Miers's Withdrawal, Cascades, and Information Markets

Cass's claim in his blog here yesterday was provocative because it claimed that one could discern a cascade.  Most of the comments were skeptical, though I must admit that I agreed with Cass's observation - hard as it is to test empirically.  The subsequent announcement by Miers of her withdrawal makes the cascade possibility yet more interesting.  So here is a further claim that I cannot prove: information markets (and perhaps other methods of communication) accelerate cascades, though we may want to call this effect something other than an information cascade.  Think of the position of a senator, and even of an academic or pundit or blogger (some overlap there), who sees an issue in play.  Some academics gain reputation from, or are by nature, contrarian, but in most walks of life it is useful to be right.  A senator normally expends political capital by backing a horse, but the expenditure is usually unprofitable if it is the wrong horse; a political commentator gains a following by predicting outcomes correctly; and even a business person or academic gains in reputation when events transpire as the person predicted.  The prognosticator is said to have good judgment.  In turn, if information markets are good predictors, perhaps because they aggregate information or judgment from dispersed parties in settings where many heads are indeed better than one, then one ought to look at the information markets (or proxies for them) and go along with the prediction.

And so, once Tradesports.com and other information markets predicted that Miers would not be confirmed, it was worthwhile for an observer who wished to save political capital or to be credited with good judgment to go with that flow.  If the market thinks failure is 55% likely, then it might well pay for the pundit to speak as if 95% certain, so to speak.  My conjecture is that a cascade-looking process can come about not so much because people are influenced by what they hear but because they want to be seen as being reliable or discerning, and so they rush to align themselves with what they think is the likely outcome.  (And yes, in the long run in a rational-actor world, listeners will want to know whether the speaker outperformed information markets, but that world is a long way off.)

Comments

[A.Hamilton]

This is interesting and I agree. I think it is important to clarify a particular point.

Your analysis seems spot on, to me, only after payoffs are taken into account. So, in the Miers case, if a correct prediction on some particular date adds 10 to a senator's political capital, but an incorrect prediction will cost the senator 1000, then the senator has to obviously weigh his or her options more carefully. The decision maker cannot only look to the odds. (Understanding of course that an initial 55% negative position on Miers could turn to a 95% negative positive quickly).

[Paul Gowder]

That's a really interesting idea -- but I'm not sure what it means. Pundits have an incentive to make accurate predictions from whatever source derived. One data point that a pundit might use to make a prediction is the result of some information aggregating market, and then presumably that prediction (to the extent it is seen as meaningful) will be fed back into that market. But why do we need tradesports.com or any other market to do this? If I get on TV and make an argument (totally unhinged from any market) that X will happen, and the argument is persuasive, others will follow with similar arguments, and the same cascade-looking effect will follow, non?

(And, of course, one must avoid the genetic fallacy in all of this.)

[Cory Hojka]

Concerning evidence of cascades, I think that one approach allowing us to measure a "cascade" effect would be to use blogs. For example, with the former Miers nomination we would find all the blog entries that discuss Miers. Then we examine them to see whether people are speaking positively or negatively about Miers (and perhaps quantify it even further by noting what basis they use to make their remarks, such as judicial qualification, cronyism, etc.). Next, since bloggers tend to link to others who they find persuasive, we could create a flowchart showing the links between all these different bloggers over time. Such a flowchart might indicate interesting patterns that could provide evidence of a cascade.

For example: Blogger A says he hates Miers because of her haircut. Bloggers B and C agree, make some comments of their own, and link back to A. Then some more bloggers jump on the bandwagon and link to A, B, and C. Etc.

Of course, there are going to be problems with trying to determine if this situation is actually a cascade. Bloggers may not always post links to others that have influenced them. Also, it may be somewhat of a judgment call what are comments consisting of the Blogger's own insight versus what came from being influenced by others.

Nonetheless, in situations where people are analyzing something where they can obtain very little meaningful information at the time (i.e, the Miers nomination, certain aspects of the build-up to the Iraq war, etc.), then I think that if such a situation occurs, and potentially other types of patterns too, that this would suggest evidence for the cascade theory.

[anon]

using blogs is an interesting idea, but you'll quickly run into google bombs. for instance, google 'failure' right now and see what you get. is there a reliable way of gathering evidence?

[Cory Hojka]

The analysis of blogs would probably take some careful considerations. I would assume that google bombs have some type of structure that is, at some point, recognizable. Thus, even if Google is initially misled by the tactics, at the time of analysis where we are looking at a past event we could probably detect and eliminate these fake blogs. In addition, we'd be looking for data that I think would stand out independent of the google bombs. There's probably no interest, at least currently, for a google bomb to represent what a cascade might look like. So even if we have misleading data in the analysis, the cascade effect might stand out as a separate and distinguishable pattern.