A Researcher Studied Us For Four Years And Never Told Us. We Asked Why.

SEATTLE, United States — Somebody has been studying this newspaper and did not mention it.

A network analysis of how contested narratives move between outlets — the kind of work that maps which sites cite which, and how a claim travels — has been running for four years. It takes in several thousand publications. This one is among them.

We found out by accident, from a figure in a conference paper in which one node was labelled with our domain.

We are node 11,442.

What she was measuring

Ludmila Ferreira-Achterberg did not have to talk to us. She said so at the start of the first call and again at the end of the last one, and she was pleasant about it both times.

Her method does not involve reading articles. It involves watching where links go.

"I am not assessing whether anything is true," she said. "I could not, at this scale, and it is not my question. My question is structural. When a claim appears, what is the shape of the path it takes, and does the shape predict anything."

Outlets in her dataset are not scored for accuracy. They are positioned — by who cites them, who they cite, and how quickly material moves between them. A site can be entirely accurate and sit in a dense cluster; a site can be nonsense and sit alone.

"People always want the map to be a league table," she said. "It is a road map. It tells you how the traffic flows. It does not tell you which towns are nice."

Our row

We asked for our own row. She thought about it for three weeks and then sent it, on the condition that we print the caveats, which we now have.

It is more specific than anything our own analytics produce.

Her data has us in a sparse position: cited by comparatively few of the outlets around us, and citing almost none of them. The material that leaves this site does not, on her figures, travel into the dense clusters at all. It goes sideways, into general-interest media and into academic reading lists, and then it stops.

"You are a cul-de-sac," she said. "I mean that neutrally. Material enters your site and does not propagate outward through the ecosystem. That is unusual for a publication with your subject matter and I do not have a settled explanation for it."

She offered two. The first is that our readers are not the ecosystem's readers. The second is less flattering.

"You label everything," she said. "Your About page is unambiguous. Material that carries an explicit fiction label does not travel well in networks that run on assertion, because there is nothing to assert. Whether that is a feature is your call, not mine."

The part we did not enjoy

Her dataset also records something we publish ourselves and had never thought to compare: our corrections.

Outlets in her sample correct at a median rate she describes as "indistinguishable from zero." We are an outlier, which we knew and are quietly pleased about, and we were prepared for that paragraph to be the flattering one.

It is not, quite.

Her figures show that our corrections do not propagate either. When this newspaper corrects something, the correction reaches a small fraction of the outlets that picked up the original — and on her numbers, the ratio is worse than the ecosystem average.

"Your corrections are excellent and nobody reads them," she said. "That is not a criticism of the corrections. It is the oldest finding in this field and you are not exempt from it because you try hard."

We have printed a corrections archive for a year on the assumption that publishing well is most of the job. On this evidence it is not most of the job.

Being studied without being asked

Bertrand Osei-Kaplan is a research ethics officer at an institution with no connection to the study. We asked him the obvious question: is it acceptable to include a publication in a dataset without telling it?

"Yes, and it would be unworkable otherwise," he said. "A published website is a public artefact. You do not need consent to count what somebody has already put in a shop window."

He drew a line at the individual.

"If she were building profiles of your named readers, that is a different conversation and she is not. Counting outlets is not surveillance. Counting people can be."

We accept that, and we record for our readers that we asked, and that the answer went against the position we started with.

What we could not establish

We could not obtain the full dataset. Ferreira-Achterberg gave us our own row and declined the rest, on the grounds that other outlets have not agreed to be identified. That is consistent and we do not dispute it.

We could not test her propagation figures independently, because doing so would require the data she has properly declined to give us. Everything above about our own reach rests on her word and her method, and we are relying on a single source, which is a thing this newspaper tells its reporters not to do.

We could not establish how many other studies we are in.

This newspaper spends a great deal of its time asking institutions what they know about people who did not consent to be known. It turns out to be a different experience from the other side of it, and the honest thing to report is that the answer we got was reasonable and we still did not like it.