hearing it even once would take about 13 hours of weekly listening. We tested for it there and found nothing to sell.

Live data · based on 11.2M five-second observations across every station measured.

Dataset updated August 9, 2026 at 1:43 PM UTC