From radio diaries to Chartbeat dashboards. Why the method always shapes the count.
Unit C: How do we research and monitor?
If nobody can actually see the audience, how does any media organisation decide what to make, who to sell it to, and how much to charge?
The answer is: they count a proxy. And whoever controls the counting controls the currency of the entire media market.
By the end of today
What you will walk out knowing
01
Trace the measurement arc
Follow audience measurement from telephone coincidentals and paper diaries through the people meter and portable people meter to real-time web analytics and social media metrics.
Reveal
02
Distinguish the three audiences
Explain the difference between the predicted, measured, and delivered audience, and why each gap carries commercial risk.
Reveal
03
Read metrics critically
Identify the difference between reach metrics, frequency metrics, and engagement metrics, and know when each is appropriate or misleading.
Reveal
04
Apply to Bangladesh and news
Show how a tool like Chartbeat changes editorial decisions, and discuss what equivalent pressures look like in a Bangladeshi newsroom context.
Reveal
Why measurement matters
The audience no one can see
Measurement is not the audience. A rating is not people watching. It is an estimate derived from a sample that everyone agrees to treat as if it were real. This agreement is what makes it a currency.
Currency. Ratings became the agreed-upon medium of exchange between broadcasters and advertisers. Both parties accept the number as the basis for transactions, even though both know it is an approximation.
The gap problem. Every measurement method leaves gaps: who counts, who is missed, what counts as "watching." Those gaps are never neutral. They systematically advantage some content and some audiences over others.
Power follows the count. Whoever defines and controls measurement has enormous influence over which content survives. A measurement system that undercounts minority audiences effectively erases their commercial value.
In Bangladesh: no independent local television-rating currency existed for most of commercial broadcasting. Ad spend was allocated on relationships, audience guesswork, and imported international proxies rather than local measurement.
Drag to scrub
Drag to scrub
A century of counting
A critical distinction
Three audiences in every transaction
📈
Predicted audience
The audience the firm and its advertisers expect to reach when they make commitments. Derived from forecasts and historical patterns. What the parties think will exist.
📊
Measured audience
The audience the measurement system actually records as having been exposed to content. The agreed-upon currency. Rarely identical to the predicted audience in either size or composition.
🎯
Delivered audience
The audience the advertiser actually wanted: people who paid attention, remembered, and were influenced. Almost always smaller than the measured audience, sometimes drastically so.
The media economy is moving from pricing by measured audiences toward pricing by delivered outcomes: clicks, conversions, attributable revenue.
Buzzard · Nielsen · Arbitron
How the ratings currency was built
The telephone coincidental (1930s). Crossley called listeners immediately after a broadcast and asked what they were listening to. Fast, but reached only telephone subscribers. Hooperatings refined this to a dominant method.
The paper diary (1940s–80s). Households recorded their own viewing. Cheap to deploy at scale, but dependent on respondent memory, honesty, and effort. Zero-cell problem: small audiences for niche content simply vanished from the count.
The Audimeter (1936). Nielsen's device attached to a radio, then a TV set, and recorded which channel was tuned. Passive: the household did not need to do anything. But it measured the set, not the person.
The people meter (1987). Viewers pressed buttons on a remote-like device to record their age and gender while watching. Demographic data per viewer, not per household. But required active participation, reintroducing the active-viewer error.
The portable people meter (PPM). A pager-like device carried by panel members detected encoded audio signals from broadcasters. Passive exposure tracking. But the PPM systematically undercounted minority audiences, leading to Congressional hearings.
Nielsen's genius was not the technology. It was the social agreement: broadcasters, advertisers, and agencies all agreed to treat the sample as if it were the truth. The rating is a currency, not a measurement.
The structural limits of counting
Sampling, representativeness, and who gets missed
Every rating depends on a panel: a small group whose behaviour is projected to represent a much larger population. The quality of that projection defines the quality of the currency.
01
The sample size problem
Nielsen's national panel covered roughly 5,000 households to represent the entire US. Small samples have wide confidence intervals: a niche programme serving 500,000 viewers might statistically not exist.
Reveal
02
Demographic ascription
Early systems assigned viewers demographic characteristics based on household profiles rather than individual measurement. A household with registered members in certain age-gender cells was credited with those viewers whether they watched or not.
Reveal
03
Minority undercounting
The PPM consistently undercounted minority audiences who listened to niche stations. Congressional hearings followed. The measurement method was not neutral: it systematically disadvantaged content serving minority communities.
Reveal
04
The zero-cell problem
Diary-based systems produced zero-cell results: if no diary respondent watched a show, the show received a zero rating even if many people actually did watch. Small audiences were literally invisible.
Reveal
05
Active-viewer error
Any system requiring participation overestimates engagement. A viewer who forgets to press the people meter button is recorded as absent even if they watched the whole programme.
Reveal
06
Panel recruitment bias
People who agree to be on a measurement panel are systematically different from those who do not. Early adopters of measurement technology, or those with more time to participate, skew every panel.
Reveal
From panels to census data
The digital measurement revolution
The unique visitor (1996). PC Meter established the thirty-day unique visitor as the first standard web metric. Chosen partly because it was "the biggest number we had" and could be compared to newspaper circulation.
Impressions and clicks. Banner ads introduced the impression (a page load) and the click-through. For the first time, advertisers could see whether anyone acted on their advertising, not just whether anyone was exposed to it.
The hybrid model. ComScore and Nielsen NetRatings merged panel-based demographic data with census-level web analytics, producing richer profiles than either could alone.
The engagement turn. After the dotcom crash, advertisers rejected pure reach as insufficient. Performance metrics (cost per click, cost per conversion) and engagement metrics (dwell time, scroll depth, recirculation) began displacing impression counts.
Real-time measurement. Chartbeat's dashboard showed editorial staff a live count of concurrent visitors, traffic sources, and engagement time. Measurement had entered the newsroom in real time.
The shift from monthly circulation audits to real-time page-view dashboards did not just add speed. It changed who the data was for. Circulation figures were produced for advertisers; live dashboards were built for editors.
Sorting game
Sorting game
Which metric family does it belong to?
Metrics to sort
Carlson · Measurable journalism
When the dashboard walks into the newsroom
01
The buyer group vs the user group
Chartbeat had to sell to two groups: senior editors and business staff who paid for the subscription, and journalists who would actually look at the dashboard. Their interests were not identical. Journalists who resented the tool could cause client churn.
Reveal
02
The "broken dial" moment
When traffic surged past the measured cap, the Chartbeat dial "broke" in a way the company called "fun." The euphoria of the broken dial bonded journalists to the tool emotionally. The analytics company was engineering its own stickiness.
Reveal
03
Habit-forming by design
Like the slot machine, the Chartbeat dashboard was designed to be compulsively watched. Foer described checking it before getting out of bed. "Stats-addicted editors" suffered from Chartbeat withdrawal during outages.
Reveal
04
The public-interest tension
Chartbeat eventually moved toward "engagement metrics" (Engaged Time, Recirculation, Visitor Frequency) and aligned itself publicly with journalistic values. But the structural pressure remained: engaging content and public-interest content are not always the same.
Reveal
Spot the misleading chart
Spot the misleading chart
Not all numbers tell the truth
Three charts about the same audience data. Click each to see what it is actually showing, and whether it is honest.
Chart A · Pageviews this month
Chart B · Unique visitors (30-day)
Chart C · Average engaged time (minutes)
Same organisation, same five months. Three stories, only one of which matters most for a public-interest editor.
How measurement works across five sectors
The currency varies by sector
Every sector has agreed on different metrics as its currency. The agreement is partly technical and partly social — it reflects what each industry values and what it can sell.
01
News
Pageviews were the original digital currency; engagement time and recirculation are replacing them. In Bangladesh, digital ad spend follows Facebook's own metrics more than any independent measurement.
Reveal
02
OTT
Netflix measures completions, not views. A "view" in Netflix's self-reported data means two minutes watched. Completion rates, binge rate, and abandonment point are the operational metrics. These numbers are proprietary and unverifiable.
Reveal
03
Games
DAU, MAU, and the DAU/MAU ratio (stickiness) are the primary engagement currency. Session length, retention at day 1, day 7, and day 30, and ARPDAU. All census-level: every player's behaviour is tracked.
Reveal
04
Music
Spotify measures streams (over 30 seconds), saves, playlist additions, skip rate, and listener-to-stream ratio. Skip rate is the clearest signal of audience rejection. These metrics now directly influence algorithmic playlist placement.
Reveal
05
Creators
YouTube Studio gives creators watch time, impressions, click-through rate, and average view duration. The algorithm weights watch time above views. A short video watched completely beats a long video abandoned at 20%.
Reveal
06
Bangladesh context
No independent digital audience measurement currency exists for Bangladesh's digital market. Publishers, advertisers, and agencies each use different metrics. The absence of an agreed currency inflates the power of whoever controls the platform analytics — primarily Meta and Google.
Reveal
Knowledge check
A Bangladeshi digital news outlet reports 5 million pageviews and 800,000 unique visitors. An advertiser asks: "How many people actually read our ad?" Which audience type does the advertiser's question refer to?
C. The 5 million pageviews and 800,000 unique visitors are both forms of the measured audience. The advertiser is asking about the delivered audience: the much smaller group who actually paid attention to the ad and were meaningfully influenced by it.
Nielsen's people meter solved the problem of measuring individual viewers rather than households by requiring no active participation from the viewer.False. The people meter did measure individual viewers rather than households, but it required active participation: each viewer had to press a button to record their presence. This reintroduced the active-viewer error the audimeter had eliminated.
A measurement system that undercounts minority audiences actively reduces the commercial value attributed to content serving those audiences.True. If a measurement system cannot count your audience, advertisers will not pay to reach it. This is precisely the critique that minority interest groups raised against Arbitron's PPM.
Your turn · 15–20 min
Activity: Count the Audience
Step 1
Choose a Bangladeshi outlet
Pick one digital news outlet, OTT platform, or creator channel. Find any publicly available data about their audience: a reported subscriber number, a social media follower count, or a claimed monthly visitor figure.
Step 2
Identify the measurement method
Work out HOW that number was produced. Was it self-reported by the platform? A third-party panel estimate? An analytics tool like SimilarWeb? Write one sentence on the method and its limitations.
Step 3
Name the gap
Identify which of the three audience types (predicted, measured, delivered) the number represents. Then describe what the other two would look like for this organisation, and what data would be needed to produce them.
Today in five lines
Recap
01Ratings are a currency, not a truth: an agreed-upon proxy that the industry treats as real because it enables transactions.
02Every measurement system has a gap: diaries missed small audiences; people meters reintroduced active-viewer error; PPMs undercounted minority listeners.
03There are always three audiences: predicted, measured, and delivered. The industry is moving from pricing by the second to pricing by the third.
04Chartbeat showed that real-time analytics change editorial behaviour: they produce compulsive dashboard-watching and create tension between clicks and public interest.
05In Bangladesh, digital ad spend follows Facebook's own metrics more than any independent measurement, making the platform the de facto ratings authority.
Next class: Listening to the Audience. We move from counting to social listening, dashboards, and real-time sentiment analysis.
Digital Audience · Lecture 09
The count is never neutral.
This week’s reading
Reading list
PrimaryBuzzard, Tracking the Audience — Introduction & Chs. 1–7 (analog and digital ratings currencies)
SupplementaryNightingale (ed.) — Ch. 14, “Ratings and Audience Measurement” (Philip M. Napoli)