A couple GUADECs ago, I agreed to do some good old fashioned research for the board on the issue of ‘crowding out of intrinsic motivations’- in other words, what happens when you start paying volunteers? I had been told by someone (I believe by the excellent Karim Lakhani) that there was good research demonstrating repeatedly that paying volunteers can have a counter-intuitive effect- that paid volunteers end up working less. I didn’t do the follow-up research, unfortunately, so I was unprepared to discuss the issue when bounties came up a few months later in a GNOME Foundation Board meeting at GUADEC.
While reading Karim’s dissertation (verdict: excellent and interesting- a blog about that when I finish it) I came across a reference to a key paper in the field. Today, in between world cup matches, I finally sat down and found the paper (here), and started reading it. Thanks to the excellent scholar.google.com, I was able to find an even better series of papers citing that one, providing context and lots more information. I strongly recommend anyone thinking about bounties or other financial incentives in a free software context read in particular ‘Motivation Crowding Theory: a Survey of Empirical Evidence’, by Frey and Jegen. (Ignore the economics; skip to the examples, which are really interesting.) Herewith some notes and comments:
- while at first the idea is counterintuitive, the basic notion is straightforward: when people do things for their own intrinsic goodness (i.e., for reasons other than payment), introducing payment can reduce the amount of invovement.
- Clearest example I’ve found in the quick literature skim so far: an experiment on parents who used Israeli child care. The parents had to pick up their kids at the end of the day; being late was not punished financially, but obviously was an inconvenience to the teachers. When fines were instituted, presumably parents would get more timely in picking up their kids, since they had whatever motivated them previously, plus the fine. The opposite occured- parents became worse at picking their kids up on time when the fine was in place. The real kicker? The fine was canceled… and parents did not return to their original level of picking kids up on time. They had been permanently (at least for the duration of the experiment) ‘ruined.’ (Referenced article is here, but not available publicly, unfortunately; I read about it in Frey and Jegen.)
- The basic psychological rationale behind crowding out theory is that as soon as you pay people (or introduce other ‘forcing’ incentives, like increased managerial oversight of employees), you’re reducing the sense of self-determination and self-esteem, and so their intrinsic motivation- doing it because they want to help others, or because it is fun, or because they are embarassed to seem like they don’t care about their kids.
- This isn’t all bad- there are also examples of ‘crowding in’, where external incentives can strengthen pre-existing intrinsic motivations- for example, if laws are structured to imply to citizens that citizens are trusted, they may in fact act in a more trust-worthy manner because they feel more valued/respected/etc.
- There is at least some controversy about this theory in academia. There is a faction of behavioralist psychologists who argues that the whole thing is BS. However, both of the papers by economists that I read suggested that this faction’s research was deeply flawed and conclusively disproved by the first paper I linked to. There are apparently also a number of economists who don’t believe that this effect can be more significant than positive economic incentives (i.e., they believe that increased pay will always increase supply, without exception); the second paper I linked to seems like a fairly conclusive debunking of this position.
- I have not yet seen references to research into community-wide crowding out; i.e., if I’m offered money to do bug work, does that crowd out incentives for Elijah and Olav? (The anecdotal evidence would suggest ‘no’, but perhaps that is partially because of the large gap between volunteering and full-time jobs- bounties may be different in this respect.)
- Directly relevant to GNOME, there is some research that indicates that paying volunteers can reduce their overall level of contribution. Specifically, it concludes that those paid small amounts in correlation with volunteering work fewer hours than the average volunteer, while those paid large amounts end up working more hours than the average volunteer. The data is survey based, rather than experiment-based, though, so I’m a little leery of placing too much weight on it- the effects may well be correlation and not causation, though the study authors do attempt to control for other causes.
So, uh, what does all this mean? Not sure it means a whole lot, exactly- none of the research is so strong as to be conclusive or suggest direct guidelines for a bounty-issuing volunteer organization. It does suggest to me that we should consider doing more reading and research before substantially expanding payments to volunteers. Given the suggestion that these communities are fragile (i.e, that once you’ve destroyed the sense of intrinsic motivation, it doesn’t come back) it suggests that this isn’t something to be trifled with without thought. Finally, it suggests that there is a very, very good psych/econ paper lurking in google’s bounty system- surely this is one of the widest-scale experiments in potential crowding out ever, and if someone can figure out a clever way of measuring the impact, they should have a very good paper and very useful advice for us on their hands.
Anyway, the last two papers (not the first one, it is by psychologists and substantially more unreadable than the other two) took me only a couple hours to read; I’d suggest that anyone interested in this topic should suck them down and give them a read.