Studying Virtual Economies

A recent article in Newsweek discussed the use of online games like Farmville and virtual worlds like Second Life for studying human behavior when buying and selling goods, a field called behavioral economics.  Consider the power of this experimental design:

Instead of dealing only with historical data, in virtual worlds “you have the power to experiment in real time,” Segerstrale says. What happens to demand if you add a 5 percent tax to a product? What if you apply a 5 percent tax to one half of a group and a 7 percent tax to the other half? “You can conduct any experiment you want,” he says. “You might discover that women over 35 have a higher tolerance to a tax than males aged 15 to 20—stuff that’s just not possible to discover in the real world.”

While that certainly sounds attractive, I have to wonder just how representative buyers of virtual goods are of the population at large.  Is the woman over 35 with higher tolerance to tax buying virtual designer clothing from the same population as the woman over 35 buying clothing at Old Navy and Gap?

It’s also important to note that online economies are typically based on microtransactions.  For those of you not familiar with microtransactions, they are exactly as they sound – buying and selling for very small amounts of money, often in a virtual currency.

As an example, consider a virtual artist’s easel that I sell through the XStreet SL marketplace for use in Second Life.

It costs me nothing ($0) to produce these easels.  It originally took me about 20 minutes to physically design it in SL.
It costs a buyer L$50 (US$0.20) to purchase a copy of the easel.
I lose L$2 (US$0.01) in fees for selling the easel on XStreet SL.

I only sell 4 or 5 of these per month, resulting in a total monthly profit for me on this item of roughly L$200-250 (US$0.80 – US$1.00).  If XStreet SL raised its fee rate from 4% to 7%, I wouldn’t really care, because it’s just not that much money to begin with.

But wait, you say – what about people who spend a lot on virtual goods?  You know someone that spends $50 per month on virtual clothing?  Because that behavior is relatively unusual, I’d argue that this just means the representativeness problem is even more of a concern.

So either way, there’s trouble, and it causes me to question results from such studies.  Either you study everyone and risk including people with spending motivations dissimilar to the real world, or you only study the big spenders and risk excluding all the “normal” purchasers.  Is there a middle ground?  Not one that I can see.