causal selection and counterfactual reasoning - william jiménez-leal

8/18/2019 Causal Selection and Counterfactual Reasoning - William Jiménez-Leal

1/20

REVISTA COLOMBIANA DE PSICOLOGÍA VOL. 22 N.º 1 ENERO-JUNIO 2013 ISSN 0121-5469 IMPRESO | 2344-8644 EN LÍNEA BOGOTÁ COLOMBIA

Causal Selection and Counterfactual ReasoningSelección Causal y Razonamiento Contrafactual

Seleção Causal e Raciocínio Contrafactual

-Universidad de los Andes, Bogotá, Colombia

Correspondence concerning this a rticle should be addressed to Will iam Jiménez-Leal, e-mail: [email protected]. Department of Psychology,Universidad de los Andes, Cr. No. A- , Building Franco, nd oor, Bogotá, Colombia.

: - :

* Special thanks to Nick Chater and an anonymous reviewer for their relevant comments.

Abstract

In this paper I defend the view that counterfactualthinking depends on our causal representation ofthe world, and in this sense, I argue that causal andcounterfactual reasoning are tightly linked. I offersome criticisms and experimental evidence againstMandel’s judgement dissociation theory (Mandel,

b), which claims functional independencebetween the process of causal selection and coun-terfactual reasoning in the context of causal selec-tion. In the experiments described, I manipulatedsome elements of the semantics of the task to showthe cases in which dissociation between causaland counterfactual reasoning does not occur. InExperiment , the level of description of the targetevent is manipulated in a list generation and ratingtask. Experiment replicates Experiment nd-ings using an alternative coding system, whereasExperiment does the same using an alternativeanswer format. e results of the experiments sup-port the picture of causal understanding proposedby the causal mental models.

Keywords: counterfactual reasoning, causality, judgement dissociation theory

Resumen

El trabajo deende la posición según la cual elpensamiento contrafactual depende de nuestra re-presentación causal del mundo y, en este sentido,argumenta que existe una estrecha relación entreel razonamiento causal y el contrafactual. Se lleva acabo una crítica a la teoría de la disociación de jui-cios de Mandel (Mandel, b), que deende laindependencia funcional entre el proceso de selec-ción causal y el razonamiento contrafactual en elcontexto de la selección causal. En los experimen-tos realizados se manipularon algunos elementosde la semántica de la tarea con el n de ilustraraquellos casos en los que no se da la disociaciónentre el razonamiento causal y el contrafactual.En el Experimento , el nivel de descripción delevento objetivo se manipuló en una tarea de ge-neración de listas y evaluación. El Experimento replicó los hallazgos del Experimento utilizandoun sistema de codicación alternativo, mientrasque el Experimento realizó lo mismo utilizandoun formato de respuesta alternativo. Los resulta-dos de los experimentos apoyan la concepción delentendimiento causal propuesta por los modelosmentales causales.

Palabras clave: razonamiento contrafactual , cau-salidad, teoría de la disociación de juicios.

Resumo

O trabalho defende a posição segundo a qual opensamento contrafactual depende de nossa re-presentação causal do mundo e, nesse sentido,argumenta que existe uma estreita relação entreo raciocínio causal e o contrafactual. Realiza-seuma crítica da teoria da dissociação de juízos deMandel (Mandel, b), que defende a inde-pendência funcional entre o processo de seleçãocausal e o raciocínio contrafactual no contextoda seleção causal. Nos experimentos realizados,manipularam-se alguns elementos da semânti-ca da tarefa com o objetivo de ilustrar aquelescasos com os quais não se dá a dissociação en-tre o racionamento causal e o contrafactual. NoExperimento , o nível de descrição do eventoobjetivo se manipulou em uma tarefa de geraçãode listas e avaliação. O Experimento repetiuas descobertas do Experimento ao utilizar umsistema de codicação alternativo, enquantoo Experimento realizou o mesmo ao utilizarum formato de resposta alternativo. Os resulta-dos dos experimentos apoiam a concepção doentendimento causal proposta pelos modelosmentais causais.

Palavras-chave: raciocínio contrafactual, causa-lidade, teoria da dissociação de juízos.


2/20


3/20

18 1

REVISTA COLOMBIANA DE PSICOLOGÍA VOL. 22 N.º 1 ENERO-JUNIO 2013 ISSN 0121-5469 IMPRESO | 2344-8644 EN LÍNEA BOGOTÁ COLOMBIA - PP. 179

CAUSAL SELECTION

reported here are based on the idea that counter-factual reasoning can help the process of causalselection when the relevant events are identiedin the same level of description, a principle im-plied by Bayes nets.

Judgement Dissociation Theory andCounterfactual Reasoning

Mandel ( b) puts forward to con-ceptualize the apparent independence betweenthe process of causal selection and counterfac-tual reasoning. e operation of is basedon two principles, actuality and substitution.According to the actuality principle, causal

selection is based on identifying sufficientantecedents for the actual outcome under con-sideration. According to the substitution princi-ple, counterfactual and covariational reasoningdepend on the generation of ad hoc categories(or norms) that systematically focus on elementsthat can undo an outcome, called preventors.

e two principles, that operate independently,are based on different information (Mandel &Lehman, ), have different targets and do notinuence each other (Mandel, a).

e alternative, currently out of favour, isthe counterfactual process view of causal think-ing or counterfactual simulation theory. is isa loose of group of proposals (N’gbala & Brans-combe, ; Petrocelli, Percy, Sherman, & Tor-mala, ; Wells & Gavanski, ; Wells, Taylor,& Turtle, ) that can be roughly characterisedby the idea that people identify the causes of anevent by performing a mental simulation of thenegation of the candidate cause.

Mandel ( a, b) points out some un-

desirable consequences of this perspective: theidea that causation is a relation of necessity, notsufficiency; and that understanding counterfac-tual statements is equivalent to understandingcausal statements. Clearly both are wrong andthus justify rejecting the simulation approachesin favour of .

Mandel uses cases of pre-emption to anal-yse and criticise counterfactual simulation ac-counts, echoing the discussion in philosophy(Collins, Hall, & Paul, ). Take as an examplethe case of Suzy and Billy. ey both throw arock at a bottle at the same time, but Suzy’s rockhits the bottle rst, breaking it. In this case, acounterfactual simulation does not allow deter-mining the cause since both counterfactual sim-ulations of the absence of the candidate causesfail to undo the effect. Consider statements( ) and ( ), as representing the correspondingsimulation.

. If Suzy had not thrown the rock, the bottle

would not have broken.. If Billy had not thrown the rock, the bottlewould not have broken.

In both cases the counterfactuals are false,indicating that eliminating the candidate causedoes not eliminate the effect, therefore the simu-lation fails to identify the cause.

In contrast, ’s actuality principle al-lows reasoners to identify the specic event thatbrings about an effect because people acknowl-

edge sufficiency as the hallmark of causality, anelement not represented in either counterfactualsimulation. e substitution principle predictsthat people will focus on preventors duringcounterfactual resasoning, elements that are alsoout of the scope of the rival theories.

Problems with JDT

ere are some problems with the way is specied. In what follows, I point out three:the notion of sufficiency and preventor, its inter-

pretation of counterfactual simulation, and thecontraposition of mechanism and dependenceinformation. e key concepts of “sufficiency”and “preventor” are underspecied. ere is nobroad agreement on how to dene them, andeven more importantly, it has been shown thatthe concepts of formal sufficiency and necessity


4/20

18 2

DEPARTAMENTO DE PSICOLOGÍA FACULTAD DE CIENCIAS HUMANAS UNIVERSIDAD NACIONAL DE COLOMBIA

WILLIAM JIMÉNEZ-LEAL

(on which is based) do not match people’sunderstanding of them (Verschueren, Schae-ken, & Schroyens, ). Some researchers haveconvincingly argued in favour of a contextualdenition of sufficiency and necessity (See Hart& Honoré, ; Hilton & Erb, ; Hilton, Jas-pars, & Clarke, ) and although Mandel’saccount of causal selection explicitly acknowled-ges it as a conversational process, he does notspecify any element of the conversational pro-cess that might inuence causal selection. us,the conversational processes inherent to causalselection as well as the characterisation of suffi-ciency and necessity under the circumstances

are le unexplained.On the other hand, Mandel’s concept ofcounterfactual reasoning is based on an illicitgeneralisation of the idea of the simulation heu-ristic (Kahneman, Slovic, & Tversky, ). With-in Norm eory (Kahneman & Miller, ), thesimulation heuristic operates as a post facto re-action that is motivationally relevant. erefore,not all counterfactual simulations aim to identifycauses (see Petrocelli et al., ). Furthermore,even when counterfactual reasoning is required

to answer a causal question, the interpretation ofthe question plays a key role in determining thenorm against which to compare. By taking the op-eration of the simulation heuristic out of context,as proposed by Norm eory, Mandel equates itsoperation with all counterfactual reasoning. Hefails to acknowledge that counterfactual contin-gencies are actualised by the demands of the task,and made available for causal selection.

Finally, Mandel wrongly opposes causalmechanism information to probabilistic in-

formation in their importance for causality at-tribution. Causal selection can rely on either ofthese sources, and what is more, in many casesthey provide equivalent information (Cheng& Glymour, ). Again, this could be due tothe lack of the conversational dynamics. WhereMandel wants to describe different questions(how vs. what) he describes different sources of

information. e difference lies in the task de-mand, not in the information itself.

ere are, however, good reasons why theobject of counterfactual reasoning can be mis-matched with the object of causal reasoning. Ibelieve the key notions that lacks are thoseof sensitive causation (Woodward, ) andmodal fragility of an event (Lewis, ). A causalrelationship is said to be sensitive to external fac-tors when it holds in the actual circumstances butwould not continue to hold in circumstances thatdepart, even slightly, from the actual. An insensi-tive causal claim holds in the actual circumstanc-es and would continue to hold across a range of

changes from the actual circumstances. To useone of Lewis’ examples, shooting a victim at pointblank through the heart is a case of causal insensi-tive causation since small variations will not alterthe outcome, whereas running into an old friendbecause you le your house late in the morning isa case of sensitive causation because it could haveeasily happened in a different way (for a similarposition see Menzies, ). Similarly, the causalrelata might be modally fragile, meaning that anevent occurring at a different time and in a dif-

ferent manner would be considered a modallydifferent event and thus would be subject to a dif-ferent set of counterfactual dependencies. us,death by a heart attack implies a different coun-terfactual history than death by gunshot, andboth are modally different than simply describingan event as an unqualied death.

I maintain that counterfactual reasoningonly allows singling out causal factors in casesof either insensitive causation or where the re-lata are modally robust. Psychologically, this

simply means that a counterfactual does notidentify the cause if the causal events can be in-stantiated in several alternative ways, and if thecausal link itself can be instantiated in severalways. In the end, this comes down to the waya causal relationship is represented in the con-text of explanation, which in turn depends onthe demands of the task or situation. can


5/20

18 3


CAUSAL SELECTION

explain cases of dissociation, clearly, but not thematches, because it lacks the theoretical tools tohandle changes in the way the representation ofthe causal relata are specied.

ere is also evidence that points to inte-gration, rather than to dissociation of causaland counterfactual reasoning. Byrne ( ) haspointed out that a good guide for understandingcounterfactual reasoning is to explore the rep-resentation of the factual possibilities associatedwith it. e appropriate modelling of the causalstructure might help to understand how causaland counterfactual reasoning relate, and at thesame time to provide a normative framework for

studying counterfactuals (Sloman & Lagnado,). is framework can easily incorporatethe notions of sensitive/insensitive causationand fragile/robust events.

An Alternative

According to the causal modelling frame-work (Glymour & Cooper, ; Pearl, ), thecausal structure of a situation constrains the kindof counterfactual inferences that are allowed.Sloman and Lagnado ( ) have used it to ex-

plore the issue of counterfactual reasoning indeterministic causal systems. eir main ndingis that “When reasoning about the consequencesof a counterfactual supposition of an event, mostpeople do not change their beliefs about the stateof the normal causes of the event” (p. ). at is,the causal structure against which counterfactu-als are judged is kept stable. Moreover, Slomanand Lagnado ( ) found that people correctlyidentify the outcome of imagined interventionsbased on counterfactual assumptions. eir con-

clusion is that causal inference follows the logicof intervention, that is, causal inference is deter-mined by counterfactually altering the values ofa variable, as part of a causal network.

How is it possible then to reconcile thesendings with Mandel’s results? Spellman and col-leagues (Spellman, Kincannon, & Stose, )proposed that dissociation is an order effect. ey

assert that in cases in which the same participantshave to complete causal and counterfactual tasksthe following congurations are possible: Whensubjects are asked rst to generate counterfactu-als, that information regarding the counterfactu-als then becomes available for causal evaluation,making it more probable to affect the performancein the causal task. However, when the causal taskis performed rst, this does not affect mutationsperformed a erwards. However, contrary toSpellman et al. ( ), Mandel ( a, b) didnot nd any order effects. is explanation is con-sidered in the following experiments.

An alternative explanation is that the mis-

match observed by Mandel occurs because causaland counterfactual queries are not usually speci-ed in the same way. In fact, Mandel’s counterfac-tual probes always refer to undoing the “outcomeof a situation” whereas causal questions refer to aparticular event (Mandel, a, p. ). In otherwords, causal queries relate to insensitive causalrelationships, whereas counterfactuals are centredon sensitive causal ones. Similarly, the instructionsfor the probability ratings he requested were notmatched. Second, the mismatch in the description

of the events in the tasks also leads to obscuringthe underlying causal structure of the situation.Once the structure is clear (causally insensitive), itis feasible that counterfactual and causal tasks canhave the same targets (they are modally robust).Causal queries convey cues that somehow specifythe content of the causal answer(s) they are in-tended to receive. us, a question about what isthe cause of a theoretical outcome has more roomfor interpretation than a question about the causeof a glass bottle breaking yesterday .

In summary, ambiguous elements that in- volve a certain degree of pragmatic interpreta-tion can be responsible for some of the cases ofdissociation between causal and counterfactualreasoning. e experiments reported below ma-nipulate these elements, the probe and the speci-city of the description to contrast the resultswith Mandel’s.


6/20

18 4



Experiment 1

Dissociation between causal and counter-factual reasoning is predicted to depend on thelevel of specicity of the description of the targetevents in each task. Results of the experiment areinterpreted contrasting with rival hypoth-eses. e presence of order effects, as suggestedby Spellman et al. ( ), is also investigated.

Method

ParticipantsSeventy-two undergraduate students (

female and male) from different programs at

the University of Warwick took part in the ex-periment in exchange for payment (£ . ). emean age was . ( = . ) and all participantssuccessfully completed the task.

Materials and Procedure Participants were tested individually in a

cubicle with a computer-based experiment. eselected stimuli were presented to participantson a computer screen using a program writtenby the author in the Delphi programming lan-

guage (Texeira & Pacheco, ). Participantsworked on the task at their own pace and all ofthem completed the tasks requested.

e complete display included eight screen-shots and followed the structure of Mandel’sexperiments. e rst screenshot contained thegeneral instructions, where it was emphasizedthat they would have the opportunity to read a vignette only once and then they would be askedquestions about it. e vignette was presented inthe second screenshot, where a criminal falls prey

to two assassination attempts. Briey, the rst as-sassin puts poison in his drink, which shouldtake one hour to have any effect. However, beforethe poison has killed him, the second assassinruns the criminal off the road. e criminal diesbecause of the explosion of the car.

Once the participants had read the scena-rio, they proceeded to complete the causal, the

counterfactual and the probability tasks. e or-der of the tasks was randomly counterbalanced.

e causal and the counterfactuals tasks consis-ted of option listing and rating the answers par-ticipants wrote.

For the causal task, participants were as-ked to list up to four factors that they “regardas causes of the ‘event’”. In the next screenshotthey were asked to rate the importance of the-se factors (from to ): “Now please rate theimportance of each factor you listed with regardto causality on a scale of - where ‘not at allcausal’ and ‘totally the cause’”.

e counterfactual task exhibited the same

structure, with participants rst asked to proposefour ways “in which the event would have beendifferent”, and then invited to rate from to “how likely those alterations would have been inchanging the ‘event’: ‘Please rate the importanceof each of the changes you listed with regard tohow likely that change would have been in alte-ring the event on a scale of - where ‘not atall a good way to undo the event’ death and ‘absolutely the best way to undo the event’”.

e description of the “event” varied in

three levels and for each level both causal andcounterfactual tasks were matched. ree levelsof specicity were dened for the event descrip-tion: For the rst or low level, the questions wereabout the “outcome of the situation”. For the se-cond or medium level, the judgements requiredwere about the “death of the main character”. Forthe third or high level, participants were askedabout the “death of the main character due to thefatal burns”.

Participants were randomly assigned to one

of the three versions dened by the specicitylevel of event description. Notice that the maindifference between these tasks and Mandel’s isthe variation of the event description and thematching of the description across tasks. In thecounterfactual task, Mandel ( b) asked peo-ple about ways the “story could be changed sothat the outcome would have been different” and


7/20

18 5


CAUSAL SELECTION

then rate how effective these were in undoingthe “character’s premature death”. In the causaltask the questions were simply about the deathof the character.

In the probability task, participants wereasked to estimate four probabilities (from to

) for the outcome, given four conditions, de-ned by the presence/absence of the actions ofthe assassins:

. None of them occurring.

. Poisoning but not car run off the roadoccurring.

. Car runoff road but not poisoning occurring.

. Both of them occurring.

For example, the sentence correspondingto condition was:

“What is the probability of the victim dyinggiven that neither Mr. Vincent added poison toMr. Wallace’s drink nor Mr. Bruce pushed Mr.Wallace’s car into a ravine.”

Coding of Causal and Counterfactual Listingse coding was done according to the ca-

tegories of interest: crime life, poison, crash and

other. Two additional elements were recorded:rst, in the case of the counterfactuals, whetheror not the manipulation actually undid the deathof the main character; second, the frequencywith which participants mentioned the con- junction of any of the targets (e.g., poison andcrash). Each participant provided at least oneanswer for the causal and the counterfactual lis-tings. Coding was performed by the author and

by an independent coder. Inter coder agreementwas (Raw agreement index. / =Overall. For causal answers , for counterfac-tual . Kappa coefficients are . , p


8/20

18 6



was not the preferred modal answer for counter-factual task, nor crash for the causal task.

ere is a signicant difference between thefrequencies of people who chose a counterfactualtarget [ Χ ( ,n= )= . , p


9/20

18 7


CAUSAL SELECTION

Table 2Percentage of Participants and Average Ratingsas a Function of the Task and Specicity Level(Experiment 1)

Judgement TypeCounterfactual Causal

Target Level % M % M

Crime life Low 54% 7.4 63% 6.7

Medium 50% 6.7 75% 6.8

High 13% 5.0 33% 8.3

Poison Low 58% 4.7 42% 6.7

Medium 70% 4.0 70% 5.9

High 83% 5.6 75% 4.8

Crash Low 63% 5.6 70% 7.0

Medium 75% 3.1 70% 7.4

High 83% 4.1 75% 7.4

Crash andPoison Low 17% 7.9 4% 5.0

Medium 20% 5.7 8% 5.5

High 33% 6.9 8% 7.7

the cause with independence of the level [ Χ ( ,n= )= . , p=. ]. e highest proportion of peo-ple in the counterfactual task corresponds to thehigh specicity level, where it was chosen by of the participants in that level. However, thedifference between the levels is not signicant[ Χ ( ,n= )= . , p=. ].

Overall, it can be seen that proportions ofpeople choosing a target are fairly similar acrossthe tasks, and that the higher number of peoplefor both tasks is concentrated around the poisonand crash targets. e distribution of the percen-tage of participants roughly mirrors the overalldistributions of answers. ere are no signicant

differences between the number of participantswho chose poison or crash as cause of the eventsin a test across levels (poison [ Χ ( , n= )= . , p=. ]; crash [ Χ ( , n= )=. , p=. ]). In thecounterfactual task the same pattern emerges(poison [ Χ ( , n= )= . , p=. ]; crash [ Χ ( ,n= )= . , p=. ]).

Although the percentage of participantsdoes not vary much within each level of spe-cicity for the counterfactual task, the causalratings do. A mixed was conductedon the importance ratings ( (judgement type)

(target) (specicity level)). e resultsshow signicant effect of judgement type [ ( ,

)= . , = . , p


10/20

18 8



in the low and medium levels, that is, the moreambiguous phrasings (signicant at p


11/20

18 9


CAUSAL SELECTION

Table 4 Mean Estimated Probability of Protagonist Deathand Probability Change (Experiment 1)

Mean ΔP

Crime life 30.43Poison 66.03 26.4

Crash but not poison 70.71

Crash and poison 83.60 12.17

Discussion

e dissociation between causal and coun-terfactual judgements predicted by was notreplicated in any of the levels of description.

e proportion of answers, participants and

importance ratings was similarly distributedacross types of judgements and varied acrossthe specicity level of description. It was alsofound that the counterfactual the task had anambiguous objective, which may possibly ex-plain the original results.

e modal responses varied as a functionof target, although not as precisely as it hasbeen expected. e main contrast between thisexperiment and Mandel’s was observed in theproportions and ratings corresponding to the

counterfactual task, specically regarding thetarget crime. is target was rated as counter-factually efficient in the low level and signi-cantly dropped in the high level. is standsin clear contrast with predictions. Insummary, cause selection and counterfactualanswers varied according to the level of des-cription, more clearly in the contrast betweenhigh and low-medium levels. Counterfactualanswers were particularly more sensitive to themanipulation.

More importantly, modal selection of targetdid not necessarily imply undoing of the targetevent. Counterfactuals generated in the low spe-cicity level did not necessarily undo the deathof the main character of the story. Mandel askedpeople about ways in which the “story could bechanged so that the outcome would have been

different”. It is possible to modify the story even ifthe main character had died anyway. e originalclassication by target (criminal life, explosion,and poison) takes for granted that people intro-duce changes to undo the death of the protago-nist. Finally, no evidence in favour of an ordereffect was found, therefore ruling out the possibi-lity of the dissociation due to an order effect.

is experiment showed how a simplemodication in the instructions for generatingcausal and counterfactual judgements can havea tremendous effect on the overall objective ofa counterfactual task, and on the focus of causaland counterfactual answers. Focus is led by the

task demand and not by a functional difference,which of course means, that dissociation can oc-cur depending on the context’s demands.

Experiment 2

In the previous experiment the dissociationbetween causal and counterfactual reasoning wasnot replicated. In fact, in most of the cases the-re was a clear match in focus between causal andcounterfactual ratings, depending on the speci-city of the description. However, there are at least

two factors that can account for this nding thatdo not have a parallel in the original study. erst one is clearly the coding system, since a con- junctive category was included, and it accountedfor an important proportion of participant res-ponses. e second one is the equivalence bet-ween the wording of the event description in thecausal and the counterfactual tasks.

e second experiment examines the im-pact of the match of the wording between thecausal and counterfactual tasks, keeping the co-

ding suggested by Mandel. It is predicted thatthe ndings of Experiment will be replicated,more specically, the medium level ndings,where there was some ambiguity in the gene-ral objective of the counterfactual task. Ordereffects and the ambiguity of the objective of thecounterfactual task were examined again.


12/20

190



Method

ParticipantsForty four undergraduate psychology stu-

dents ( female and male) were given cour-se credit to participate in the experiment. eirmean age was . ( = . ) and all of them suc-cessfully completed the task.

Materials and Proceduree materials are the same as described

for Experiment . e only difference is thatall participants work through causal and coun-terfactuals tasks that focus on the death of the

main character, that is, the medium level in Ex-periment . As in Experiment , the order of thetasks was randomly counterbalanced.

e coding was done according to the cate-gories of a priori interest, as in Mandel’s ( b)original experiment, in order to make the resultsmore directly comparable. As a measure of theambiguity of the task, it was recorded whetheror not the counterfactual manipulation actuallyundid the death of the main character. Answerswere again coded by the author and the same co-

der of the rst experiment.

Results

e results will be presented in an ordersimilar to Experiment . I rst present a generaloverview of the frequencies of responses. en,the proportion of participants and the meanratings per target are summarized and compa-red with the previous experiment, as well as theorder effects. is is followed by the ambiguitymeasure of the counterfactual task objective. Fi-

nally the probability ratings are examined.

Proportions of Answersere is no signicant difference between

the number of overall causal and counterfactualresponses [ χ ( , n= )= . , p=. ]. However,the number of answers differ depending on the

target [ Χ ( , n= )= . , p


13/20

19 1


CAUSAL SELECTION

e ratings show a curious pattern thatdoes not match the proportion of participants.Results were submitted to a repeated measures

. Target ratings changed depending onthe task [ ( , )= . , = . , p


14/20

192



participants who nonetheless recognized thatmodifying the crash is not enough to undo thedeath of the main character. It shows dissocia-tion, not as a function of the judgements, but ofthe evaluation considered in a causal networkand its relative weight in producing an effect.

e attributions to crime were again higher forthe causal than for the counterfactual task, inline with the probability ratings.

Probability ratings exhibit the same patternas in Experiment and in Mandel’s experiment.However, a very important difference consists inthe base rate probability of dying given that theprotagonist is a gangster. In both Experiments

and , the base rate probabilities are nearly dou-ble the base rate in Mandel’s experiments. isis coherent with the high proportion of partici-pants and ratings attributed in the causal task tocrime life.

In both Experiments and the proba-bility ratings do follow the predictions of .However, given the rest of the results, it makessense to affirm that this is not due to a functio-nal dissociation. One possibility is that the dis-sociation does not emerge due to problems with

the coding. Many of the cases classied as othermight have altered the answer and ratings distri-butions if switched to a different category. ishypothesis is tested in Experiment by usingclosed answers.

Experiment 3

Results of the second experiment showedthat the dissociation between causal and coun-terfactual reasoning is explained by the matchbetween the descriptions of the events. e

pattern of the modal responses exhibits a trendsimilar to the medium specicity level in Expe-riment . at is, the match between causal andcounterfactual reasoning, at least in the modalresponses, is not that clear.

It is possible that the methodology used tocollect the data has some effect on this result.Previous research has either used closed answers

or listing generation (Mandel & Lehman, ;N’gbala & Branscombe, ; Wells & Gavanski,

), but not both. Mandel ( b) has rightlyargued that the demand in each case is different,and that part of the success of is being ableto predict performance across differential de-mand. e differential demand consists of theavailability of the options for listings, since thegeneration of answers depends more directly onwhat the focus of attention is. However, what isnot taken into account is the possible interferen-ce of one task over the other. In all of Mandel’sexperiments, as well as in Experiments and ,participants were rst asked to list factors and

then to rate them. It is probable that the demandon focus to list the factors had altered the attri-buted importance. e use of predened catego-ries would make the relevant causal categoriesobvious, and consequently alleviate any contex-tual inuences.

Hence, the use of predened categories isimplemented in this experiment. It is predictedthat the match between causal and counterfac-tual judgements will be more clearly replicatedif the task demands are independent. Otherwise,

the pattern should change. In this experimentthe causal and counterfactual tasks consist onlyof the ratings of the content categories of a prioriinterest. No listings or probability ratings wererequested from the participants.

Method

ParticipantsForty one volunteer undergraduate stu-

dents ( female and male) from several pro-

grams at the University of Warwick were paid £to complete Experiment .

Materials and Proceduree method was generally the same as des-

cribed for Experiment . e causal and coun-terfactual tasks consisted exclusively in ratingthe factors provided so the wording was adjusted


15/20

19 3


CAUSAL SELECTION

slightly to reect this, while keeping the matchon focus between tasks unchanged. No codingwas necessary since the closed answer methodwas adopted. Probability ratings were not re-quested, making the overall duration of the ex-periment shorter.

Results

e mean counterfactual ratings for cri-me life, poison and crash were . , . , and .respectively, while the mean causal ratings were

. , . and respectively; thus, rating resultsof Experiment were closely replicated, the-reby not providing support to the idea of task

interference.Results were submitted to a repeated mea-sures . Ratings varied by target [ ( ,

)= . ., = . , p


16/20

194



and counterfactual tasks. e importance ofthis factor was lessened only when the taskswere phrased in the most specic level of des-cription (i.e., Experiment ), where crime lifewas given the lowest ratings and chosen by thesmallest proportion of people. In other words,only when the task referred to the specic wayin which the main character died, was the causaldependence on crime life not considered. is isalso reected in the mean base rate of death thatparticipants provided in the probability ratings.In comparison to the change in probability fromthis base rate compared to the action of any ofthe other elements, the probability was not as

large as in Mandel’s experiments.is calls into question the a priori dis-tinction between what constitutes a sufficientelement for an effect and what is a “preventor”.

proposes that counterfactual reasoningwill focus on prevention, and preventors seemto be elements that enable a causal relation.Even accepting this hypothesis, it is the statusof the preventor itself that is at stake when thedissociation occurs. e experiments presen-ted show that what is considered a preventor

(and a sufficient cause) does depend on theparticular description of the event, the instruc-tions and the type of task demand. at is, the-se notions cannot be dened independently ofthe context (Hart & Honoré, ), but only in very limited settings.

ere was, however, a consistent dissocia-tion across the experiments regarding the im-portance of the direct cause (crash) in causal andcounterfactual judgements. Whereas in the rstexperiment, crash was considered both the cause

and the best way to undo the outcome only inthe high level condition, in Experiments and

crash was not considered as effective in thecounterfactual as in the causal task. However,this tendency is actually complementary to theresults observed in terms of attributions to cri-me, and is consistent with the primacy constraint

hypothesis. e rst element of the chain wasconsidered both the most causal and the mosteffective in undoing the outcome, whereas thelast element of the chain is totally determined byits predecessors and cannot effectively undo theoutcome.

e variation is partially explained by thetask demands. e instructions were matched,and that factor alone reduced the extent of thedissociation (Experiment ). is result is rela-ted to the overlap between the questions aboutcauses, explanations and counterfactuals, aspointed out by Hilton et al. ( ; McClure etal., ). Participants could have interpreted

the questions about the factors causally respon-sible for the outcome as a more general questionabout an explanation of why things turned outto be that way. is is then the contrast betweenanswering how and why questions about a cau-sal situation. Although Mandel recognizes thisimportant distinction in his later work (Mandel,

, ), he does not relate it to the link bet-ween causal and counterfactual reasoning, andclaims that dissociation is functional, not the re-sult of task demands and pragmatics.

Furthermore, it is possible that people re-presented the scenario in different ways. In fact,there are at least two alternative representationsthat consequently imply slightly different coun-terfactual dependencies. ese are presented inFigure .

e rst row of Figure represents the scena-rio as a typical case of pre-emption, where coun-terfactuals dependencies do not allow singlingout a cause, and both poison and crash are inde-pendently caused. e second row incorporates

a new arrow to represent the possible inuenceon both the elements a er poison and its impacton the nal outcome. In this case, there is not anobvious candidate for causal selection. In thiscase, the immediate cause of the death (crash) isrepresented as an effect itself of poison. Take forexample the answer of one of the participants:


17/20

19 5


CAUSAL SELECTION

Figure 1 . Alternative representation of the scenario used in the experiments

“If Mr Wallace had not drunk the poison,his reexes would have been better, and he couldhave driven away from the van to his secondappointment”.

What these representations have in com-mon is that counterfactual modication of anythe nodes does not undo the nal outcome un-less the rst one is modied. It then appears that

participants considered this causal representa-tion and recognized that interventions in anynode were not sufficiently effective to undo thedeath of the protagonist. at is, in pre-emptionscenarios, people cannot hold constant otherelements in the causal network in order to assessthe counterfactual impact of the interventionson it, particularly when it is possible that thetwo cause candidates are not independent fromeach other.

is also seems to suggest that cases of pre-

emption are not suitable for testing theories ofcausal selection. It is very telling that cases ofpre-emption are customarily used in discussionsabout philosophy of causation given that sharedcommon intuitions about causality fail whenanalysing these cases.

It is still possible that the dissociation foundby Mandel is a side product of the differential

demand imposed by the causal and counterfac-tual tasks combined with the mismatch in thedescription in his causal and counterfactualtasks. Another possible explanation is that thegeneral description of the outcome triggered acausal model that made available counterfactualoptions that were not available when causal fac-tors involved in the death of the main charac-

ter were requested (Sloman, ). Even in thiscase, the dissociation seems to be due more toa task demand than necessarily to a functionaldifference.

References

Ahn, W., Kalish, C., Medin, D. L., & Gelman, S. A.( ). e Role of covariation versus mechanisminformation in causal attribution. Cognition, ( ),

- .Friendly re’ hits Kurdish convoy. ( , April).

News. Retrieved from http://news.bbc.co.uk/ /hi/middle_east/ .stm

Byrne, R. ( ). Mental models and counterfactualthoughts about what might have been. rends inCognitive Sciences, ( ), - .

Cheng, P. ( ). From covariation to causation: Acausal power theory. Psychological Review, ( ),

- .

Crime

Poison

Crash

Death

Crime

Poison

Crash

Death


18/20

196



Cheng, P. & Glymour, C. ( ). Causal mechanismand probability: A normative approach. In M.Oaksford & N. Chater (Eds.), Rational models ofcognition (pp. - ). Oxford ; New York: OxfordUniversity Press.

Collins, J. D., Hall, E. J., & Paul, L. A. (Eds.). ( ).Causation and counterfactuals . Cambridge, Mass.:

Press.Einhorn, H. J. & Hogarth, R. M. ( ). Judging pro-

bable cause. Psychological Bulletin PsychologicalBulletin, ( ), - .

Glymour, C. & Cooper, G. F. ( ).Computation,causation, and discovery . Menlo Park, Calif. Cam-bridge, Mass.; London: Press; Press.

Hart, H. L. A. & Honoré, T. ( ).Causation in the law.Oxford: Clarendon Press.Hilton, D. & Erb, H. P. ( ). Mental models and

causal explanation: Judgements of probablecause and explanatory relevance. Tinking andReasoning , , - .

Hilton, D., Jaspars, J. M. F., & Clarke, D. D. ( ). Prag-matic conditional reasoning: Context and contenteffects on the interpretation of causal assertions. Journal of Pragmatics, ( ), - .

Hoerl, C., McCormack, T., & Beck, S. R. ( ).Unders-

tanding counterfactuals, understanding causation:Issues in philosophy and psychology . Oxford; NewYork: Oxford University Press.

Kahneman, D. & Miller, D. T. ( ). Norm theory:Comparing reality to its alternatives. PsychologicalReview, ( ), - .

Kahneman, D., Slovic, P., & Tversky, A. ( ). esimulation heuristic. In D. Kahneman & A. Tvers-ky (Eds.), Judgment under uncertainty:Heuristicsand biases (pp. - ). Cambridge; New York:Cambridge University Press.

Kelley, H. ( ). Perceived causal structures. In J. M.F. Jaspars, F. D. Fincham, & M. Hewstone (Eds.), Attribution theory and research: Conceptual, deve-lopmental, and social dimensions. London; NewYork: Academic Press.

Lewis, D. ( ).Counterfactuals. Oxford: BasilBlackwell.

Lewis, D. ( ).Philosophical papers volume II . NewYork: Oxford University Press.

Mandel, D. ( a). Effect of counterfactual and factualthinking on causal judgements. Tinking & Reaso-ning , ( ), - .

Mandel, D. ( b). Judgment dissociation theory: Ananalysis of differences in causal, counterfactual,and covariational reasoning. Journal of experimen-tal psychology. ( ), - .

Mandel, D. ( ). Counterfactual and causal explana-tion: From early theoretical views to new frontiers.In D. R. Mandel, D. Hilton, & P. Catellani (Eds.),Te psychology of counterfactual thinking . London;New York: Routledge.

Mandel, D. ( ). Mental simulation and the nexusbetween causal and counterfactual explanation.In C. Hoerl, T. McCormack, & S. R. Beck (Eds.),Understanding counterfactuals, understanding cau-sation: Issues in philosophy and psychology . Oxford;New York: Oxford University Press.

Mandel, D. & Lehman, D. R. ( ). Counterfactualthinking and ascriptions of cause and preventabi-lity. Journal of personality and social psychology , ( ), - .

McClure, J., Hilton, D., & Sutton, R. M. ( ).

Judgments of voluntary and physical causes incausal chains: Probabilistic and social functionalistcriteria for attributions. European journal of social psychology , ( ), .

Menzies, P. ( ). e role of counterfactual de-pendence on causal judgements. In C. Hoerl, T.McCormack, & S. R. Beck (Eds.),Understandingcounterfactuals, understanding causation: Issuesin philosophy and psychology . Oxford; New York:Oxford University Press.

Miller, D. T. & Gunasegaram, S. ( ). Temporal order

and the perceived mutability of events: Implica-tions for blame assignment. Journal of personalityand social psychology , ( ), - .

N’gbala, A. & Branscombe, N. R. ( ). Mental simu-lation and causal attribution: When simulating anevent does not affect fault assignment. Journal ofExperimental Social Psychology , ( ), - .


19/20

19 7


CAUSAL SELECTION

Pearl, J. ( ). Causation, action and counterfactuals.Proceedings of the th conference on Teoreticalaspects of rationality and knowledge. e Nether-lands: Morgan Kaufmann Publishers Inc.

Petrocelli, J. V., Percy, E. J., Sherman, S. J., & Tormala,Z. L. ( ). Counterfactual potency. Journal of personality and social psychology , ( ), - .

Sloman, S. A. ( ).Causal models: How people thinkabout the world and its alternatives. Oxford: OxfordUniversity Press.

Sloman, S. A. & Lagnado, D. A. ( ). Do We “do”?Cognitive Science, ( ), - .

Spellman, B. A., Kincannon, A., & Stose, S. ( ). erelation between counterfactual and causal reaso-

ning. In D. R. Mandel, D. Hilton, & P. Catellani

(Eds.), Te psychology of counterfactual thinking (pp. - ). London; New York: Routledge.

Teixeira, S. & Pacheco, X. ( ).Borland Delphi .Indianapolis Ind.: Sams Publishing.

Verschueren, N., Schaeken, W., & Schroyens, W.( ). Necessity and sufficiency in abstract condi-tional reasoning. Te European Journal of CognitivePsychology , ( ), - .

Wells, G. & Gavanski, I. ( ). Mental simulation ofcausality. Journal of Personality and Social Psycho-logy , ( ), - .

Wells, G., Taylor, B. R., & Turtle, J. W. ( ). e un-doing of scenarios. Te Journal of Personality andSocial Psychology , ( ), - .

Woodward, J. ( ). Sensitive and insensitive causa-tion. Philosophical Review Philosophical Review, ( ), - .


20/20

C o p y r i g h t o f R e v i s t a C o l o m b i a n a d e P s i c o l o g í a i s t h e p r o C o l o m b i a , D e p a r t a m e n t o d e P s i c o l o g i a a n d i t s c o n t e n t m am u l t i p l e s i t e s o r p o s t e d t o a l i s t s e r v w i t h o u t t h e c o p y r i g h t H o w e v e r , u s e r s m a y p r i n t , d o w n l o a d , o r e m a i l a r t i c l e s f o r

causal selection and counterfactual reasoning - william jiménez-leal

Documents