Earlier in the month, there was a very interesting discussion over at Flowing Data entitled the Rise of the Data Scientist. We decided to highlight it in this post because it raises important issues regarding the relationship between Computational Legal Studies and other movements within law.
As we consider ourselves empiricists, we are strong supporters of the Empirical Legal Studies movement. For those not familiar, the vast majority of existing Empirical Legal studies employ the use of econometric techniques. For some substantive questions, these approaches are perfectly appropriate. While for others, we believe techniques such as network analysis, computational linguistics, etc. are better suited. Even when appropriately employed, as displayed above, we believe the use of traditional statistical approaches should be seen as nested within a larger process. Namely, for a certain class of substantive questions, there exists tremendous amounts of readily available data. Thus, on the front end, the use of computer science techniques such as web scraping and text parsing could help unlock existing large-N data sources thereby improving the quality of inferences collectively produced. On the back end, the use of various methods of information visualization could democratize the scholarship by making the key insights available to a much wider audience.
It is worth noting that our commitment to Computational Legal Studies actually embraces a second important prong. From a mathematical modeling/formal theory perspective, at least for a certain range of questions, agent based models/computational models ≥ closed form analytical models. In other words, we are concerned that many paper & pencil game theoretic models fail to incorporate interactions between components or the underlying heterogeneity of agents. Alternatively, they demonstrate the existence of a P* without concern of whether such an equilibrium is obtained on a timescale of interest. In some instances, these complications do not necessarily matter but in other cases they are deeply consequential.
6 thoughts on “The Rise of the Data Scientist [From Flowing Data]”
agent based models/computational models > closed form analytical models
This seems way too strong. There are lots of well-known problem with agent based/computational models too, like extreme sensitivity to initial conditions and general opacity and lack of generality. Why not the more moderate “there’s a time and a place for each sort of modeling?”
Maybe that is too much… I made an edit. It is often an issue which is rarely ever discussed. In observing the presentation of a formal model…I very rarely hear a discussion of the impact of dependancy, heterogeneity, etc.?
Certainly to the extent those are problems, they should be discussed. It’s worth some more thought about the scope of those issues, though.
I’m not totally sure what you mean by dependency/interactions with components. At least for game theoretic models, that kind of dependency/interaction is, of course, the whole point of the method, if we’re talking about interaction between agents. Are you talking about interactions between some other features of the model, perhaps parameter values or something? It would be helpful to have an example of the sort of interaction you’re concerned with.
I certainly agree that heterogeneity is a problem with some analytic models. But many game theoretic models are about a fairly small number of players, so that the heterogeneity of agents issue shouldn’t be too much of a problem, at least if I have the right sense of what you mean. They become a bigger issue when we’re talking about, e.g., large-scale coordination games, but many of those models explicitly incorporate some kinds of heterogeneity, i.e., diverse preferences. (For an example that just happens to be on the top of my head, see Barry Weingast’s “The Political Foundations of Democracy and the Rule of Law.” APSR 91:245-63 (1997), which builds a model of constitutionalism that permits agents with diverse preferences to coordinate on sanctioning the state.) I’m not sure how large the set of models is where this issue is both important and unaddressed.
So sorry, not trying to have an argument with you … We are going to have to respectfully disagree. Maybe we will post on this in the coming weeks…
If you want to learn more, check out the syllabus I posted earlier in the month … I believe the readings (particularly the foundation readings) will lay out the case and clear up much of your confusion.
Sure, will doubtless be worth a read.
Aloha!wpzy! http://pwdnoz.com bfvnm tktfr http://pfzofa.com zvbqd yoeun http://pnvgxt.com narue hpasn http://yqfdlo.com mjqxd rfaff http://zxefoo.com awmeu oykqa