Applying a 47-second office-multitasking figure to the claim that nobody can finish a novel anymore is not a subtle statistical stretch; it’s measuring one activity and reporting the result as if it described a completely different one.
Part 2 of an occasional series examining The Atlantic’s “The End of Reading Is Here”
Part 1 of this series looked at the single example Rose Horowitch reaches for early in her essay – Doctor Zhivago as proof that Americans once had the appetite for demanding prose. The Cold War apparatus behind that bestseller (a Soviet ban, a smuggled manuscript, a CIA operation, a refused Nobel Prize) turned out to explain the sales figures rather better than any theory of national reading stamina did.
But the Zhivago comparison isn’t a one-off slip. It’s a symptom of a wider methodological habit running through the whole essay: treating sentence length, and screen time, as reliable stand-ins for something they don’t actually measure. Neither thread stands up to scrutiny.
Short sentences are not a symptom
Horowitch’s implicit measure of literary seriousness is syntactic complexity – long, subordinate-clause-heavy sentences, of the kind Pasternak’s Russian-to-English translators produced. Short, plain sentences become, by the same logic, evidence of a shrinking attention span and a diminished culture.
This doesn’t survive thirty seconds’ contact with the actual canon. Hemingway built an entire style, and a Nobel Prize, on short declarative sentences; nobody has ever accused The Old Man and the Sea of being unserious.
Steinbeck’s Of Mice and Men – for decades the single most widely taught novel in British secondary schools, not despite its brevity but partly because of it – runs to barely a hundred pages and asks nothing less of its reader than any 700-page epic.
Kazuo Ishiguro has built a Nobel-winning career on prose a ten-year-old could parse word by word, while asking adult readers to sit with some of the most quietly devastating ideas in contemporary fiction.
Cormac McCarthy could write a sentence a page long, then answer it with three words, and nobody who has read Blood Meridian or Child of God would call the result undemanding.
Sentence length correlates with plenty of things – period, translation convention, genre, house style. It has never reliably correlated with the difficulty of the ideas a book asks you to hold in your head. Treating it as a proxy for reading capability, as Horwowich does, doesn’t just risk error; it gets the direction of the craft backwards. Cutting a sentence to its essentials is usually harder than padding it out, not easier.
The doorstopper problem
If Horowitch’s thesis were right, we’d expect today’s bestselling fiction to be visibly shrinking, in both sentence length and overall scale, as readers lose the stamina for anything demanding.
What we actually have is the golden age of the fantasy doorstopper. Brandon Sanderson’s Stormlight Archive novels run past 400,000 words apiece, tracking a dozen point-of-view characters across multiple timelines and an invented cosmology with its own internal physics.
George R. R. Martin’s A Song of Ice and Fire and Robert Jordan’s Wheel of Time ask readers to hold hundreds of named characters and decades of invented history in mind simultaneously – a feat of sustained structural attention that has no real equivalent in 1958’s bestseller list. Doctor Zhivago itself, for context, is shorter than the average Game of Thrones volume.
None of this is a claim that fantasy readers are smarter or more literary than anyone else. It’s simply that the argument that “readers today can’t sustain long, complex texts” is measurably false at the level of the bestseller list Horowitch herself is using as evidence – she has just measured the wrong variable. Sentence-level complexity has fallen in places; structural and narrative complexity, in some of the most commercially dominant fiction of the last two decades, has gone in the opposite direction.
Which screen?
The essay’s other major evidentiary pillar is attention research – specifically the finding, from UC Irvine informatics professor Gloria Mark, that the average attention span “on a screen” is now around 47 seconds before switching, down from roughly two and a half minutes when she began measuring in the early 2000s. It’s a startling number, and it’s genuine, peer-reviewed, and repeatedly replicated.
And it’s total BS in the way Horowitch deploys it.
Mark’s foundational studies measured something specific: how long office workers and university students stayed on a single screen or task before switching to another, tracked via logging software installed on workplace computers, later supplemented by prompts sent to computers or phones. It is, in other words, a study of task-switching during knowledge work – email, documents, browser tabs, notifications – not a study of sustained leisure reading on any device.
Applying a 47-second office-multitasking figure to the claim that nobody can finish a novel anymore is not a subtle statistical stretch; it’s measuring one activity and reporting the result as if it described a completely different one. It’s BS.
It also erases distinctions that matter enormously in practice, and that any working reader could point to immediately. Mark does not define what she meant by a “screen”, and Horowitch had an agenda that precluded asking.
A cinema screen demands total, uninterrupted attention by design – no pause, no rewind, and social pressure against checking a phone or a holding a conversation.
A television at home offers the same kind of content with a remote control in hand and a household of distractions nearby.
A dedicated e-reader, or a reading app on a tablet, is built for the opposite of task-switching: no notifications competing for space, full user control over pace, and the ability to pause and reflect rather than being swept along.
A smartphone, the device Horowitch’s essay implicitly conjures with the phrase “on a screen,” is the one genuinely engineered around interruption – and, not coincidentally, close to the device on which most of Mark’s data was collected in the first place.
“Screen time” as a single category was never a coherent unit of analysis. Conflating a 47-second office-attention statistic with an inability to read novels treats every screen as functionally identical to a smartphone notification feed, when the whole point of Mark’s own distinction between different technologies is that they aren’t.
The essay’s own unresolved evidence
The clearest sign that something is wrong with Horowitch’s framework is that her own essay contains the evidence against it, unreconciled. She notes, in passing, the boom in long-form writing on Substack. She notes independent bookstores opening, not closing. Neither observation gets folded back into the “postliterate America” thesis; both sit there as loose threads the essay never ties off.
That’s not a minor inconsistency. If Americans were genuinely losing the capacity for sustained, demanding reading, a platform built entirely around long-form written essays wouldn’t be thriving, and a retail category built entirely around selling physical books would be contracting rather than expanding. The data Horowitch herself cites describes a reading culture that has changed shape – not one that is disappearing.
Reading hasn’t gone anywhere
Put the two flawed measures together – sentence length as a proxy for capability, “screen time” as a single undifferentiated category – and the essay’s central claim stops holding its shape.
What’s actually changed is where reading happens, in what form, and who gets counted as doing it: a doorstopper fantasy series counts, a Kindle app counts, an hour of Substack counts, and none of them look like the 1950s bestseller list Horowitch is quietly using as her baseline for what “real” reading looks like.
The next instalment in this series turns to a different sleight of hand – the industry habit of calling audiobooks “reading,” and why that conflation matters more than it might first appear.
This is Part 2 of an occasional TNPS series examining the evidence behind “The End of Reading Is Here.” Part 1 looked at what The Atlantic’s use of Doctor Zhivago as a bestseller comparison leaves out
This post first appeared in the TNPS LinkedIn Analysis Newsletter.
