The Role of Explanatory Considerations in Updating

June 10, 2015

Heads up for any C&I readers interested in probabilistic models and how they relate to the psychology of updating, check out the following two articles coauthored with Igor Douven.  Both were published in the last month, and both are freely available at the moment.

I’m especially proud of this joint work, which defends explanationist (and probabilist) models of human learning over and above prevailing Bayesian models.  For more detail, abstracts are below the fold…

Read the rest of this entry »

Invariant Generalizations and Causal Explanation

February 20, 2010

I am re-reading the chapters devoted to causal explanation of Woodward’s Making Things Happen: A theory of causal explanation.  The book is very interesting and ambitious and probably Woodward is offering there one of the most complete and attractive theories of causal explanation available today.  I am sympathetic to some of the main ideas.  The purpose of this note is to indicate some potential problems with the general account defended by Woodward.

Some background first.  Woodward proposes necessary and sufficient conditions for a generalization to represent a causal or explanatory relationship.  The idea is that the generalization should be invariant under some testing interventions on variables occurring in the relationship (page 253).  This simple formulation presupposes a lot.  For example, it presupposes that the generalization is expressible in a functional form Y = f(X). The technical term “testing intervention” needs to be explained as well. Consider an intervention (meeting the conditions specified by Woodward in chapter 3 of his book) that changes the value of X, say x_{0}, that presently holds to some other value x_{1}, where x_{1} is claimed by the generalization to be associated with a value of Y that is different from the value associated with x_{0}. So, if G abbreviates our generalization we have that x_{0} is different from x_{1} and G(x_{0}) = y_{0} is also different from  G(x_{1}) = y_{1}. Such interventions are called “testing interventions.”

Read the rest of this entry »