We thought about this question quite extensively and there are a few reasons why we sampled the top 1% (which we didn't get around to discussing in this brief paper). First, because of the high degree of contribution inequality in Wikipedia's editing community, we were primarily interested in how status rewards affect the all-important core of highly-active editors. There is also a lot of turn-over in the long tail of the distribution, and even among the most active editors, there is considerable heterogeneity. Focusing on the most active users ensured us sufficient statistical power. (Post-hoc power analysis suggests that our sample size would need to be several thousand users in the 80-90th percentiles, and several hundred in the 90-99th percentiles, to discern an effect of the same strength.) Also, we considered the question of construct validity:  which users are deserving (so to speak) of receiving an editing award or social recognition of their work?

You are right that it should be fairly easy to extend this analysis beyond just the top 1%, but just how wide a net to cast remains a question. The issue of power calculation and sample size becomes increasingly difficult to manage for lower deciles because of the power-law distribution. And I don't think it would be very meaningful to assess the effect of barnstars on the bottom half of the distribution, for example, for the substantive reasons I mentioned above. Still, I'd be curious to hear what you think, and whether there might be some variations on this experiment that could overcome these limitations.

In terms of data dredging, that is always a concern and I completely understand where you are coming from. In fact, as both and author and consumer of scientific knowledge, I'm rarely ever completely satisfied. For example, a related concern that I have is the filing cabinet effect - when research produces null (or opposite) results and hence the authors decide to not attempt to have it published.

In this case, I actually started this project with the hunch that barnstars would lead to a slight decline in editing behavior; my rationale was that rewards would act as social markers that editors' past work was sufficient to earn social recognition and hence receiving such a reward would signal that the editor had "done enough" for the time being. In addition to there being substantial support for this idea in the economics literature, this intuition stemmed from hearing about an (unpublished) observational study of barnstars by Gueorgi Kossinets (formerly at Cornell, now at Google) that suggested editors receive barnstars at the peak of their editing activity. Of course, we chose an experimental design precisely to help us to tease out the causal direction as well as what effect barnstars have for recipients relative to their unrewarded counterparts. We felt like no matter what we found - either a positive, negative, or even no effect - it would have been interesting enough to publish, so hopefully that alleviates some of your concerns.

Please let me know if you have any other questions, and I'd love to hear your thoughts about potential follow-ups to this research.