Making Things Happen: A Theory of Causal Explanation by ...

Mind Association

Making Things Happen: A Theory of Causal Explanation by James Woodward Review by: Peter Menzies Mind, Vol. 115, No. 459 (Jul., 2006), pp. 821-826 Published by: Oxford University Press on behalf of the Mind Association Stable URL: . Accessed: 30/11/2014 12:35 Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . . JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@. .

Oxford University Press and Mind Association are collaborating with JSTOR to digitize, preserve and extend access to Mind.



This content downloaded from 130.49.198.5 on Sun, 30 Nov 2014 12:35:09 PM All use subject to JSTOR Terms and Conditions

BookReviews 821

Making Things Happen: A Theory of Causal Explanation, by James Woodward.Oxfordand New York:OxfordUniversityPress,2004. Pp.vi + 410.

H/b ?37.50.

A revolution has taken place in the study of causation over the last decade. Drawingon an older traditionof researchin experimentaldesign and econometrics,researchersin computerscience,philosophy,and statisticshavedeveloped the theory of causalBayesnets as a frameworkfor the scientificstudy of causation.Twobooks stand out as landmarksin this field of research.One is

PeterSpirtes,ClarkGlymourand RichardScheines'1993book (now in its second edition 2000) Causation,Predictionand Search.The other is JudeaPearl's

2000 book Causality.These books develop in differentways an extensiveand complex theory of causalinference.They show how causalBayesnets, which are essentiallydirectedacyclicgraphswith nodes representingrandom variables,can playa systematicrole in tasksof causaldiscovery,inferenceand prediction. Very little of this technical knowledge has, however, percolated throughto philosophersuntil now.

The book under reviewwill help to makethis forbiddinglytechnicalliterature accessibleto philosophers,not becauseit is a philosophicalprimerto the theory of causalBayesnets, but becauseit decomposesthe theory,subjectsits underlyingconceptsto carefulscrutiny,andthen reassemblesthem in an original theory of causation,all of this in a way that is especiallysensitiveto philosophical issues. WhereasSpirteset al. and Pearlareconcernedwith practical problemsof causalinferencefromstatisticaldata,Woodwardis concernedwith

the semanticor interpretativetaskof explainingthe meaningof causalclaims. Forexample,Spirteset al. and Pearlassumea primitiveconceptof causalrelevance or of causal mechanism and use this concept to define other concepts requiredto estimate quantitativemagnitudes and to make causal inferences and predictions.In contrast,Woodward'sgoal is to tryto explainthe veryconceptof causalrelevanceor causalmechanismtakento be primitiveby the theoristsof causalBayesnetsbylinkingtheseconceptswith thatof intervention.On his account,a claim that a variableX is causallyrelevantto anothervariableY means roughly that an ideal intervention that changes the value of X will change the value of Y.What is distinctive about causal relationshipson this account is that they arepotentiallyexploitablefor manipulationand control. Woodwardcallshis theory a manipulabilityor interventionistaccountof causation. One of his goals is to show that this theory is free from many of the problemsthathavemadeearliermanipulabilityor agencytheoriesseem objectionableto philosophers.Thus he arguesthat when the notion of an intervention is appropriatelydefined, his account of causation does not attempt to reducecausalconceptsto conceptsof agencyor manipulation;nor is it anthropocentricin linkingcausationto actualhumanmanipulativecapacities.

I think thatWoodwardis magnificentlysuccessfulin his goals in this book. The interventionist theories of causation and explanation developed in the book aresubstantialand importantcontributionsto the literature.In my view,

Mind, Vol. 115 . 459 . July2006

? Mind Association 2006

This content downloaded from 130.49.198.5 on Sun, 30 Nov 2014 12:35:09 PM All use subject to JSTOR Terms and Conditions

822 BookReviews

this is one of a smallhandfulof philosophicalbooks on causationandexplanation publishedin the lasttwentyyearsthatmakesa significantandlastingcontribution to the study of these subjects.Becauseof the clarityand cogency of its argument and the richness of its theoretical framework, the book will

inform and guide philosophical thinking about causation, explanation and relatedsubjectsfor a long time to come.

Chapter one sets out the philosophical orientation of the work and lays down some desideratafor an acceptabletheory of causationand causalexplanation.Woodwardemphasizesthatour practicesof makingcausalandexplan-

atory claims have a practical purpose or point, which can be deployed to motivatedescriptivetheoriesof whatwe mean by these claims,but alsoto jus-

tify normativerecommendationsaboutwhatwe should meanby them.

Chaptertwo providesthe basic frameworkfor the interventionistapproach

to causationand explanation.The chapterexplainsthatthe fundamentalpoint of makingcausalclaims,as opposed to simple claimsabout correlations,is to isolate those relationships that are invariant under intervention and hence potentiallyexploitablefor the purposesof manipulationand control. To provide a formal frameworkfor articulatinghis theory of causation,Woodward introducesthe formalapparatusof directedgraphsand structuralequationsof the causal Bayesnets tradition. These are useful structuresfor representing

general and specific causal dependences between variables. Woodward's immediategoal is to spell out exactlywhat these causaldependencesamount to. He distinguishesa numberof differentcausalconcepts-direct cause,total cause,and contributingcause-and givesa precisedefinitionof eachin terms of whatwould happenwhen thereis an interventionon one or morevariables. Forexample,X is definedto be a directcauseof Ywith respectto some varia-

ble set V if and only if thereis a possible interventionon X that will change Y (or the probabilitydistribution of Y)when all other variablesin V besides X and Yareheld fixed at some value by interventions.The causalconcepts that Woodwardis mostlyconcernedwith aretype-causalconcepts.But some space is devotedto developingan accountof whatWoodwardcallsactualcausation, or what is more commonly called singular or token-causation. Woodward's accountof actualcausationis closelyrelatedto ones developedby Pearlandhis collaboratorJosephHalpern,and by ChristopherHitchcock.

Chapterthreeis devotedto providingan accountof the fine-grainedcausal structureof an intervention.The roughideaWoodwardtriesto captureis that an interventionon a variableX with respectto a secondvariableYis a process that fixesthe value of X in an exogenousway and changesthe value of Yonly by virtue of changingthe value of X. Clearly,this notion of an interventionis causalin characterb, ut this factdoes not renderthe interventionistaccountof

causationvacuousbecausein generalwe need not knowwhetherX causesYin orderto determinewhethersomethingcounts as an interventionon X. Woodward arguesthat earliermanipulabilityor agencytheories of causation have gone astrayin tryingto providereductiveanalysesof causalconceptsin terms

Mind, Vol. 115 . 459 . July2006

? Mind Association 2006

This content downloaded from 130.49.198.5 on Sun, 30 Nov 2014 12:35:09 PM All use subject to JSTOR Terms and Conditions

BookReviews 823

of agency-relatedconcepts.A virtue of his accountis that it can makesense of causalrelationshipsholdingeven in circumstancesin whichit is nomologically impossible for human beings to intervene and manipulate variables. He attemptsto distancehimself from the anthropocentrismof earliermanipulability theoriesby stressingthe objectivecharacterof the causalrelationship:it is built into the very concept of causationthat it is a relationshipthat would hold irrespectiveof whether it is physicallypossible for us to carryout any

manipulations.

Chapters four and five move the focus of discussion from causation to

causalexplanation.Chapterfour developsa powerfulcaseagainstthe standard deductive-nomological model of explanation:it arguespersuasivelythat the subsumptionof explanansand explanandumunder a law is neithernecessary nor sufficientfor causalexplanation.Chapterfive presentsa positive account of causalexplanation,accordingto which causalexplanationconsistsin exhibiting patternsof counterfactualdependence.To explainsome phenomenon is to locate it in spaceof alternativepossibilitiesand to show how it would have been differentif differentinitialconditionshadbeen realized.Woodward'sslo-

gan is that causalexplanationis successfulto the extentthat it answersa series of implicit what-if-things-had-been-differenqt uestions about the explanandum. Such information is explanatorypreciselybecause it is relevantto the manipulationand controlof the explanandum.

Chaptersix exploresthe notion of invariancethat is centralto Woodward's framework.It arguesthat a necessaryand sufficientconditionfor a generalization to represent a causal or explanatory relationship is that it be invariant undersome interventionson variablesoccurringin the relationship.Generalizations can be invariantto differentdegreesdepending on the number,kind and significanceof the interventionsunderwhich they arestable.Woodward employsthe notion of invarianceto reframefamiliarphilosophicaldoctrines. Forexample,he notes that many of the generalizationsthat play a role in the specialsciencesdo not satisfymost of the standardcriteriaforbeing a law.The usual move by philosophers intent on saving the doctrine that successful explanationsappealto laws is to treatthese generalizationsas second-grade, ceterisparibuslaws. However,this conceptual doublethink is not necessary, Woodwardargues,if we see thatthe distinctivefeatureof these generalizations is that they are relatively invariant under interventions on their variables, despitethe factthatthey arenot exceptionless.

Chaptersevenappliesthe ideasaboutinterventionsandinvarianceto illuminatethe natureof structuralequations,includingthe regressionequationsthat are used extensivelyin the biomedical, behaviouraland social sciences. It is arguedthatif a structuralequationis to be interpretedcausallyit mustbe invariantundersome interventionson its right-handsidevariables.The chapteralso exploresthe desirabilityof requiringthat explanatorysystemsof equationsbe modular in the sense that intervening on one equation does not disrupt the otherequations.Thesignificanceof imposingthis requirementis thatit ensures

Mind, Vol. 115 . 459 . July2006

? Mind Association 2006

This content downloaded from 130.49.198.5 on Sun, 30 Nov 2014 12:35:09 PM All use subject to JSTOR Terms and Conditions

824 BookReviews

that eachequationin a set of equationsrepresentsa distinctcausalmechanism thatcanbe replacedor disruptedindependentlyof the others.

Chapter eight compares the proposed account of explanation with the causalmechanicalaccountdevelopedby WesleySalmonand the unificationist account developedby Philip Kitcher.It arguessuccessfully,to my mind, that the causal-mechanicalaccountcannot,without being completelyrecast,make the crucialdiscriminationbetween those propertiesthat arecausallyrelevant to some outcome and those that arecausallyirrelevant;and that the unificationist account exploits the wrong sort of generality-a generalityof scope rather than invariance-in trying to explain how one explanation can be deeperthan another.

As this briefsummaryindicates,Woodward'sbook castsnew light on many familiarcontroversiesabout causation and explanation,but also raisesa fair numberof interestingnew problemsand issues,especiallyin connection with its reformulationof the manipulabilityapproachto causation.I want to consider brieflyjust one issue concerning this reformulation.Woodwardmakes much of the subjectivismof earlyagencyand manipulabilitytheoriesof causation, targetingin particulara theory once defendedby Huw Priceand myself to the effectthat our concept of token causationis groundedin the primitive conceptwe haveas agentsof interveningin the worldto bringaboutone event by meansof another.I thinkthatmanyof Woodward'scriticismsof this theory strikehome. I am persuadedthat it is muchbetterto formulatea manipulability accountthateschewsreductionby usinga notion of interventionthatbears its causalcharacteron its face;and that is betterto avoidanthropocentrismby

appealingto a notion of interventionthat is generalenough to applyeven in cases where human manipulation is physicallyimpossible. None the less, it seems to me thatWoodward'stheory is not as full-bloodedlyrealistas he sug-

gests.

Forinstance,considerhow Woodward'smanipulabilitytheory of causation would answerthe Euthyphroquestion: Does an intervention on X change Y becauseX causesY,or doesX cause Ybecausean interventionon X changesY?

Traditionally,realistschoose to answerthe Euthyphroquestion by accepting the first alternative. Woodward writes at times as though this is what he believes. For example: 'When an intervention changes C and in this way changesE,this exploitsan independentlyexistingcausallinkbetweenC andE. One can perfectly well have the link without the physical possibility of an

interventionon C' (p. 132).And elsewherehe writes:'... quite independently of our experienceor perspectiveas agents,there is a certainkind of relationshipwith intrinsicfeaturesthatwe exploitor makeuse of when we bringabout B by bringingaboutA' (p. 125).But I wonderwhetherhe reallywantsto maintain this. To do so would be to suppose that the causal relation has some intrinsic naturethat is only contingently associatedwith invariantrelationships between variables. But this supposition surely runs counter to whole tenor of his interventionisttheory,which might meaningfullybe expressedas

Mind, Vol. 115 . 459 . July2006

? MindAssociation 2006

This content downloaded from 130.49.198.5 on Sun, 30 Nov 2014 12:35:09 PM All use subject to JSTOR Terms and Conditions

BookReviews 825

the view that the whole nature of the causal relation between X and Y is

exhaustedby the fact that an actual or hypothetical experiment on X would change Y It seems to me that the manipulabilitytheorist of causationshould answerthe Euthyphroquestionby embracingthe second alternative:X causes Ybecausean interventionon X would changeY.But embracingthis is to take one stepbackfrom a full-bloodedrealismaboutcausation.

Another respectin which Woodwardoverstateshis theory'srealistcredentials, or ratherunderstatesits anti-realistimplications,is by downplayingthe representation-dependentfeaturesof the causalconcept. It turns out that his definitionsof importantcausalconceptssuch as those of a directcause,a total cause,and a contributingcausearerelativeto a variableset:whethera variable is a direct,totalor contributingcauseof anotheris sensitiveto whichvariables areused to representthesecausalrelationships.(Indeed,it turnsout thatthese causalrelationsalso dependon the rangeof allowablevaluesthe variablescan take (p. 56).) As Woodwardremarks,this fact is a surprisingbut inevitable consequenceof using systems of structuralequations and directedgraphsto representcausalrelationships.Within this framework,the choice of variables is crucialfor the representationof causalrelationships:the variableschosen reflect the possibilities one is willing to take seriously. Indeed Woodward exploits this feature of the framework for representing causal relations to explain a featureof the concept of actual or token causationthat is puzzling from the perspectiveof his theory (pp. 86-91). Our application of the token causal concept tends to be highly selective. For example, we judge that the omission of a supervisingdoctor to administera medicine to a patientcaused the patient'ssubsequentillness,but we do not judgethat a similaromission by a doctor who had no responsibility for the care of the patient caused the patient's illness, even though it is true in both cases that if the doctor had administered the medicine, the patient would not have suffered the illness. Woodwardexplainsthe differencein our causaljudgementsaboutthese examples in terms of our willingnessto takeseriouslycertainpossibilities.It was a seriouspossibilitythatthe firstdoctorshouldhaveadministeredthe medicine, but not a seriouspossibilitythatthe seconddoctorshouldhavedone so. These judgementswould be reflectedin our representationof the causalstructureof the situation by the inclusion of a variable representing the first doctor's action,but not a variablefor the seconddoctor'saction.

All of this seemsplausibleenough as an explanationof the selectivityof our actualor tokencausaljudgements.None the less, the crucialroleplayedby the choice of variablesin the representationof causalrelationsraisessome serious

questionsaboutthe extentto whicha full-bloodedrealismaboutcausationcan

be sustained.The fact that the correctapplicationof causalconceptsdepends on certainkey representationalchoices seems to imply that the truth-conditions of causalconceptsarenot completelyobjectiveor mind-independent.

It is a pity that Woodwarddoes not explorethese mattersat greaterlength since they impingeon foundationalquestionsabouthis chosen frameworkfor

Mind, Vol. 115 . 459 . July2006

? Mind Association 2006

This content downloaded from 130.49.198.5 on Sun, 30 Nov 2014 12:35:09 PM All use subject to JSTOR Terms and Conditions

826 BookReviews

explainingcausalconcepts.Otherresearchersin the causalmodellingtradition

are more up-front about the representation-dependent characterof causal conceptsin assertingthatcausalclaimsmust alwaysbe understoodrelativeto a model. On this surfacethis talkis puzzlingfrom a philosophicalpoint of view.

What is a model?Is it an abstractmathematicalobject or a mental construc-

tion?How is causationrelativeto a model?Does this just mean thatthe truthconditions of causal concepts are relativizedto models in the same way that context-sensitiveexpressionsarerelativizedto contextualparameters?Or does

it have some more radically subjectivist meaning? Such questions that are raised by the recent philosophical borrowings from the modelling tradition deservesustainedphilosophicaldiscussion.Woodwarddoes not addressthese questions partlybecausehe is eagerto downplayany departurefrom realism on the part of his theory. This is a pity because I am sure that Woodward would havegivenan extremelyacutephilosophicalexplanationof this talk.

In conclusion,these minor quibblesaside,this is a splendidbook. It is beautifully and clearlywritten; and in countless ways sheds a flood of light on a rangeof topics to do with causation and explanation.It representsthe most

significant and substantial philosophical contribution to the study of these conceptsin recentyears.Readingand studyingthis book will be obligatoryfor everyone whose work bears directly or indirectly on the topics of causation and explanation.

Department of Philosophy

MacquarieUniversity NorthRyde,NSW2109 Australia

pmenzies@scmpm.q.edu.au

doi:lo0.o93/mind/fzl821

PETERMENZIES

Mind, Vol. 115 . 459 . July2006

? Mind Association 2006

This content downloaded from 130.49.198.5 on Sun, 30 Nov 2014 12:35:09 PM All use subject to JSTOR Terms and Conditions

 ................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download