The Goldfish Myth

What it is

The route is the lesson, because every step in it is a normal, defensible act of citation and the result is a fabrication that reads as science.

It begins with a Microsoft Canada marketing report on digital advertising. Not a peer-reviewed paper and not a psychology experiment, but a document produced to sell attention to advertisers, which puts a conflict of interest at the origin of a claim about attention. That report attributes the figure to a secondary source that cannot be substantiated, and the trail ends there rather than at a dataset, a method or a sample. The goldfish comparison has no identifiable origin at all: nobody has produced the study that timed a goldfish, and it is a rhetorical flourish that acquired a number. A news outlet then reported the marketing document under a headline beginning with the word Science. And a book cites the news article in its endnotes while calling it a Microsoft study in its text, so the reader sees a study, a company, a year and a number.

Notice what survives each step and what does not. The number survives every time. The provenance is stripped when the press reports it, and the genre is upgraded when a book cites the press.

The vault met the last step in Zelana Montminy's Finding Focus, which opens its concentration chapter at page 78 with the figure attributed in the prose to a 2015 study from Microsoft. The endnote does not cite a study. It cites a news article in Time from May 2015. There is no Microsoft study anywhere in the notes and no primary source anywhere in the book.

In effect

The premise fails as well as the provenance, and that matters more. Attention researchers reject the idea that a single average attention span is a meaningful quantity at all. Attention duration is task dependent: the same person sustains attention for very different lengths on a spreadsheet, a conversation, a film and a driving hazard, so a population average across all tasks is not a measurement of anything. That is why the figure cannot be fixed by finding a better study. There is no number to find.

The contrast with real measurement is instructive. Larry Rosen, L. Mark Carrier and Nancy Cheever observed two hundred and sixty-three school and university students studying in their own homes for fifteen minutes, with trained observers logging the technology present and then recording on-task and off-task behaviour minute by minute. Students averaged under six minutes on task before switching, most often to social media or texting. That figure has a sample, a method, a setting and stated limits: the observation window is only fifteen minutes, so a switch at six minutes is close to the floor of what the design can resolve, and the accompanying finding about grade averages is correlational. The figure that circulates from that study, including in its own first author's later trade book, is three to five minutes. The published number is under six minutes. Quote it as under six minutes from a fifteen-minute observation, or do not quote it.

The sibling case is worth reading alongside this one, because it is the same laundering caught at a different link. The claim that we have around sixty thousand thoughts a day and that most of them are negative terminates in a business blog rather than a marketing report, and it is credited to a national funding agency that does not run and publish studies under its own name, which makes that citation not merely unverified but structurally impossible. The number also drifts between retellings, which is the giveaway. One question about the nature of the institution being credited would have stopped a claim that has circulated for decades.

What it does not say

It does not say the intuition behind it is silly. People genuinely do switch more often on screens than off them, and there is real measurement behind that. The specific claim, the specific numbers and the goldfish comparison are not evidence for it, and using them to make a roughly true point with a false statistic is how the genre keeps discrediting itself.

It does not belong in the replication crisis. You cannot fail to replicate a study that was never run. Replication is a filter positioned downstream of a gate this claim walked straight past, which is why it is filed as contested rather than unverified: unverified means nobody has opened the paper yet, and there is no paper to open.

It does not need a hedge to be usable. The figure may appear in one form only, as a worked example of citation laundering with the chain shown, which is what this page is.

What makes the case valuable is that the appearance of sourcing and the fact of sourcing turn out to be separate things. The book in which the vault met it has unusually good endnotes, mostly peer-reviewed journal articles, and the reader is never shown them. The apparatus that would have exposed the claim is in the same volume as the claim.

---