Wikipedia in theory (experimental economics edition)

(See my earlier posts: Wikipedia in theory, and Wikipedia in theory (Marxist edition).)

Wikipedia works (if imperfectly) in practice, even though some relevant theories say it shouldn’t.  Take game theory.

We can think of Wikipedia as a public goods game: contributing time and effort into improving it doesn’t have a direct benefit, but the result of many people doing that creates a resource for everyone.  There’s little direct incentive to contribute; whatever I might do to make it better is based on what I already know, and my personal improvements only come back to benefit me with a fraction of the value I put into the project.  So despite that the optimal situation for the public as a whole is if everyone who could contribute to Wikipedia did so, for any individual the strictly rational choice is not to contribute–to be a “free rider” or “defector”.

As in the canonical version of the public goods game, the Nash equilibrium for Wikipedia is zero contributions.  In a world of strictly rational, self-interested players of the Wikipedia game, the projects dies a silent theoretical death in 2001–which we know, in practice, is not what happened.  Experimental economics has been focused on just this divergence between theory and practice for several decades now, and there may be a lot of insights in that body of literature for how to make Wikipedia work better.

What factors make a public goods game more successful?

One study looked a “partners condition” versus a “strangers condition”: in repeated plays of a public goods game, players were either matched with the same group from earlier games or a new group of strangers.

  • The result: players consistently contribute more to the public pot when playing with people they have become familiar with.
  • Applied to Wikipedia: Wikipedians who are familiar with each other will contribute more.  We should provide ways to knit Wikipedian identities into the broader social fabric, so that relationships of trust and familiarity created outside of Wikipedia can be ported in.  We hold the right of anonymity dear, but that doesn’t preclude doing more to support and encourage real identities for those who are willing to use them.

Another study focused on “inequality aversion”: what do players do if they know how much (or little) others are contributing and can incrementally increase contributions in response to other players?

  • The result: players raise their contributions if others do the same, but “most players are willing to contribute to the public good at a level at or slightly above the contribution of the lowest contributor in the group”.
  • Applied to Wikipedia: in recruiting potential contributors from specific groups (e.g., academics) we should highlight contributions their peers have made.  We should also do more to (automatically) catalog individuals’ contributions in ways that are easy to understand, share, and compare within peer groups.  “X number of experts in your field have made at least Y number of contributions” could be an effective pitch.

Other studies, including this one, have looked at “endowment heterogeneity” and “endowment origin”: does it matter whether players earned what they might contribute or received it as a windfall, or whether the players have unequal potential for contribution?

  • The result: studies have come up with conflicting answers on whether endowment origin (windfall vs. earned) matters, with some (but not the linked one) finding that people are more generous and more willing to take the risk of contributing heavily to a public pot when spending from a windfall.  The linked study does find, however, that in groups with heterogeneous endowments there is less contribution–likely because of the “inequality aversion” factor discussed above, since those with the most to contribute scale back based on what they expect others to contribute.
  • Applied to Wikipedia: if the windfall effect is real, it might be most effective to target recruitment efforts at people who have received “windfalls” of free time and knowledge as opposed to those who have earned it.  Smart, geeky people, to whom learning comes easy, would seem to fit the bill best: they’ve received windfalls of free time and knowledge both because they learn quickly (leaving more time for other things) and because they often get years of free schooling funded by grants and scholarships.  But the over-educated are already overrepresented on Wikipedia (we think) and the level of contribution from those people may be held back by the endowment heterogeneity effect, with highly educated people holding back because those with less free time and/or knowledge contribute so little.  In that case, the key to getting experts to contribute more would be (as Erik Möller suggested at Wikimania) finding ways light-weight ways to get non-experts and readers (free-riders) more involved first.

Still other studies have explored factors that make players change their behavior: this paper examines punishment mechanisms, and this one looks at history and how group behaviors change after multiple rounds of a public goods game.

  • The results: the ability for contributors to punish free riders (even at a cost to themselves) results in higher levels of cooperation and less free riding (although opportunities for counter-punishment make this less effective).  Conversely, players who begin the public goods game with a high tendency to contribute tend to gradually contribute less in later rounds the more they play with free riders.  But matching high contributors with other high contributors in round after round leads to growing contributions among that group.
  • Applied to Wikipedia: punishment mechanisms for free riders are conceivable (e.g., pop-ups asking for monetary or editing contributions after a certain number of page views by a reader), but probably wouldn’t be compatible with Wikipedia culture and the project’s purpose.  Burnout from active contributors getting dispirited when they do more than everyone else they see probably is a problem, and the solution would be to facilitate and strengthen the social ties between active users.  This already happens naturally on talk pages and user pages,  outside communication channels like email and Skype, and physical meetups, but Wikipedia could make it a lot easier to see and connect with other contributors through software improvements.

This just scratches the surface of experimental research on public goods games; a more systematic survey could turn up a lot more relevant data for how better to structure Wikipedia and other collaborative knowledge projects.