On this page
The three figures every community engagement plan quotes, ninety per cent reading, nine per cent contributing now and then, one per cent writing most of what everyone else reads, come out of one article: Jakob Nielsen's "Participation Inequality: Encouraging More Users to Contribute", published on the Nielsen Norman Group site in October 2006. Nielsen never called it a law. He assembled it from counts he had to hand, and the studies published before and after it agree on the shape of the distribution while scattering its values across almost the whole range.
Where the 2006 figures came from
The evidence under the split comes to four observations, with no study behind any of them. Nielsen put Wikipedia's active contributors at 68,000 against 32 million unique visitors, which is 0.2 per cent, and its 1,000 heaviest editors, 0.003 per cent of users, at around two-thirds of all edits. On Usenet he counted 27 per cent of postings coming from people who had posted exactly once, while the busiest 3 per cent of posters produced 25 per cent of the messages. For blogs he suggested the ratio ran closer to 95-5-0.1, and he put Amazon reviews at under 1 per cent of customers. He wrote all of that up as participation inequality, a pattern he had seen, and the caveat is the half of the citation that stopped travelling.
Two studies that bracket the range
| Study | Sample | Range or finding |
|---|---|---|
| Nonnecke & Preece, CHI 2000 | 109 email discussion lists, 60,000+ members, 19,000 posters, over twelve weeks | Mean lurking 46 per cent on health lists, 82 per cent on software lists; individual lists ran 0 to 99 per cent |
| van Mierlo, JMIR 2014 | 63,990 users, 578,349 posts, four health support networks | Top 1 per cent of posters produced between 59.0 and 75.0 per cent of all content |
Blair Nonnecke and Jenny Preece got there six years before the rule existed. Their CHI 2000 paper "Lurker demographics: counting the silent" tracked 109 email discussion lists over twelve weeks, covering more than 60,000 members and 19,000 posters, and counted as a lurker anyone who posted nothing in that window. Mean lurking was 46 per cent on the health support lists and 82 per cent on the software support lists. Individual lists ran from 0 to 99 per cent. On that evidence the subject of a list predicted who posted better than the software it ran on did.
Trevor van Mierlo's "The 1% Rule in Four Digital Health Social Networks", published in the Journal of Medical Internet Research in 2014, pulls the other way. Across 63,990 users and 578,349 posts on four health support networks, the top one per cent of posters produced between 59.0 and 75.0 per cent of all content, and the four networks had no reason to resemble each other that closely.
The two papers do not disagree about concentration. Both find a small band writing the bulk of what is on the screen, which is the part of Nielsen's summary that has held up everywhere anyone has counted. What they disagree about is the silent share, and Nonnecke and Preece's spread puts the 90 in 90-9-1 at one point inside a distribution wide enough to contain almost any board. A support forum for a chronic illness and a bug-reporting list for a build tool are both communities, and the 2000 paper says they will not share a lurking rate.
The part of the ratio a community can move
Participation costs time and exposes you to judgement, and reading costs almost nothing. That asymmetry survives a healthy community, a paid moderation team and a redesign, which is why the ratio holds across software that has nothing else in common.
Reading it as a failure produces three expensive moves: gating the archive so readers have to register, prompting members to post, and measuring the board by posts per member until it looks like it is dying. The 90 per cent are the reason the 1 per cent write at all. An answer read four thousand times by people who never reply has done the job it was posted for, and much of that reading arrives from search, which makes thread indexing and search visibility a question about how large the reading band gets.
What is worth attacking is the absolute size of each band and the rate at which people cross between them. Those are the numbers that move on a real board, and they are measurable in a way the ratio is not: how many new members post once, how many first posts get an answer inside a day, how many occasional posters became regulars this quarter.
The reply to a first post
Elisabeth Joyce and Robert Kraut followed 2,777 newcomers across six public newsgroups for "Predicting Continued Participation in Newsgroups", published in the Journal of Computer-Mediated Communication in 2006. Sixty-one per cent of those newcomers received a reply to their opening message. The ones who did posted again at 56 per cent against 44 per cent for the ones who did not, a gap of 12 percentage points, and a newcomer who asked a question or wrote a longer message was more likely to be answered at all.
The finding that decides how a moderation rota is staffed is the one about quality. Joyce and Kraut measured the emotional tone of the reply and how helpful it was, and neither made a measurable difference to whether the newcomer came back. A short, plain answer counted the same as a careful one. What the 2006 data separates is answered from unanswered, across 2,777 first posts on six newsgroups.