Measuring a Moving Target: Why ATUS, Nielsen, eMarketer, and GWI All Disagree About How Much Media We Consume

One question, four different numbers

The apparent disagreement between media-measurement sources is not usually a factual dispute. It is a definitional one. "Time spent with media" can mean a respondent's self-reported primary activity across a 24-hour diary, a device's metered usage inside a household panel, an analyst's modeled synthesis of multiple inputs, or a respondent's self-reported estimate on a survey. Each produces a real, internally consistent number that is not directly comparable to the others without adjustment.

Method one: the 24-hour diary (ATUS)

The US Bureau of Labor Statistics' American Time Use Survey is the only nationally representative, government-run time-diary dataset covering US daily activity, running continuously since 2003. It recorded US adults' TV time averaging 2 hours 46 minutes per day in the 2013 to 2017 period, rising to a pandemic-era peak of 2 hours 52 minutes in 2021, then declining to 2 hours 36 minutes by 2024, the lowest level in the survey's history.

ATUS has a documented structural limitation worth stating plainly: it records only a respondent's stated primary activity in each time block, with one narrow exception for secondary childcare. It does not capture simultaneous, background, or second-screen use. That means ATUS systematically undercounts multi-device attention relative to a metered panel or a self-report survey that asks about concurrent exposure.

Method two: the metered panel (Nielsen)

Nielsen measures a different thing again: device-level usage inside a metered household panel, disclosed and industry-standard. Its monthly report, The Gauge, tracks how total TV usage splits across streaming, cable, and broadcast, and shows streaming crossing 40.3 percent of total TV usage in June 2024, a share Nielsen itself flags as a new record. This method captures what a device is doing, not what a person says their primary activity was, which is a materially different question from ATUS's diary.

A parallel example outside television shows how much a chosen metric can shape a headline even within the "metered" family: comScore reported in March 2015 that US mobile-only internet users first exceeded desktop-only users, a user-composition metric, while StatCounter reported in October 2016 that mobile-plus-tablet usage share, a page-view-based metric, exceeded desktop worldwide for the first time. Both are real, correctly measured, and describe two different milestones commonly conflated into one "mobile overtook desktop" headline.

Method three: the vendor-modeled estimate (eMarketer, and similar trackers)

eMarketer's "Time Spent with Media" series, the most frequently cited US media-time figure in trade press, is explicitly an analyst's modeled synthesis of multiple inputs, not a raw panel or diary. By that method, 2013 was the year US adults' average daily digital media time (5 hours 46 minutes) first surpassed television time, and by 2024, total daily media time reached 12 hours 37 minutes, of which 63.7 percent was digital.

That 12-hour-plus total is more than four times ATUS's 2 hour 36 minute TV figure for the same year, which is not a contradiction once the methods are named: eMarketer is modeling total media time across every format and device, much of it concurrent, while ATUS is diarizing only a respondent's single stated primary activity. Self-report survey vendors that track digital behavior, including trackers such as GWI, sit in a related but distinct category again, relying on respondents' own recollection of their usage across a structured questionnaire rather than a diary or a device meter, which tends to produce yet another set of totals.

Method four: the self-report survey, and a live example of the risk

Survey-based tracking has its own comparability pitfalls, visible in Pew Research Center's own data. Pew's long-run tracker reports US adult social media adoption at 72 percent in its chart series, while Our World in Data's synthesis of Pew and company-reported figures cites 79 percent for 2019. The gap most likely reflects different survey years or question wording rather than an actual decline, but the two figures should not be quoted together as if they describe the identical measurement without first reconciling the exact survey waves behind each one.

What this means for reading any "time spent with media" claim

None of the four methods above is more "true" than the others; each answers a genuinely different question. A diary captures self-reported primary activity with high fidelity but undercounts multitasking. A metered panel captures device-level usage with precision but says nothing about a person's intent or attention quality. A vendor model captures a broad, useful magnitude but is a synthesized estimate, not a raw measurement. A self-report survey captures perceived behavior but is vulnerable to recall bias and wording effects.

The discipline this forces on anyone building an attention map, including this one, is to name the source and the method next to every figure, not just the headline number, and to avoid combining totals from different methods as though they were interchangeable.

This piece, held to its own rule

This piece itself follows the rule it describes: every figure above is attributed to its specific source and, where relevant, flagged as vendor-modeled, government-diary, metered-panel, or comparability-flagged. GWI and similar self-report trackers are referenced here only as an example of the survey-based methodology category, not as a source of any specific figure cited in this piece, since no GWI-specific dataset was verified in this pass.