The following reprints and technical reports are now
available on our web site:
http://bayes.cs.ucla.edu/jp_home.html
Hard copies can be obtained from kaoru@cs.ucla.edu.
Would value your comments and thoughts,
====== Judea Pearl
Annotations begin with *****
- --------------------------------------------
Pearl, J., ``Graphical Models for Probabilistic and
Causal Reasoning,'' in Allen B. Tucker, Jr. (Ed.),
The Computer Science and Engineering Handbook,
Chapter 31, CRC Press, Inc., 697--714, 1997.
**** Surveys the transition from Bayesian-nets to causal-nets.
Galles, D. \& Pearl, J., ``Axioms of Causal Relevance,''
{Artificial Intelligence}, 97(1-2), 9--43, 1997.
*****Proposes semantics for causal and counterfactual
expressions, axiomatizes the expression: "X has no effect
on Y once we fix Z", and introduces graphical methods of
proving theorems on such expressions.
Pearl, J., ``Graphs, Structural Models and Causality,''
Prepared for "Causation, Bayes Networks, and Machine Discovery"
AAAI/MIT Press.
*******Extends my Biometrika paper (1995) with examples,
explanations and alternative approaches.
Balke, A. \& Pearl, J., ``Bounds on Treatment Effects from Studies with
Imperfect Compliance,'', Journal of the American Statistical
Association (JASA)}, 92(439), 1171--1176, 1997.
***** Results originally reported in UAI-94 are here
presented to hard-core statisticians .
Pearl, J., ``TETRAD and SEM,'' Commentary on The TETRAD Project.
R-244, June 1997.
To appear in Multivariate Behavioral Research, Volume 33(1).
***** Attempting to show structural-equation practitioners
that projects like Tetrad can help revitalize their methodology.
Greenland, S., Pearl, J., and Robins, J., ``Causal Diagrams for
Epidemiological Research.'' Epidemiology 1998, in press.
(**********Attempting to show epidemiologists that graphs
can answer questions that rank-and-file epidemiologists are
eager but afraid to ask.
Galles, D. \& Pearl, J., ``An Axiomatic Characterization of
Causal Counterfactuals,''
R-250, 1998. To appear in Foundations of Science,
Kluwer Academic Publishers, vol 4(1) 1998.
****** Arguing for the thesis that human
judgment of counterfactuals is not based on assessing similarity
between worlds but on mental manipulation of causal structures.
The axioms that govern the two conceptions are strikingly
similar though, and the end of the paper presents examples of how
probabilities of counterfactuals can be derived by either
graphical or symbolic methods.
Pearl, J., ``Graphs, Causality, and Structural Equation Models,''
Technical Report R-253, 1998. To appear in
Sociological Methods and Research, Special Issue on
Causality, Nov. 1998.
**** Attempting to convince social scientists that
causality can now come out of the closet, and that much
of their frustration with available methodologies can be relieved
by taking seriously the notion that a set of equatsions is
"structural". Bags of tools are then provided for
readers who are convinced.
Pearl, J., ``Why There Is No Statistical Test For Confounding,
Why Many Think There Is, and Why They Are Almost Right,''
Technical Report R-256, 1998.
**** Written for epidemiologists, the paper argues that the
only statistical tests that have anything to do with
confounding are those that insist on "stable unbiasedness",
namely, effect estimates that remain unbiased under all values
of the model parameters.
Pearl, J. and Meshkat, P., Report R-257, "Testing regression
models with fewer regressors" June 1998.
******We normally test Bayesian networks using the parent-screening
condition: Given its parents, each variable ought to be independent of
its other predecessors. More economical tests can be applied
in the case of Gaussian variables (which yields compositional
graphoids); we show that almost any separation set between
nonadjacent variables would be sufficient for testing the
network structure.
Pearl, J., "On the Definition of Actual Cause",
Tech Report R-259, June 1998 (Draft).
*****What do we (or lawyers) mean when we argue
that an action was the "ACTUAL CAUSE" of a given event?
or that one events "EXPLAINS" another?
Evidently, commonsense (and juries) accept causal connections
which do not pass the standard counterfactual test
that the effect should not occur in the absence of the
cause (e.g., this house would have burned down anyway)
This renders the standard calculations of counterfactual
probabilities (e.g., Balke-Pearl 1994) incomplete for
evaluating probabiities of "actual causation."
The paper proposes an explication of such causal
expressions in the language of structural models .