7

I am trying to find some data structures/mathemetical theories to represent causal relationships which differ from graphical models or Bayesian Networks. Any ideas?

Neel Krishnaswami
  • 9,013
user6813
  • 71
  • There may be some usefule information found at an earlier question (which generated quite a bit of debate): http://mathoverflow.net/questions/22490/correlation-and-causation-when-can-we-believe-correlation-reasonably-at-least – Harald Hanche-Olsen Jun 15 '10 at 11:52
  • I would say that most of it can be cast in terms of adapted filtrations. – Grant Izmirlian Jan 07 '14 at 21:20

3 Answers3

4

The standard account of causality is Lewis's theory of counterfactuals. He wrote a small, very readable book called Counterfactuals, which the SEP summarizes here. The idea is to take the viewpoint of modal logic, and interpret the counterfactual conditional $A \;\boxdot\!\!\!\to B$ (read "if A were the case, then B would be the case") as holding if $B$ holds in the nearest world in which $A$ holds. Obviously this requires enriching modal logic with a notion of similarity of worlds, as well.

A lot of other people have worked on this subject; personally I am most fond of McCain-Turner causality, since Graham White has given it a nice proof-theoretic formulation.

Neel Krishnaswami
  • 9,013
  • 2
    I wouldn't say there's a single standard account of causality. For a discussion of further accounts - mechanistic, probabilistic, agency - try http://www.kent.ac.uk/secl/philosophy/jw/2004/causality.pdf. – David Corfield Jun 15 '10 at 09:00
  • That's a fair point -- I meant "standard" in the sense of "the one everyone feels obliged to argue with", rather than "the one everyone accepts". – Neel Krishnaswami Jun 15 '10 at 09:41
2

With several variables connected by asymmetric causal relations, it's not so likely that a mathematical theory of causality will escape graphical representation. Neel mentioned above Lewis's counterfactual analysis, and this has a close affinity with aspects of Judea Pearl's work on casual Bayesian networks, see p. 239 of http://books.google.co.uk/books?id=wnGU_TsW3BQC.

Among quantitative approaches, a variant on the usual statistical approach is Janzing and Schoelkopf's use of algorithmic dependence to determine causal relations: http://www.kyb.mpg.de/publications/attachments/paper_IEEE_version3_webseite_6526%5B1%5D.pdf.

David Corfield
  • 4,901
0

Burks, A.W. (1977), Chance, Cause, Reason : An Inquiry into the Nature of Scientific Evidence, University of Chicago Press, Chicago, IL.

Jon Awbrey
  • 393