Back to the directoryFind in the timeline
The MusicLab artificial music market
In an artificial music market, 14,341 participants downloaded unknown songs. Seeing others' download counts made success more unequal and less predictable, though the best songs rarely did badly. Any link to feeds or charts is interpretation.
- Where it sits
- Academic psychologyResearch published for other researchers and students.
- Evidence level
- Independent experimentIndependent research, with its design and limits stated.
- Sources inspected
- 2 of 21 with the relevant text, company document or archive item checked, 1 as an abstract or reference article, 0 not verified.
What happened
In 2006 Salganik, Dodds and Watts reported two experiments in an artificial music market called MusicLab. In total 14,341 participants, recruited mostly from a teen-interest Web site, saw a menu of 48 previously unknown songs by unknown bands. They could listen to a song, rate it from one to five stars, and choose whether to download it for free.
The psychological idea
Participants were randomly assigned either to an independent condition, where they saw only band and song names, or to one of eight separate social influence worlds, where they also saw how many times each song had been downloaded in their own world. About 20 per cent were in the independent condition and 10 per cent in each influence world. In the second experiment the influence worlds listed songs in descending order of downloads, a stronger signal.
In practice
No platform or company use is documented. Success was more unequal and less predictable in the social influence worlds than in the independent world, and more so in the second experiment. Quality, measured by a song's share in the independent world, was only partly linked to success: the best songs rarely did poorly and the worst rarely did well, but almost any other outcome was possible.
Why it mattered
The design created several parallel histories from identical starting conditions, whereas in reality only one history is ever observed. The authors conjecture that experts fail to predict hits because individual decisions are influenced by others, not because the experts judge poorly. Any application to feeds, rankings or recommendation systems is interpretation.
Limits and qualifications
The authors say real markets differ: marketing, product placement, critical acclaim and media attention are absent here, and the effects of social influence were probably diminished by the small number of songs and because participants could take part only once and could not share opinions. The songs were by unknown bands, the downloads were free, and social influence meant only download counts.
Evidence behind the claims
The MusicLab experiments involved 14,341 participants, recruited mostly from a teen-interest Web site, who saw 48 previously unknown songs by unknown bands, rated them from one to five stars and could download them for free.
Participants were randomly assigned to an independent condition or to one of eight social influence worlds, with about 20 per cent in the independent condition and 10 per cent in each influence world, and participants in influence worlds saw the download counts of their own world only. This created parallel histories from identical starting conditions, whereas in reality only one history is ever observed.
In experiment 2 the influence worlds listed songs in descending order of downloads, and inequality and unpredictability were higher in the influence worlds than in the independent condition and higher in experiment 2 than experiment 1.
Quality was only partly related to success: the best songs rarely did poorly and the worst rarely did well, but almost any other result was possible.
The authors conjecture that experts fail to predict success because individual decisions are subject to social influence, not because experts are incompetent.
The authors state that real markets include marketing, product placement, critical acclaim and media attention that were absent, and that social influence effects were probably diminished by the small number of songs and participants joining only once without sharing opinions.
People involved
- Matthew J. Salganik. Sociologist, lead author
- Peter Sheridan Dodds. Co-author
- Duncan J. Watts. Co-author
In the Psy101 course
- OpenStax Psychology 2e, section 12.4: Conformity, Compliance, and ObedienceChapter 12, Social Psychology. Context. Nearest related section; it does not discuss this topic by name.Nearest related section: it covers conformity and informational social influence, but not cultural markets, popularity rankings or music downloads.
This is course relevance only. It is not evidence for the historical or corporate claims on this page.See the full course map.
Sources and inspection
2 of 2 sources were opened and read at the level shown. A source marked as not verified supports no claim on this page by itself.
- [1]Experimental Study of Inequality and Unpredictability in an Artificial Cultural MarketSalganik, M. J.; Dodds, P. S.; Watts, D. J.. Science, 311(5762), 854-856, 2006. Journal article. Primary source.Relevant text checkedRead the whole three-page report (abstract, Figures 1 to 3, closing discussion, references) as images, plus the first two pages of the Supporting Online Material (experimental design, subject experience). The supplement tables and figures were not read. The copy is hosted on the first author's university page.
- [2]Experimental study of inequality and unpredictability in an artificial cultural market (Europe PMC record)Salganik, M. J.; Dodds, P. S.; Watts, D. J.. Science, 311(5762), 854-856, 2006. Journal article. Primary source.Abstract or publisher record onlyRead the abstract as returned by the Europe PMC record, which agrees with the PDF.
Access note: No link between MusicLab and any platform's ranking system is claimed. Supplementary tables and later follow-up studies were not read.
Independent source review: Claude review agent, independent of the research agent, 2026-10-21. Verdict: pass-with-edits.