Further to the Google prediction market

We were surprised to note the widespread discussion throughout the blogsphere regarding the Google prediction market paper. Something there clearly captures the imagination in a way that the art and science of forecasting rarely otherwise can.

Today’s discussions have also surfaced a fascinating item not included in the original paper itself: an influence diagram of successful and unsuccessful forecasters. Surprisingly, represented as a heat map, one clearly sees an unprecedented proximity effect in clusters of both winners and losers. There is also a strong temporal correlation to these trades on the prediction market, further reinforcing the idea of action from a shared information base.

The representation also reminds us of other recent comments regarding geographic impact on social networks at the esteemed Network Weaving blog.

We have long been proponents of virtual distributed intelligence production – especially given the realities of life in the DC metro. And given the personalities of most intelligence analysts (stereotyped or otherwise), the prevalence of VTCs and telcons for interactions even within the national capital region, and the highly asynchronous schedules of so many professionals working in 24 / 7 environments such as watch desks or fusion centers, we had frankly thought little lost when virtual teams were properly organized and managed.

Now the ‘plex no doubt has a very different organizational culture than the IC, and we are uncertain that these findings may be applied more generally within our field. We would particularly question cross-domain validity given the number of critical interactions already mediated by networks even in the most physically cohesive of teams, such as the requirements and evaluation process, or almost every element of coordination and review.

Yet still we find our gaze retuning time and again to the heat map. There is clearly an aspect of information contagion at work – for both good and ill. Bad ideas seem to have the same stickiness and spread as did accurate analysis. And despite the high levels of reorganization and mobility within specific offices, the contagion seems to have persisted. While in the IC we might immediately recognize the office as a “sick shop” with poor management or destructive internal dynamics, we cannot so blithely make the same statements regarding a corporation with which we are nowhere near so familiar.

Overall, and irrespective of specific elements of the discussion, we find ourselves wondering at this glimpse of what production management and consumer outcomes could look like in a transformed IC. For those that have advocated tracking more closely metrics for assessing analytic performance, this is perhaps the best model of the potential utility of such efforts we have ever seen. One could imagine similar representational overlays charting production output, consumer feedback, citation, or even employee retention / satisfaction.

This is definitely something to ponder at length.