Journal of Personality and Social Psychology 1996, VoL 7 t, No. 3, 450--463

Copyrighl 1996 by the American Psychological Association, Inc. 0022-3514/96/$3.00

Counterfactual Thinking and Ascriptions of Cause and Preventability D a v i d R . M a n d e l a n d D a r r i n R. L e h m a n University of British Columbia Research suggests that counterfactuals (i.e., thoughts of how things might have been different) play an important role in determining the perceived cause of a target outcome. Results from 3 scenario studies indicate that counterfactual content overlapped primarily with thoughts of how an outcome might have been prevented (preventability ascriptions) rather than with thoughts of how it might have been caused (causal ascriptions). Counterfactuals and preventability ascriptions focused mainly on controllable antecedents, whereas causal ascriptions focused mainly on antecedents that covaried with the target outcome over a focal set of instances. Contrary to current theorizing, causal ascriptions were unrelated to counterfactual content (Study 3). Results indicate that the primary criterion used to recruit causal ascriptions (covariation) differsfrom that used to recruit counterfactuals (controllability).

negation or "undoing" of the target effect), the stronger one's belief should be that C was a necessary cause of E. To use a wellknown example from Kahneman and Tversky (1982), if Mr. Jones was in a car accident, he might imagine having taken a different route (especially if the route he took was unusual; Kahneman & Miller, 1986). To the extent that imagining doing so also leads Jones to imagine not having been in the car accident, he may be especially likely to view his decision to take that unusual route as a cause of his accident. Mackie's (1974) notion that people use counterfactuals to infer necessary causes represents an extension of Mill's ( 1843/ 1973) method of difference to events that can be simulated in psychological space. Following Mill's method of difference, suflport for a necessary C is strengthened if C is the case, then E is also true. Logically, counterfactuat evidence that C implies E provides indirect support for the belief that E implies C (i.e., C was necessary for E). This form of causal reasoning amounts to a modus tollens argument (Copi, 1986; Wason, I966) in which the negation of an implicate (in this case, C) leads to the valid conclusion that the implicans (in this case, E) must also be negated. Logically, counterfactual tests of hypothesized sufficient causes are also possible (Mackie, 1974). Accordingly, one could construct a counterfactual in which E is negated. The easier it is to imagine that negating E would have undone C as well, the stronger one's belief should be that C was a sufficient cause of E. For example, Jones might think, "If only I hadn't gotten into this accident, I wouldn't have been in the path of a drunk driver." Once again, a modus totlens argument is constructed, but this time the hypothesized rule is that C implies E (i.e., C is a sufficient cause of E). Thus, following modus tollens logic, the

People often think about what might or could have been if only things had been slightly different than they were (see, e.g., Hofstadter, 1979, 1985; Kahneman & Miller, 1986; and Steiner, 1975 ). These thoughts, called counterfactuals, suggest possible worlds and ways in which they might have been realized. Counterfactual thinking thus involves imagining alternatives to one or more features of a perceived event. Over the past decade, several researchers (Einhorn & Hogarth, 1986; Hilton, 1990; Hilton & Slugoski, 1986; Kahneman & Tversky, 1982; Lipe, 1991 ; McGill, 1989, 1990; McGiil & Klein, 1993, 1995; Roese & Oison, 1995a; Wells & Gavanski, 1989) have suggested that people use counterfactual thinking in the process of ascribing causes to target outcomes (viz., causal ascription). These psychological accounts have much in common with each other, and each owes greatly to Mackie's ( 1974; as well as Hart & Honort's [ 1959 ]) philosophical account of causation. C o u n t e r f a c t u a l s as Tests o f Necessary a n d Sufficient C a u s a t i o n Mackie (1974) argued that people construct counterfactuals to test whether a particular antecedent was in fhct a cause of a target outcome. Specifically, in testing whether a hypothesized cause (C) is a necessary cause of target effect (E), one would construct a counterfactual in which C is negated (denoted C). The easier it is to imagine C followed by E (i.e., the

David R. Mandeland Darrin R. Lehman, Department of Psychology, University of British Columbia, Vancouver,British Columbia, Canada, This research was supported in part by doctoral fellowshipsfrom the Natural Sciences and Engineering Research Council of Canada (NSERC) and the Social Sciencesand Humanities Research Council of Canada (SSHRC) and by SSHRC Research Grant 410-93-1295 and NSERC Research Grant 95-3054. We thank Neal Roese and Jim Sherman for helpful comments on a draft of this article and Nancy Dwornick and Sarah Smith for their assistance with this research. Correspondence concerning this article should be addressed to David R. Mandel, who is now at the Department of Psychology, Stanford University, Stanford, California 94305-2130.

J An implicate is an event whose occurrence is implied by the occurrence of another event, called the implicans. The terms antecedent and consequent are often used synonymously for the terms implicans and implicate, respectively.However,because an implicans may not be temporally antecedent to an implicate, we use the unambiguous terms ( i.e., implicans and implicate) when referring to logical implication and the terms antecedent and consequent when referring to temporal order. 450

COUNTERFACTUALS, CAUSES, AND PREVENTABILITY negation of E implies the valid conclusion that C must also not have occurred. Although counterfactual tests of necessary and sufficient causes share the same valid modus tollens logic, they do not share the same degree of psychological plausibility. Most theorists and laypersons alike would probably agree that the type of counterfactuals required to test hypothesized sufficient causes are exceedingly rare. That is, people hardly ever imagine consequent events having implications for antecedent events, such as "'if only E had not occurred, then C would also not have occurred," The most likely reason for this asymmetry in counterfactual constructions is that people tend to think about events unfolding forward rather than backward in time. People, therefore, prefer to think of antecedents as having logical implications for consequents (Tversky & Kahneman, 1980), despite the fact that logical implication is not temporally bound (see Footnote 1 ). Whereas invoking a counterfactual criterion for tests of necessary cause accomplishes congruence between antecedent and implicans and between consequent and implicate, invoking a counterfactual criterion for tests of sufficient cause induces incongruence between temporal order and logical order. Perhaps this is why, in the literature on causal reasoning (e.g., Einhorn & Hogarth, 1986; Hilton, 1990; Hilton & Slugoski, 1986; McGill, 1989, 1990; McGill & Klein, 1993, t995), tests of necessity are framed as counterfactual conditional questions (i.e., if C, then E?), whereas tests of sufficiency are framed as contrastive but explicitly nonimplicative questions (i.e., what distinguishes E cases from E cases?). At best, then; everyday counterfactual (if only . . . ) thinking can aid in ascriptions about necessary causes but not in ascriptions about sufficient causes. This limitation poses a problem for theorists (e.g., Lipe, 1991 ) who argue that causal ascriptions depend heavily on counterfactual thinking because, in general, people think about causes in terms of necessity and sufficiency, and perhaps even more in terms of the latter (Hart & HonorS, 1959; Hilton, 1990; Hilton & Jaspars, 1987; Kanouse, 1972; Mandel, 1995; Nisbett & Ross, 1980). Another problem with the notion that counterfactual thinking represents a causal test strategy stems from the fact that counterfactuals often have a confirmatory rather than testlike quality. As noted earlier, psychological literature on counterfactual thinking (for reviews, see Roese & Olson, 1995a; and Sherman & McConnell, 1995 ) and causal reasoning (e.g., Einhorn & Hogarth, 1986; Hilton, 1990; Lipe, 1991; McGill, 1989, 1990; McGill & Klein, 1993, 1995) suggests that counterfactuals represent, among other things, a method for testing the plausibility of various hypothesized causes. From this perspective, counterfactual thinking is a negative hypothesis test strategy (see Klayman & Ha, 1987)--a method for answering the counterfactual conditional question, if C, then still E? in hindsight. That is, counterfactual thinking assesses P(EI C ) - - t h e conditional probability of the target effect still occurring, given the causal candidate did not occur. If counterfactual thinking really does test for the necessity of various causal candidates, one would expect considerable variability in the test results. That is, a sizable proportion of the counterfactuals that people construct should n o t undo the target outcome. On the contrary, counterfactuals usually are experienced as compelling possibil~

451

ities from the moment they become the focus of attention (Hofstadter, 1979). They have more of the quality of a confirmation than of a test. This is not to say that counterfactuals always successfully undo a target outcome. Indeed, people sometimes have counterfactual thoughts that change an antecedent but leave the outcome as it was perceived to have occurred (Roese & Olson, 1995a). In covariational terms, these counterfactuals represent c-cell instances (i.e., antecedent C is negated but effect E still occurs; ~C n E), which provide disconfirmatory information about the necessity of a causal relation. Much more common, however, are counterfactuals that undo a target outcome by changing an antecedent factor. These counterfactualsrepresent d-cell instances ( i.e., C is negated, and E is undone; C A E) that provide confirmatory information about the necessity of a causal relation when compared with the c cell. Therefore, even in theory, the notion that counterfactual thinking serves as a method for testing the plausibility of causal hypotheses faces some serious limitations. Next, we summarize the work of investigators who have examined the link between counterfactual and causal thinking. Empirical Research Only two studies ( N'gbala & Branscombe, 1995; Wells & Gavanski, 1989) have addressed the relation between counterfactual availability and causal ascriptions, and these two studies have provided contradictory results. Wells and Gavanski found support for the notion that counterfactuat thinking influences causal ascriptions. In their second study, for example, participants read a scenario in which a cab driver refused the fare of a paraplegic couple. The couple decided to drive their own specially equipped car, and they were subsequently killed when the bridge that they drove over collapsed. In the undoing condition, the cab driver reached the bridge 15 min before the couple and made it across safely. In the nonundoing condition, the cab driver also drove off the collapsed bridge. After reading the vignette, participants listed four ways that the couple's death could have been avoided and rated the extent to which the cab driver's refusal to take the couple caused their deaths. Wells and Gavanski found that the cab driver's refusal to accept the couple's fare was mentioned more often as a way of avoiding the couple's death and was viewed as more causal in the undoing condition than in the nonundoing condition. More important, they demonstrated that participants who undid the cab driver's decision rated the causal role of the cab driver's decision more strongly. N'gbala and Branscombe ( 1995 ), however, did not replicate these basic findings. In their second experiment, for example, participants read a vignette in which a cab driver did not take the fare of a handicapped couple because either he had made a mistake and picked up another couple on the same street (controllable) or because his car broke down (uncontrollable). The couple then took another cab that crashed off the bridge either because the second cab driver was drunk (controllable) or because the bridge collapsed (uncontrollable). Participants generated a single counterfactual statement (the outcome could have been different if o n l y . . . ) and then rated the extent to which the first driver, the second driver, the bridge, and the cou-

452

MANDEL AND LEHMAN

pie were responsible, blameworthy, and causally implicated. N'gbala and Branscombe combined these three ratings into a single measure of fault and found that, although significantly more participants mutated the first driver's actions than the second driver's actions or the collapsed bridge, fault was most strongly ascribed to either the second driver (when he was drunk) or the collapsed bridge. In contrast to Wells and Gavanski (1989), N'gbala and Branscombe did not find that participants who mutated a particular event rated that event as more at fault. N'gbala and Branscombe (1995) suggested that the first cab driver's actions represented the necessary cause, whereas the second driver's actions or the collapsed bridge represented the sufficient cause in the scenario. Accordingly, they argued that people focus primarily on necessary causes when mutating events but that they focus primarily on sufficient causes when ascribing fault. It is unclear, however, whether participants viewed the second driver's actions or the collapsed bridge as sufficient for the outcome. That is, neither the second driver's actions nor the collapsed bridge, by themselves, were sufficient to enable the outcome (e.g., iftbe first driver had taken the couple's fare, then the couple may have been alright, even though the second driver was drunk or the bridge collapsed). Moreover, if one considers what was conditionally necessary under the circumstances (see Mackie, 1974), the actions of the second driver (when drunk) or the collapsed bridge may be viewed as just as necessary as the first driver's actions. Scope o f the R e s e a r c h Similar to the studies by both Wells and Gavanski (1989) and N'gbala and Branscombe (1995), our notions about the relations between counterfactuals and preventability and causal ascriptions focus more specifically on upward counterfactuals that undo negative outcomes. The emphasis on cognitive reactions to negative outcomes is driven in part by findings which indicate that, in comparison with positive or neutral events, negative events elicit more counterfactual thinking (e.g., Boninger, Gleicher, & Strathman, 1994; Gavanski & Wells, 1989; Gieicher et al., 1990; J. T. Johnson, 1986; Kahneman & Miller, 1986; Landman, 1987; but see Markman, Gavanski, Sherman, & McMullen, 1993; and Roese & Olson, 1993, 1995b, for notable counterexamples) and more attributional thinking (see Peeters & Czapinski, 1990; Taylor, 1991; and Weiner, 1985, for reviews). Because negative events can significantly impede psychological well-being, there may also be important practical advantages to understanding how people process negative-outcome information. Because only upward counterfactuals (i.e., thoughts that improve on reality; Markman et al., 1993) can mentally restore negative outcomes to positive or neutral states, we focus exclusively on this type ofcounterfactual thinking. D i s t i n g u i s h i n g Between C a u s a l i t y a n d P r e v e n t a b i l i t y In contrast to both Wells and Gavanski (1989) and N'gbala and Branscombe ( 1995 ), we provide an account of the relation between counterfactual thinking and causal ascription that is based on the distinction between facilitative and inhibitory causes ( Kelley, 1971 a, 1973 ). Facilitative causes bring about or

enable target outcomes and are what people generally mean when they refer to causes ( Mandel, 1995 ). In contrast, inhibitory causes prevent target outcomes. In line with everyday language, we refer to facilitative causes simply as causes and inhibitory causes as preventors (keeping in mind that preventors are causes in the broader sense of the term). Just as everyday counterfactuals (i.e., counterfactuals in which the implicans is an event that is antecedent to a target outcome) have implications for hypotheses about necessary causes, they also have implications for hypotheses about sufficient preventors. Logically, the negation of any necessary cause may be reinterpreted as a sufficient preventor and vice versa. For example, consider the relation between oxygen and fire. Oxygen is a necessary cause of fire because its absence implies the absence of fire. This is logically identical to stating that the absence of oxygen is sufficient to prevent fire. In psychological terms, however, causal and preventability ascriptions may focus on different events. That is, although antecedent X may be treated as a good causal candidate, X (i.e., the negation of X ) may be viewed as a poor preventability cand~date. Similarly, although Y may be viewed as a good preventability candidate, Y may be viewed as an implausible causal candidate. This is not to say that preventability ascriptions cannot focus on the negation of perceived causes or that causal ascriptions cannot focus on foregone preventors. Indeed, the findings of Wells and Gavanski (1989) indicate that sometimes causal ascriptions overlap considerably with counterfactual preventability ascriptions. Nevertheless, we believe that the perception of preventors as negated causes, or the perception of causes as counterfaetual preventors, is the exception rather than the rule. Moreover, we anticipate that when people generate counterfactuals, they often try to simulate ways in which an outcome could have been prevented, without necessarily mutating the hypothesized cause of the outcome. Thus, in the process of mentally undoing focal outcomes, counterfactual thinking may often represent the post hoc cognitive sense that preventative actions might have been taken rather than the post hoc cognitive negating of hypothesized causes. This interpretation is consistent with Kahneman and Varey's (1990) notion of competitive causation but extends beyond their focus on close counterfactuals (i.e., thoughts about what almost happened) to conditional counterfactuals as well. In line with the classic social psychological works of Lewin (1936) and Heider (1958), Kahneman and Varey (1990) argued that people perceive target outcomes as the result of the competition between opposing forces. The notion that counterfactuals are aligned with post hoc preventability ascriptions suggests that people often think about competitive causation, even in cases where counterfactual outcomes did not just nearly occur. We suggest that even when a counterfactual outcome is not supported by a suddenly disconfirmed propensity (i.e., the perception of a dynamic shift in the probability of an outcome) but only by a disconfirmed disposition (i.e., the perception of the prior probability of an outcome), people still tend to think about preventative actions that they (or another focal person) might have taken that would have disabled, overtaken, or otherwise competed with the perceived causes of an unwanted outcome.

COUNTERFACTUALS, CAUSES, AND PREVENTABILITY Note that the distinction between negating necessary causes and adding possible preventors is unrelated conceptually to the distinction between additive and subtractive counterfactuals (Roese & Olson, 1993; cf. Dunning & Parpal, 1989, who also use the terms additive and subtractive to refer to frames of reference). That is, both perceived causes and perceived preventors may focus on either the addition or the subtraction of events. In counterfactual terms, an omitted event may be mentally added and a committed event may be mentally subtracted. We have no reason to believe that distinctions between causes and preventors depend on whether a target antecedent represents an omission or a commission. Indeed, as Davis, Lehman, Wortman, Silver, and Thompson ( 1995 ) noted, the same event may often be stated in either additive or subtractive terms (e.g., "If only I had stayed awake" vs. "If only I had not fallen asleep") with no apparent change in meaning. Controllability and Covariational Criteria We expect that people use different criteria to arrive at causal and preventability ascriptions. Causal ascriptions are likely guided by a covariational criterion, whereas preventability ascriptions, as well as counterfactuals, are likely guided by a controllability criterion. In the former case, the selection of causal candidates depends on the degree to which they covary with the target outcome. We do not assume, however, that covariational information need be explicit or data-driven. That is, world knowledge that provides theory-driven, covariational information can be used to arrive at causal ascriptions. For example, most people know that reckless driving is directly related to the likelihood of being in a car accident. People can then use this covariational knowledge in reasoning about the cause of a specific car accident involving a reckless driver. In contrast, we expect that the selection of preventability ascriptions depends on the factors that are perceived to be controllable from a focal person's point of view. Returning to the previous example, if Jones, while driving home by an unusual route, is hit by a reckless teenager, he then might be inclined to view the teenager's reckless driving as the cause of his accident, despite the fact that he might also think counterfactually that, if he had taken his normal route home, then the accident would have been prevented. Unlike the teenager's reckless driving, Jones's choice of route was under his own control. Jones's choice of route, however, is unlikely to covary with the occurrence of accidents. In contrast, reckless driving is likely perceived as covarying with accident frequency. In our view, Jones is likely to focus on the teenager's reckless driving (an uncontrollable covarying factor) in ascribing a cause to the accident, and he should be likely to focus on his choice of route (a controllable, noncovarying factor) or on other controllable factors when thinking about how the accident could have been prevented. Of course, if an antecedent is both controllable and covarying, then people may focus on the same event in both their causal and preventability ascriptions (e.g., if Joan does not pass a test because she hardly studied, even though she had more time to do so). In line with our view that counterfactuals are more representative of the taking of preventative actions than of the undoing of causal factors, we also expect that counterfactual content

453

(like preventability ascriptions) to focus on controllable factors. This view is supported by literature that indicates that perceived controllability is an important determinant of both counterfactual availability (Girotto, Legrenzi, & Rizzo, 1991; Markman, Gavanski, Sherman, & McMullen, 1995) and counterfactual directionality (i.e., whether the counterfactual improves reality or provides a worse scenario; Roese & Olson, 1995b). The findings of Roese and Olson (1995b) are particularly supportive because they show that upward counterfactuals (i.e., counterfactuals that improve reality) were more likely to focus on controllable antecedents than were downward counterfactuals (i.e., counterfactuals that simulate worse possible worlds). If, as we suggest, upward counterfactuals focus on how a negative outcome might have been prevented, and if preventability ascriptions follow a controllability criterion, then we would expect upward counterfactuals to be significantly more likely to focus on controllable actions than downward counterfactuals would, which may be implicated in nonattributional processes. One reason for the appeal of the counterfactual availability to causal ascription link is that, like counterfactuals, causal ascriptions usually focus on a single instance (Mackie, 1974). Counterfactual thinking (or mental simulation, more generally; Kahneman & Tversky, 1982 ) is a useful cognitive tool in reasoning about single-case situations for which covariation information is unavailable (Fearon, 1991; Hart & HonorS, 1959; cf. Lipe, 1991, who argued that covariation information is used as a proxy for counterfactual information). Note, however, that just because people often may focus on interpreting a particular outcome, they may still rely on world knowledge about a larger set of events of which the target outcome is an exemplar. That is, although the focus of a causal judgment may often be particularistic, the information used in arriving at a causal judgment may focus on a larger set of events that are semantically related to the target outcome. This view is consistent with an emerging knowledge structure account of causal reasoning (Abelson & Lalljee, 1988; Read, 1987, 1992; Read & Marcus-Newhall, 1993; Thagard, 1989) and with Cheng and Novick's (1990, 1991, 1992) covariational account of causal attribution and causal induction. Indeed, as Cheng and Novick ( 1991, 1992) have argued, for people to distinguish between causes, enabling conditions, and noncausal factors, it seems necessary that they consider covariation between hypothesized causes and target outcomes over different focal sets. A counterfactual criterion alone, however, is insufficient to explain these commonly made distinctions. Consider, for example, an inference made about the effect of oxygen on the occurrence of fire. Using only a counterfactual criterion, one would have to conclude that oxygen was a necessary cause of fire because whenever oxygen is absent, fire is also absent. In contrast, by applying a covariational criterion over different focal sets, the effect of oxygen on fire would likely be judged as an enabling condition: First, because oxygen is present in everyday situations (viz., one focal set), regardless of whether there is a fire or not, it would not be ascribed a causal role because of weak covariation. However, because oxygen does covary with fire over a broader focal set in which instances of oxygen being absent do occur, it would be viewed as an enabling condition, rather than a noncausal factor, within the former focal set of

454

MANDEL AND LEHMAN

everyday occurrences. In other words, the causal judgments people make about a focal set o f interest or even a particular case seem to depend on their knowledge of causal and covariational relations in broader focal sets that provide frames o f reference. The use o f knowledge about covariation over a focal set as a criterion for causal ascriptions suggests that people seek causes for events that have, all other things being equal, the greatest predictive value or explanatory breadth (see Read, 1992; Read & Marcus-Newhall, 1993; and Thagard, 1989). Study 1 Study 1 was designed to test three interrelated hypotheses: First, counterfactual content overlaps more with the content o f preventability ascriptions than with the content of causal ascriptions. Second, both counterfactuals and preventability ascriptions focus on controllable actions that might have been taken from a focal actor's perspective. In other words, counterfactual thinking and preventability ascription are guided primarily by a controllability criterion. Third, causal ascriptions focus on actions that general world knowledge indicates covary with a target outcome over a focal set of instances, regardless of the controllability of these actions. In other words, causal ascription is guided primarily by a covariational criterion.

Method Participants. One hundred and thirty eight ( 83 female and 42 male, 13 participants did not state their gender) University of British Columbia (UBC) undergraduates participated in the study for course credit. Materials and procedure. Participants first read the following modified version of Kahneman and Tversky's (t982) Jones's unusual route scenario: Mr. Jones is 47 years old, the father of three and a successful banking executive. His wife has been ill at home for several months. On the day of his accident, Mr. Jones left his office at his regular time. He occasionally left early to take care of home chores at his wife's request, but this was not necessary on that day. Mr. Jones did not drive home by his regular route. The day was exceptionally clear, so Mr. Jones decided to drive along the shore to enjoy the view. The accident occurred at a major intersection. The light turned yellow as Mr. Jones approached. Witnesses noted that he braked hard to stop at the crossing, although he could have easily gone through. His family recognized this as a common occurrence in Mr. Jones's driving. As he began to cross after the light changed, a truck charged into the intersection at high speed and rammed Mr. Jones' car from the left, Mr, Jones was seriously injured. It was later ascertained that the truck was driven by Mark Smith, a teenager who was under the influence of alcohol. Mark was on his way to a beach party that his friend had told him about earlier that day. After reading the scenario, participants answered one of six questions depending on the condition to which they were randomly assigned. Three of the six conditions focused on the thoughts of Jones (viz., Jones conditions), whereas the remaining three conditions focused on the thoughts of Smith (viz., Smith conditions). Crossed with the two levels of person focus were three levels of thought focus (viz., counterfactual vs. preventability vs. cause): In the counterfactual conditions, participants read, "'As commonly happens in such situations, (Mr. Jones)

[Mark Smith, the teenage driver], ot~en thought, "If only.. ?, (while he was recovering in the hospital ) during the days that followed the accident. How do you think (Mr. Jones) [Mark] continued this thought? Please write one or more likely completions." In the preventability conditions, participants read, "As commonly happens in such situations, ( Mr. Jones) [ Mark Smith, the teenage driver], often thought about how (his misfortune) [the accident] could have been prevented, (while he was recovering in the hospital) during the day,s following the accident. Please write what you think (Mr. Jones) [Mark] probably thought about when he thought about how the accident could have been prevented." In the cause conditions, participants read, "As commonly happens in such situations, (Mr. Jones) [ Mark Smith, the teenage driver], often thought about the cause of(his misfortune ) [ the accident ], (while he was recovering in the hospital) during the days that followed the accident. Please write what you think (Mr. Jones) [Mark] probably thought about when he thought about how the accident was caused?" After reading the question, participants received six blank lines to respond in an open-ended manner. Participants were then debriefed and thanked for their participation. Data coding. Two independent raters coded participants' responses in two ways. First, each participant's first statement was coded for whether it focused on (a) a controllable action of Jones (e.g., "He should have taken his regular route home," "He shoul6n't have stopped at the yellow light," or "'He should have left earlier that day"), (b) a controllable action of Smith (e.g., "He shouldn't have drank alcohol before driving," "He should have been more careful," or "He shouldn't have gone to the party"), or (c) something different (e.g., "If only there wasn't a beach party that day"). Second, the number of statements about either Jones's actions or Smith's actions was tabulated. Interrater reliability was high for both methods of coding. Agreement for the first method was 92%. Reliability was .93 for number of statements about Jones and .95 for number of statements about Smith. We resolved disagreements for both methods of coding by discussion.

Results and Discussion If, as we suggest, participants focused on controllable actions when thinking counterfactually and when generating preventability ascriptions, then participants in the Jones-counterfactual and preventability conditions should be especially likely to focus on controllable actions that Jones could have taken to avert the accident. Similarly, we anticipated that participants in the Smith-counterfactual and preventability conditions would be especially likely to focus on controllable actions that Smith could have taken to avert the accident. These predictions are in line with the focus rule ( K a h n e m a n & Tversky, 1982), which states that people focus on some property o f the main object o f concern and attention (viz., either Jones or Smith). However, our predictions diverge from the focus rule in the following manner: Because causes need not be controllable actions but are likely to covary over a focal set (e.g., drunk driving covaries with motor vehicle accidents), we expected that participants in the Jones-cause condition would be relatively more likely to focus on Smith's actions than would participants in the other two Jones conditions. Finally, we predicted that, like participants in the Smith-counterfactual and preventability conditions, participants in the Smith-cause condition would focus on Smith's drunk and reckless driving because, in general, d r u n k driving covaries with car accidents; hence we did not predict significant differences in response patterns between the three Smith conditions. To test these predictions, we analyzed participants' first state-

COUNTERFACTUALS, CAUSES, AND PREVENTABILITY

455

Table 1

Number of Participants" First Responses as a Function of Response Type and Condition in Study I Jones

Smith

Response type

CF

Prev

Cause

CF

Prev

Cause

Total

Jones's unusual route Jones's indecisive driving Jones's time of departure Smith's drunk/reckless driving Total

9 6 6 0 21

7 7 3 1 18

8 5 0 10 23

2 2 1 16 21

0 1 I 22 24

0 2 0 12 14

26 23 11 61 123

Note. CF = counterfactual; prev= preventability. Data from 17 participants are not shown because the content of their first statements represented small categories of responses accounting for no more than 1 or 2 participants.

ments using t w o ~(~2tests. As anticipated, participants in the Jones conditions were significantly more likely to focus on Jones's controllable actions (as opposed to Smith's controllable actions) in the counterfactual condition (100%) and in the preventability condition (94%) than in the cause condition (55%), x 2(2, N = 61 ) = 17.71, p = .0001.2 Also as predicted, participants in the Smith conditions were highly likely to focus on Smith's controllable actions across all three thought conditions: 87% for the counterfactual condition, 92% for the preventability condition, and 74% for the cause condition, x2(2, N = 67) = 2.95, p > .20. The specific types of first statements offered by participants in each condition are shown in Table 1. Most of the statements focusing on Jones's actions fell into three categories: statements concerning his choice of an unusual route home that day, statements concerning his indecisive driving habits, and statements concerning his time of departure from work. With few exceptions, the statements focusing on Smith's actions dealt with his d r u n k or reckless driving. In line with our hypothesis that participants would focus on covarying antecedents when ascribing causality to target outcomes, Smith's drunk-reckless driving was the most frequently mentioned response in both the Jonescause and the Smith-cause conditions. Because participants' first statements may not be reflective of their overall responses, we also analyzed the total n u m b e r of statements about Jones's and Smith's controllable actions. The mean n u m b e r of statements focusing on Jones, M = 1.16, SD = 1.17, and Smith, M = 1.28, SD = 1.13, respectively, did not differ significantly, t < 1. For each participant, we calculated the proportion of Smith to Jones statements by dividing the n u m ber of Smith statements + 1 by the n u m b e r of Jones statements + 1 ( 1 was added to the numerator and denominator to avoid undefined divisions). These proportions were then log transformed to normalize the distribution of values. The transformed proportions were analyzed using a 2 (person focus ) × 3 (thought focus) factorial analysis of variance (ANOVA). Not surprisingly, we obtained a significant main effect for person focus. As can be seen by comparing the row totals in Table 2, participants in the Jones conditions provided a significantly greater mean proportion of statements that focused on Jones's controllable actions than did participants in the Smith conditions, F ( 1,131 ) = 162.13,p < .001.

More important, and corroborating the previous × 2 analyses, a significant Person-Focus × Thought-Focus interaction was obtained, F ( 2 , 131 ) = 9.00, p < .00 I. As can be seen in Table 2, whereas each of the three mean proportions in the Smith conditions are substantially positive (indicating a greater proportion of statements focused on Smith's actions), only the Jones-counterfactual and -preventability conditions had high mean proportions of statements focused on Jones's actions; in the Jones-cause condition, the mean proportion was close to 0, indicating a roughly equal split between statements focusing on Jones and Smith, respectively. Pairwise tests confirmed this account: None of the pairwise tests between thought-focus conditions were significant for participants in the Smith conditions, smallest p > . 10. The mean proportion of Jones statements in the Jones-cause condition, however, was significantly lower than in either the Jones-counterfactual, t( 131 ) = 3.24, p < .005, or Jones-preventability, t( 131 ) = 3.71, p < .005, conditions (the mean proportions in the latter two conditions did not differ significantly, t < 1 ). The findings from Study 1 indicate that counterfactuals, like preventability ascriptions, tend to focus on controllable actions. In contrast, causal ascriptions are more likely to focus on antecedents that covary with a target outcome over a focal set of cases. In this study, covariation was implied because it is comm o n knowledge that driving while intoxicated increases one's chances of driving recklessly and causing an accident. Note that N'ghala and Branseombe's (1995) suggestion that counterfactuals tend to focus on necessary causes, whereas causal attributions tend to focus on sufficient causes, c a n n o t account for our findings. Jones's choice of an unusual route c a n n o t be objectively described as a necessary cause of his accident because he may have been in an accident, even if he took his usual route. Similarly, Smith's actions, by themselves, were insufficient to cause the accident. Moreover, if one considers what was necessary for the accident under the circumstances (see Mackie, 1974), then Jones's actions would have to be viewed as no more necessary for the outcome than would Smith's reckless driving because without the latter antecedent that particular accident would not have occurred. 2 The othercategory of responses, which represented only 7% of participants' statements, was excluded from analyses.

456

MANDEL AND LEHMAN

Table 2 Mean Log Proportion o f Smith to Jones Statements as a Function o f Person Focus and Thought Focus in Study I Thought focus Person focus

Counterfactual

Preventability

Cause

Overall

Jones Smith Overall

-.33 .36 .03

-.37 .33 .01

-.08 .22 .06

-.26 .30 .03

Also note that, a l t h o u g h the focus rule can a c c o u n t for the significant m a i n effect for person focus, the focus rule c a n n o t explain the significant Person-Focus × T h o u g h t - F o c u s interaction. T h a t is, the focus rule c a n n o t explain why p a r t i c i p a n t s in the Jones-cause condition focus a b o u t as m u c h on S m i t h as Jones. We suggest t h a t this finding is d u e to the differing criteria used to generate c o u n t e r f a c t u a l a n d preventability ascriptions, on the one h a n d (viz., a controllability c r i t e r i o n ) , a n d causal ascriptions, on the other h a n d (viz., a covariational c r i t e r i o n ) . Indeed, the near-equal split o b t a i n e d in the Jones-cause condition may reflect a c o m p e t i t i o n between the focus rule (leading to a n e m p h a s i s on Jones's a c t i o n s ) a n d a covariational criterion (leading in this case to a focus o n S m i t h ' s actions). Alternatively, it is also possible t h a t p a r t i c i p a n t s a s s u m e d t h a t J o n e s would have learned a b o u t S m i t h ' s behaviors (e.g., driving d r u n k ) b u t t h a t S m i t h would n o t have k n o w n a b o u t Jones's behaviors (e.g., taking a n u n u s u a l route a n d Jones's indecisive driving h a b i t s ) . T h i s potential i n f o r m a t i o n a s y m m e t r y may have resulted in Jones-cause p a r t i c i p a n t s focusing m o r e on S m i t h ' s b e h a v i o r t h a n S m i t h - c a u s e p a r t i c i p a n t s focusing on Jones's behavior. In the following two studies, all participants adopt the same focus, t h u s ruling o u t this i n t e r p r e t a t i o n o f the results, Study 2 In Study 2, we sought to replicate the basic findings o f Study 1, using a different vignette a n d a within-subjects design. O n c e again, we a n t i c i p a t e d t h a t p a r t i c i p a n t s ' c o u n t e r f a c t u a l s would c o r r e s p o n d m o r e closely in c o n t e n t to t h e i r preventability asc r i p t i o n s t h a n t h e i r causal ascriptions. Moreover, we expected that, as in Study 1, p a r t i c i p a n t s ' c o u n t e r f a c t u a l s a n d preventability ascriptions would focus on controllable actions, whereas t h e i r causal ascriptions would follow a covariational criterion. Unlike Study 1, however, this study i n c l u d e d a m a n i p u l a t i o n check o n the perceived controllability o f key events in the scenario.

Method Participants. Fifty-seven ( 34 female and 23 male) UBC undergraduates participated in the study for course credit, Materials" and procedure. Participants first read the following vignette: Mrs. Wallace was somewhat distressed about her husband flying from Vancouver to Calgary for a convention. She herself was afraid of flying, and neither her nor her husband had ever flown anywhere

before, in the past, they had either driven or taken the train to their destination. On this occasion, however, Mrs. Wallace was not accompanying her husband on this two day business trip. Mr. Wallace told his wife that he was going to book a flight because he didn't want to spend so many hours driving or taking the train. Mr. Wallace booked a flight even though he had originally considered driving. Mr. Wallace, although a little uneasy about flying for the first time, tried to assure his wife that everything would be alright. These assurances did little to put her at ease. She thought about pleading with him to take the train instead. But she didn't because she felt silly doing so, even though she knew that her husband would definitely have changed his plans at her request if she pleaded. One week later, Mr. Wallace took his flight to Calgary. About midway through the flight, the plane crashed, tragically killing Mr. Wallace and all others on board. In the weeks following the crash, a formal investigation found evidence from the debris and flight recorder that indicated that the plane's engine had spontaneously malfunctioned, leaving the pilot no time for an emergency landing. The investigation also determined that the plane's engine was inspected thoroughly by a qualified maintenance team prior to takeoff. Other plane engines of the same model were also inspected for structural flaws in manufacturing, but this inspection revealed that the engines were well constructed. Next, participants read the counterfactual question, "'As commonly happens in such situations, Mrs. Wallace often thought '!fonly ...', while she grieved the loss of her husband. How do you think she continued this thought? Imagining you are Mrs. Wallace, please write one or more likely completions." Participants were given four if-only stems to complete. Participants were then asked the preventability question: "'If you were Mrs. Wallace, and you were thinking about how Mr. Wallace's death could have been prevented, what would you think about?" Next, participants were asked the cause question: "If you were Mrs. Wallace, and you were thinking about the cause of Mr. Wallace's death, what would you think about?" For each of the latter two questions, participants received five blank lines to respond in an open-ended manner. Finally, participants were asked to rate "how controllable" (a) "was the engine malfunction?" (b) "'was Mrs. Wallace's decision not to plead with her husband?" and (c) '~was Mr. Wallace's decision toffy to Calgary?'" on a 7-point scale ranging from ( 1 ) not at all controllable, through ( 4 ) rnoderateh' c¢mtrollable, to ( 7 ) totally controllable. Participants were then debriefed and thanked for their participation. Data c~)ding. A rater who was blind to the purpose of the study coded responses to the counterfactual, preventability, and cause questions for the presence or absence of statements regarding (a) Mr. Wallace's actions (e.g., "He should have taken the train" or "He shouldn't have gone on the trip at all"), (b) Mrs. Wallace's actions (e.g., "'She should have pleaded with him not to go by plane" or "She should have gone with him"), and (c) the engine malfunction (e.g., "The plane engine was the problem'" or "'The maintenance team could have found the engine problem"). These three categories were of interest for the following reasons: Mr. Wallace was a focal target because each question was oriented toward his death. Mrs. Wallace was a focal target because participants were asked to respond from her point of view. We anticipated that participants would view the actions of Mr. Wallace and Mrs. Wallace as controllable from a focal person's perspective, even though these actions do not covary with plane crashes. In contrast, the engine malfunction was a factor that general world knowledge indicates would covary with plane crashes, even though it was uncontrollable from either of the two focal actors' perspectives.

COUNTERFACTUALS, CAUSES, AND PREVENTABILITY Table 3

Proportion of Participants as a Function of Response Type and Question Type in Study 2 Response category Question type

Mr. Wallace

Mrs.Wallace

Engine

Overalla

Counterfactual Preventability Causality Overallb

.37 .28 .11 .58

.93 .72 .26 .97

.11 .07 .32 .40

.97 .90 .58

a Proportion of participants mentioning one or more of the three response categories, b Proportion of participants mentioning a response category for one or more of the three questions.

Results and Discussion Manipulation check. We anticipated that participants would rate both Mr. Wailace's and Mrs. Wallace's actions as significantly more controllable than the engine malfunction. As expected, a repeated-measures ANOVA on the three controllability ratings was highly significant, F(2, 86) = 74.68, p < .001. This effect was clearly attributable to the lower mean rating of controllability for the engine malfunction, M = 2.02, SD = 1.49, than for either Mr. Wallace's actions, M = 5.27, SD = 1.56, or Mrs. Wallace's actions, M = 5.16, SD = 1.64. Analyses within question type. We hypothesized that people would be guided by a controllability criterion when generating both counterfactuals and preventability ascriptions and by a covariational criterion when generating causal ascriptions. Accordingly, we should find that participants focus more on both Mr. Wallace's actions and Mrs. Wailace's actions (each relatively high in perceived controllability) than on the engine malfunction (relatively low in perceived controllability) when answering both the counterfactual and preventability questions. Alternatively, we should find that participants focus more on the engine malfunction (relatively high in implicit covariation) than on either Mr. Wallace's actions or Mrs. Wallace's actions (both relatively low in implicit covariation ) when answering the causal question. Table 3 shows the proportion of participants' statements as a function of response type and question type. As predicted, when participants generated counterfactuals, the proportions of participants who mentioned Mr. Wallace's actions and Mrs. Wallace's actions, respectively, were greater than the proportion mentioning the engine malfunction (see the first row of Table 3). The difference between these three related proportions was significant, Cochran's Q(2, N = 57 ) -- 66.50, p < .0001, and all three pairwise comparisons were significant, largest p < .002. An analysis of participants' preventability statements revealed a similar pattern of findings. As expected, the proportions of participants who mentioned Mr. Wallace's actions and Mrs. Wallace's actions, respectively, were greater than the proportion mentioning the engine malfunction (see the second row of Table 3 ). The difference between these three proportions was significant, Cochran's Q(2, N = 57 ) = 42.76, p < .0001, and all three pairwise comparisons were significant, largest p < .005. Finally, as anticipated, the proportion of participants who

457

mentioned the engine malfunction was greater than the proportions mentioning either Mr. Wallace's actions or Mrs. Wallace's actions in response to the causal question (see the third row of Table 3).3 The difference between these three proportions was significant, Cochran's Q(2, N = 57) = 7.55, p < .03. The pairwise comparison between the proportion of participants mentioning Mrs. Wallace's actions and the proportion mentioning the engine malfunction, however, was nonsignificant,Cochran's Q < l; the remaining two pairwise comparisons were significant, largest p < .03. The preceding analyses examined differences in the proportion of participants mentioning each response category within each question type. The results of these analyses clearly support the notion that counterfactuals and preventability ascriptions focus more on controllable actions than relatively uncontrollable actions. The findings also provide some additional support for the notion that causal ascriptions focus more on antecedents that covary with a target outcome over a focal set of instances.4 Analyses within response type. In addition to examining whether counterfactuals and preventability ascriptions focus more on controllable events than uncontrollable events, we ask whether controllable events are more likely the focus of counterfactuals and preventability ascriptions than causal ascriptions, which we would also expect. The proportion of participants who mentioned Mr. Wallace's actions was, in fact, greater in response to both the counterfactual and the preventability questions than to the causal question (see the first column of Table 3). The difference between these three related proportions was significant, Cochran's Q(2, N = 57) = 10.94, p < .005. The smaller of the two predicted differences (viz., preventability vs. cause questions) was also significant, Cochran's Q( l, N = 57) = 5.00, p < .03; however, the proportion of participants mentioning Mr. Wallace's actions did not differ between responses to the counterfactual and preventability questions, Cochran's Q( l, N = 57) = 1.19, p > .25. An analysis of statements about Mrs. Wallace's actions revealed a similar pattern. As expected, the proportion of participants who mentioned Mrs. WaUace's actions was greater in response to both the counterfactual and the preventability questions than in response to the causal question (see the second column of Table 3). The difference between these three proportions was significant, Cochran's Q(2, N = 57) = 50.31, p < 3 Because when asked the cause question 42% of participants did not mention any of the three response categories that were of a priori interest, we examined the other responsesoffered. A considerablepercentage of participants mentioned two categories of responses: First, 25% mentioned that the pilot was causally implicated (only 9% of participants mentioned the pilot in response to both the counterfactual and preventability questions). The percentages across question type for this category are consistent with our predictions because the pilot's action, although not controllable from either of the two focal actors" perspectives, is consistent with world knowledgeabout events that covary with plane crashes. Second, 33% of the participants mentioned that the cause of the accident was fate (0% and 11%of the participants, respectively,mentioned this category in response to the counterfactual and preventability questions). 4 The data also indicate that across all question types participants focused more on Mrs. Wallace'sactions than Mr. Wallace'sactions. This finding, however,is of no theoretical interest.

458

MANDEL AND LEHMAN

.0001, and all three pairwise comparisons were significant, largest p < .002. Finally, as anticipated, the proportion of participants who mentioned the engine malfunction was greater in response to the causal question than to either the counterfactual or the preventability question (see the third column of Table 3). The difference between these three proportions was significant, Cochran's Q(2, N = 57 ) = 14.96, p < .001. The smaller of the two predicted differences (viz., counterfactual vs. cause questions) was also significant, Cochran's Q( l, N = 57) = 8.00, p < .005; however, the proportion of participants mentioning the engine malfunction did not differ between responses to the counterfactual and preventability questions, Cochran's Q < 1. The preceding set of analyses examined whether controllable events were more the focus of counterfactuals and preventability ascriptions than causal ascriptions. Without exception, the obtained findings supported this prediction. The within-response-type analyses also indicated that the nonsignificant difference in proportions between participants mentioning Mrs. Wallace's actions and participants mentioning the engine malfunction in response to the causal question was attributable to the overall high proportion of participants mentioning the former category and the overall low proportion mentioning the latter category. That is, whereas only 27% (26/.97) of participants mentioned Mrs. Wallace's actions in the cause condition, 80% (32/.40) of participants mentioned the engine malfunction in the cause condition. Taken together, the results of the within-question-type and within-response-type analyses suggest that both counterfactuals and preventability ascriptions focus on events high in perceived controllability from a target person's perspective. Moreover, as in Study l, participants were significantly more likely to focus on antecedents that general world knowledge indicates would covary with the target outcome when generating causal ascriptions than when generating counterfactuals or preventability ascriptions. A limitation of Study 2 is that all participants received the questions in the same order. However, because participants responded to the cause question after the counterfactual and preventability questions, we suggest that, if anything, this ordering would have increased the overlap between their causal ascriptions and their preventability ascriptions or their counterfactual statements. That is, participants' preventability ascriptions would be highly available when answering the cause question. Obtaining the predicted divergence in the content of participants' causal ascriptions from the content of their counterfactual and preventability statements under these conditions, therefore, seems particularly compelling.

One-way dependence of causal and preventability ascriptions on counterfactuals. Because Study 2 relies on a within-subjects design, we can also address more specifically the likelihood of particular causal and preventability ascriptions, given that participants offered the same response content to the counterfactual question. If our account is correct, then given a particular response to the counterfactual question (e.g., one focusing on Mr. Wallace's actions), the probability of that same response to the preventability question should be greater than the probability of that same response to the cause question. Table 4 shows these conditional probabilities and the weighted difference between them (last column). Positive difference scores (or

Table 4

Conditional Probability of a Particular Response Type on Preventability or Cause Ascriptions Given Same Response to Counterfactual Question in Study 2 Question type Response type

Preventability

Cause

2uTM

AP X P(R) b

Mr. Wallace Mrs. Wallace Engine M

.38 .75 .17 .43

.10 .26 .50 .29

.28 .49 -.33 .15

.10 .46 -.04 .17

a AP = difference between the two conditional probabilities, bP(R)= the base probability of offeringa givenresponse type on the counterfaetual question (see top row of Table 3). Therefore, AP × P(R) represents a weighted differencescore.

weighted difference scores) indicate that the likelihood of a particular preventability ascription, given the same counterfactual content focus, is greater than the likelihood of a particular causal ascription, given the same counterfactual focus. That is, positive difference scores indicate a greater one-way dependence of preventability-ascriptive content on counterfactual content than ofcausal-ascriptive content on counterfactual content. As expected, across the three response types, the mean weighted difference score was positive. This finding lends further support to the notion that counterfactual thinking has more to do with thinking about how an outcome might have been prevented than how it might have been caused.

Content overlap between counterfactuals and preventability ascriptions. Although the within-question-type analyses indi• cate that both counterfactuals and preventability ascriptions are more likely to focus on controllable antecedents (viz., Mr. Wallace's or Mrs. Wallace's actions) than causal ascriptions are, neither these nor the within-response-type analyses directly address the overlap between counterf~.ctuals and preventability ascriptions in terms of their focus on the same controllable antecedents. If counterfactuals and preventability ascriptions do overlap, then we would expect that participants who focus on Mr. Wallace's actions in response to either the counterfactual or the preventability question should be more likely to focus on Mr. Wallace's actions in response to the other question than to focus on Mrs. Wallace's actions in response to that other question. Similarly, participants who focus on Mrs. Wallace's actions on one of the two questions should be more likely to focus on Mrs. Wallace's actions in response to the other question than to focus on Mr. Wallace's actions. To test this hypothesis, we first calculated phi coefficients for the association between (a) Mr. Wallace responses on the counterfactual question and Mr. Wallace responses on the preventability question, (b) Mr. Wallace responses on the counterfactual question and Mrs. Wallace responses on the preventability question, (c) Mrs. Wallace responses on the counterfactual question and Mr. Wallace responses on the preventability question, and (d) Mrs. Wallace responses on the counterfactual question and Mrs. Wallace responses on the preventability question. For a-d, the phi coefficients were. 17,. 15, .02, and .29, respectively. Although, as expected, the difference between the

459

COUNTERFACTUALS, CAUSES, AND PREVENTABILITY mean phi coefficient for content-congruent responses (i.e., a and d, M = .23) was greater than the corresponding value for content-incongruent responses (i.e., b and c, M = .09), it was not a reliable effect (t test for nonindependent correlations <1).

Table 5 Mean Causal Strength as a Function of Mutability Condition and Causal Focus in Study 3 Plead Fly

Study 3 The findings of Studies 1 and 2 support the notion that counterfactuals offer a natural cognitive and verbal syntax for expressing hindsight ascriptions o f preventability for negative events. These studies also provide evidence that preventability ascriptions do not simply focus on the negation o f what is believed to be the cause of a target outcome. Like participants' responses to explicit questions about preventability, their responses to nonleading i f only prompts led them to focus more on controllable events than events to which they ascribed a primarily causal role. Thus, rather than representing tests in which hypothesized causes are negated, counterfactuals may more often represent the post hoc cognitive sense that controllable, preventative actions might have been taken. These findings raise questions about the generality of the only previous study (Wells & Gavanski, 1989) that empirically demonstrated an influence of counterfactual availability on causal thinking. The purpose of Study 3 was to investigate the relation between counterfactual availability and judgments of causal significance. Like Wells and Gavanski, we manipulated counterfactual availability by constructing default events that were more or less mutable, However, unlike Wells and Gavanski's vignettes, our vignette included both a controllable antecedent that did not implicitly covary with the target outcome and an uncontrollable antecedent that did implicitly covary with the target outcome, thus allowing participants more choice in terms o f what they ascribe as preventative and causal, respectively.

Method Participants. Ninety-nine (71 female and 28 male) UBC undergraduates participated in the study for course credit. Materials and procedure. We constructed four modified versions of the vignette used in Study 2 that manipulated the mutability of both Mr. Wallace's decision to fly and Mrs. Wallace's decision not to plead with her husband. The mutability of Mr, Wallace's decision to fly was manipulated in the following manner: In the low-mutability (Flylow) conditions, the last sentence of the first paragraph read, "Mr. Wallace told his wife that he was going to book a flight because he didn't want to spend so many hours driving or taking the train?' In the high-mutability (Flyhigh) conditions, an additional sentence was added: "Mr. Wallace booked a flight even though he had originally considered driving?' The mutability of Mrs. Wallace's decision not to plead with her husband to take an alternative form of transportation was manipulated in the following manner: In the low-mutability (Pleadlow) conditions, the last sentence of the second paragraph read, "But she didn't [ plead ], because she knew that once her husband made up his mind he definitely would not change his plans at her request?' In the high-mutability (Pleadhigh) conditions, the last sentence of the second paragraph read, "But she didn't [plead], because she felt silly doing so, even though she knew that her husband would definitely have changed his plans at her request if she pleaded?' The two mutability variables (Fly and Plead) were fully crossed. The vignettes were identical to the vignette used in Study 2 in all other respects.

Low

High

Overall

Engine malfunction Low High Overall

6.72 6.96 6.84

6.64 7.35 6.98

6.68 7.14 6.91

Mr. Wallace's decision to fly Low High Overall

2.52 2.23 2.37

2.52 2,26 2.40

2.52 2.24 2.38

Mrs. Wallace's decision not to plead Low High Overall

1.36 1.38 1.37

1.96 1.57 1.77

1.66 1.47 1.57

Note. Fly = mutability of Mr. Wallace's decision to fly to the convention; Plead = mutability of Mrs. Wallace's decision not to plead with her husband to take the train.

Participants were randomly assigned to one of the four mutability conditions and read the assigned vignette. Next, they were asked (as in Wells & Gavanski's, 1989, study) to "list four things that could have been different so that Mr. Wallace's death might have been avoided. "" Finally, participants responded, on 9-point scales ranging from (0) not at all the cause, through (4) moderately the cause, to ( 8 ) very much the cause, to each of the following three questions about the cause of Mr. Wallace's death: (a) "To what extent is the engine malfunction the cause of Mr. Wallace's death?" (b) "To what extent is Mrs. Wallace's failure to plead with her husband not to fly to Calgary the cause of his death?" and (c) "To what extent is Mr. Wallace's decision to fly to Calgary the cause of his own death?" Participants were then debriefed and thanked for their participation. As in Study 2, a rater who was blind to the purpose of the study coded participants' open-ended statements about avoidability for the presence or absence of statements regarding (a) Mr. Wallace's actions, (b) Mrs. Wallace's actions, and (c) the engine malfunction.

Results and Discussion Effects o f mutability Supporting the effectiveness o f our mutability manipulations, we found that a significantly higher percentage of participants in the Flyhigh conditions (100%) mentioned Mr. Wailace's actions in their counterfactual statements about avoidability than in the Flflow conditions (86%), ×2( 1, N = 99) = 7.38, p < ,007. A significantly higher percentage of participants in the Pleadhigh conditions (75%) mentioned Mrs. Wallace's actions in their counterfactual statements about avoidability than in the Pleadlow conditions (55%), ×2( 1, N = 99) = 4.37, p < . 0 4 . Table 5 shows participants' mean causal judgments as a function o f mutability condition. In contrast to the effect of the mutability manipulations on counterfactual statements about avoidability, and in contrast to the findings of Wells and Gavanski (1989), we found that neither mutability manipulation nor

460

MANDEL AND LEHMAN

the interaction of these dichotomous variables had a significant effect on any of the three causal ratings, smallest p > . 15. Controllable versus covarying antecedents. Replicating the findings of Study 2 and supporting our hypothesis that people focus on controllable antecedents in their counterfactual thoughts about avoidability, we found that, compared with the proportion of participants mentioning the engine malfunction (.21), a significantly greater proportion of participants mentioned Mrs. Wallace's actions (.65), Cochran's Q( 1, N = 99) = 34.89, p < .0001, and a significantly greater proportion of participants mentioned Mr. Wallace's actions (.93), Cochran's Q( 1, N = 99) = 67.21,p < .0001. If people rely on world knowledge about covariation in ascribing causality, then we should find that participants rate the engine malfunction as more causal than the actions of either Mr. Wallace or Mrs. Wallace. It is also plausible, however, that counterfactual availability has an effect on causal thinking. If this were so, we would expect that participants who mentioned a particular antecedent in their counterfactual statements about how Mr. Wallace's death might have been avoided would have rated that antecedent as more causal than participants who did not mention the same antecedent. To test these two hypotheses, we calculated a repeated measures analysis of variance on participants' three causal judgments. We entered each of the three dichotomous content-coding variables (viz., Mr. Wallace's actions, Mrs. Wallace's actions, and engine malfunction) first as covariates; we made type of causal ascription (Mr. Wallace's decision to fly vs. Mrs. Wallace's decision not to plead vs. engine malfunction) the repeated measure. Contrary to the notion that counterfactual availability significantly influences causal judgments, the influence of the three covariates on mean causal ratings was nonsignificant, F < I. In contrast, the effect of type of cause was highly significant, F ( 2 , 195 ) = 179.0 l, p < .001, and as can be seen in Table 5, this effect was due primarily to the higher mean causal rating of the engine malfunction than either Mr. Wallace's decision to fly or Mrs. Wallace's decision not to plead. The results of Study 3 add further support to the notion that counterfactual thinking often focuses on preventability ascriptions rather than causal ascriptions. Manipulating the mutability of default events had an anticipated effect on participants' counterfactual thoughts about avoidability (i.e., preventability) but not on their causal judgments. Moreover, consistent with N'gbala and Branscombe (1995), counterfactual availability had no significant relation to causal judgments. Finally, as in Studies 1 and 2, the results of Study 3 suggest that, whereas post hoc preventability ascriptions focus on controllable actions, post hoc causal judgments focus on antecedents that world knowledge indicates would covary with a target outcome over a focal set o f instances. General Discussion Over the past decade, many researchers (Einhorn & Hogarth, 1986; Hilton, 1990; Kahneman & Tversky, 1982; Lipe, 1991; McGilI, 1989, 1990; McGill & Klein, 1993; Roese & Olson, 1995a; Wells & Gavanski, 1989) have posited that causal ascriptions depend in a significant way on counterfactual thinking. From a logical perspective, it is indisputable that count-

erfactuals have causal implications (in the broad sense of the term cause), insofar as they deal with conditional relations between antecedent and consequent events (Roese & Olson, 1995a). From a psychological perspective, however, it is unclear whether counterfactual thinking plays a significant role in the process of ascribing (facilitative) causes to target events. In this article, we proposed that counterfactuals relate differentially to facilitative and inhibitory causal ascriptions. Specifically, we argued that counterfactuals correspond more closely in content to preventability (inhibitory causal) ascriptions than to (facilitative) causal ascriptions and that the stronger correspondence between the former two types of thoughts is due in part to their shared focus on controllable events (i.e., a controllability criterion ). We anticipated, however, that (facilitative) causal ascriptions would be guided primarily by a covariational criterion rather than by counterfactual or controllability criteria. The findings strongly support these hypotheses. Across studies, we found a significantly greater correspondence between counterfactuals and preventability ascriptions than between counterfactuals and causal ascriptions. More important, we also found that counterfactuals and preventability ascriptions focus primarily on controllable events and that these two types of thoughts focus on controllable events significantly more often than do causal ascriptions. Causal ascriptions, however, tend to focus more on antecedents that general world knowledge indicates would covary with the target outcome over a focal set of events. We obtained this overall pattern of findings using different scenarios, study designs, and data-analytic techniques.

Implications for Ca usal Ascriptions These findings, corroborating N'gbala and Branscombe (1995), suggest that people do not ascribe causes according to a counterfactual criterion. That is, they are not significantly more likely to consider a given antecedent as a facilitative cause simply because that antecedent is the focus of their counterfactual thoughts. Nevertheless, we interpret our results cautiously: We do not deny that counterfactual availability may, under some conditions, have an influence on causal ascriptions (as was demonstrated by Wells & Gavanski, 1989). Nor do we deny claims that causes must be mutable (see, e.g., Hart & HonorS, 1959; Kahneman, 1995; and Kahneman & Miller, 1986). What we have questioned, however, is the importance of counterfactual availability in augmenting support for a particular causal candidate when participants are given the opportunity to respond according to an alternative covariational criterion that may be more central to the process of causal ascription. We suggest that, whereas some degree of perceived mutability of a particular antecedent may be necessary for it to be deemed a cause, mutability--even extreme mutability--is generally insufficient to elicit a corresponding causal ascription. Thus, although causal ascriptions may be constrained by a mutability criterion, they are not likely constructed according to a counterfactual-availability criterion. In line with both knowledge structure and covariational accounts of causal reasoning, we suggested that causal ascriptions would focus on events that world knowledge indicates would covary with the target outcome over a focal set of cases. Using a covariational criterion for arriving at causal ascriptions would

COUNTERFACTUALS, CAUSES, AND PREVENTABILITY result in greater explanatory breadth than using either a counterfactual-availability or controllability criterion. This was evident in our research: In Studies 2 and 3, for example, the actions of Mr. Wallace and Mrs. Wallace provided little explanatory breadth, even though these actions were counterfactually available. That is, these actions could not explain why other people were killed in the same plane crash. In contrast, the less counterfactually available engine malfunction could account for these other deaths as well as Mr. WaUace's death, and even more generally, plane crashes resulting in death are likely to covary with engine malfunctions. In short, focusing on the engine malfunction as a facilitative cause yields the greatest explanatory breadth in that scenario. Also note that a su~ciency criterion applied to a single case (see N'ghala & Branscombe, 1995) cannot achieve explanatory breadth precisely because of its narrow focus.

Implicationsfor Counterfactual Thinking As noted earlier, recent literature on counterfactual thinking and causal reasoning suggests that counterfactuals represent, among other things, a method for testing the plausibility of various hypothesized causes. In contrast, we characterize counterfactuals not as test results indicating whY an outcome occurred but as highly available mental simulations that often represent how various outcomes could have been prevented (cf. Kahneman & Miller, 1986). As our findings indicate, counterfactuals are closely aligned with post hoc preventability ascriptions that focus on controllable events. In contrast to the notion that counterfactual thinking drives the causal ascription process, we suggest that counterfactual thinking may represent a cognitive syntax for expressing perceptions of cause-effect relations in a conditional schematic form. This view is supported by research indicating that people are geared toward processing counterfactual conditional information. For example, information from counterfactual conditional statements is encoded in memory as semantically complex propositions that retain the linguistic form of the counterfactual (Carpenter, 1973), and recall accuracy is better for counterfactual statements than causal statements (Fillenbaum, 1974). The preceding view is also consistent with some literature on reasoning (e,g., Cheng & Holyoak, 1985; Lehman, Lempert, & Nisbett, 1988; Nisbett, Fong, Lehman, & Cheng, 1987 ), which suggests that people rely on pragmatic inferential rules or reasoning schemas ( see also Kelley, 1971 b) that are isomorphic with either conditional or biconditional implication.

FunctionalImplications Our findings (and our interpretation of them) raise the question, Why should counterfactual preventability ascriptions focus on controllable antecedents and causal ascriptions focus on covarying antecedents? Although our data do not directly address this issue, we suggest that, at least in the case of negative events, there is functional value in differentiating the criteria used for constructing preventability and causal ascriptions, respectively. By focusing on a covariational criterion in causal judgments, people reduce future unpredictability. In contrast, by focusing on controllable antecedents that can nullify the

461

effect of a potential cause, people may garner knowledge that they personally can use to prevent unwanted outcomes of a similar nature from occurring in the future. Thus, whereas causal ascriptions may aid in predicting events, preventability ascriptions may aid in intervening in events. The processes of predicting and intervening functionally complement each other. Predicting negative events without intervening is functionally futile, and intervention simply cannot proceed effectively without predictive knowledge. This functional view is consistent with Taylor's (Taylor & Pham, in press; Taylor & Schneider, 1989) notion that counterfactual thinking (or mental simulation, more generally) provides a thought-to-action link, facilitating the execution of goaldirected, controllable plans (for empirical support, see Roese, 1994). Returning to Kahneman and Varey's (1990) notion of competitive causation, it appears that people seek an understanding of the facilitative causes of an event as well as a competitive plan for how they might have disabled or overcome that cause, so as to have prevented an undesirable outcome from occurring. From this perspective, both counterfactual and causal thinking may represent different functional responses to expectancy disconfirmations involving negative outcomes (see Olson, Roese, & Zanna, in press). As an emerging functional perspective on counterfactual thinking (see, e.g., Gleicher et al., 1990; M. K. Johnson & Sherman, 1990; Markman et al., 1993; Roese, 1994; Roese & Olson, 1993, 1995b; Taylor & Pham, in press; and Taylor & Schneider, 1989) suggests, these cognitive reactions may ultimately be motivated by an intention to avert similar outcomes in the future. Although our account is consistent with a functional perspective, we draw attention to some of the potential dysfunctional outcomes of counterfactual thinking, especially when it focuses on controllable antecedents (for a detailed discussion of dysfunctional aspects of counterfactuai thinking, see Sherman & McConnell, 1995 ). For example, people may focus on controllable antecedents that are of little functional value in controlling future events precisely because they are unrelated to the target outcome over a broader focal set of cases. Consider Jones's reaction that "if only he had taken a different route, the accident wouldn't have occurred." Or consider Mrs. Wallace's reaction that "if only she had pleaded with Mr. Wallace not to fly, he would still be alive." It is true in each particular case that, had those controllable actions happened, the negative outcomes that occurred might have been prevented. However, accidents caused by another reckless driver are no more likely to occur on unusual routes than normal routes, and accidents caused by mechanical malfunctions can happen in cars and trains as well as planes. In other words, these controllable actions are not likely to serve as a basis for functional behaviors in the future. A more adaptive approach may be to focus on the set of antecedents that are jointly controllable and covarying with the target outcome. Above and beyond not recruiting (and possibly even obscuring) useful information in controlling future outcomes, the counterfactual ascriptions of preventability offered in the previous examples also point to other unwarranted attributions and negative emotional reactions that might result from those thoughts. For example, thinking about how one personally could have prevented a serious negative outcome may result in

462

MANDEL AND LEHMAN

self-blame and feelings of personal responsibility where, in fact, one's actions might have had no predictive value in foresight

(Davis, Lehman, Silver, Wortman, & Ellard, 1996). Nevertheless, these feelings that one could have done something may trigger considerable negative affect. In terms of the distinction between self-blame and other-blame, one might expect that controllable preventability ascriptions would p r o m p t the former but not the latter. Alternatively, covariation-driven causal ascriptions might p r o m p t either the former or the latter, depending on whose behavior was most predictive of a target outcome. This research has focused on clarifying a distinction between how people perceive causes versus how they perceive preventors and on how these different types o f ascriptions relate to counterfactual thinking. It will be i m p o r t a n t in future research on counterfactual thinking and attribution to delineate further the differences between people's lay conceptions o f attributional constructs, such as causality, preventability, blame, responsibility, intenti0nality, and motive (e.g., see Shaver, 1985; Shaver & Drown, 1986). In particular, because most psychological accounts of these constructs have their roots in philosophy, it is important to contrast how logicians and laypersons define the same labels. The divergences between these two systems o f thought may provide the basis for constructing a detailed taxono m y o f lay attributional constructs. In turn, this should lend greater precision to psychological accounts o f attributional thinking and behavior.

References Abelson, R. E, & Lalljee, M. G. (1988). Knowledge structures and causal explanations. In D. J. Hilton (Ed.), Contemporary science and natural explanations: Commonsense conceptions of causality (pp. 175-203 ), New York: New York University Press, Boninger. D. S., Gleicher, E, & Strathman, A. (1994). Counterfactual thinking: From what might have been to what may be. Journal of Personality and Social Psychology, 67, 297-307. Carpenter, E A. (1973). Extracting information from counterfaetual clauses. Journal of Verbal Learning and Verbal Memory, 12, 512521. Cheng, P. W., & Holyoak, K. J. ( 1985 ). Pragmatic reasoning schemas. Cognitive Psl~hotogy, 17, 39 t-416. Cheng, E W,, & Novick, L. R, (1990). A probabilistic contrast model ofcau sal induction. Journal of Personality and Social Psychology, 58, 545-567. Cheng, E W., & Novick, L. R. ( 1991 ). Causes versus enabling conditions. Cognition, 40, 83-120. Cheng, E W., & Noviek, L. R. (1992). Covariation in natural causal induction. PsychologicalReview: 99, 365-382, Copi, I. M. (1986). Introduction to logic (7th ed.). New York: Macmillan. Davis, C. G., Lehman. D. R., Silver, R. C., Wortman, C. B., & Ellard, J. H. (1996). Self-blame following a traumatic life event: The role of perceived avoidability. Personality and Social Psychology Bulletin, 22, 557-567. Davis, C. G., Lehman, D. R., Wortman, C. B., Silver, R. C., & Thompson, S. C. ( 1995 ). The undoing of traumatic life events. Personality and Social Psychology Bulletin. 21, 109-124. Dunning, D., & Parpal, M. (1989). Mental addition versus subtraction in counterfactual reasoning: On assessing the impact of personal actions and life events. Journal of Personality and Social Psychology, 57. 5-15, Einhorn, H, J., & Hogarth, R. M. (1986). Judging probable cause. Psychological Bulletin, 99. 3-19.

Fearon, J. D. ( 1991 ). Counterfactuals and hypothesis testing in political science. WorldPolitics, 43. 169-195. Fillenbaum, S. (1974). Information amplified: Memory for counterfactual conditionals. Journal of Experimental Psychology, 102, 44-49. Gavanski, 1., & Wells, G. L. (1989). Counterfactual processing of normal and exceptional events. Journal of Experimental Social P~ychology, 25, 314-325, Girotto, V., Legrenzi, P., & Rizzo, A. ( 1991 ). Event controllability in counterfactual thinking. Acta Psychologica, 78, 111 - 133, Gleicher, E, Kost, K. A., Baker, S. M., Strathman, A. J., Ricbman, S. A., & Sherman, S. J. (1990). The role of counterfactual thinking in judgments of affect. Personality and Social Psychology Bulletin, 16, 284-295. Hart, H. L. A., & Honor6, A. M. (1959). Causation in the laa: Oxford, England: Clarendon. Heider, E (1958). The psychology o['interpersonalrelations. New York: Wiley. Hilton, D. J. (1990). Conversational processes and causal explanation. Psychological Bulletin, 107, 65-81. Hilton, D. J., & Jaspars, J. M. E (1987). The explanation of occurrences and nonoccurrences: A test of the inductive logic model of causal attribution, BritishJournalofSociaIPsycholog.v, 26, 189-201. Hilton, D. J., & Slugoski, B. R. (1986). Knowledge-based causal attribution: The abnormal conditions focus model. PsychologicalReview 93, 75-88. Hofstadter, D. R. (1979). GiMet, Escher, Bach:An eternalgolden braid. New York: Basic Books. Hofstadter, D. R. ( I985 ). Metamagical themas: Questingfor the essemz, q/mind and pattern. New York: Basic Books. Johnson, J. T. (1986). The knowledge of what might have been: Affecrive and attributional consequences of near outcomes. Personality and Social PsychologyBulletin, 12, 51-62. Johnson, M. K., & Sherman, S. J. (1990). Constructing and reconstructing the past and the future in the present. In E. T. Higgins & R. M. Sorrentino (Eds.), Handbook of motivation and cognRion: Foundations o['social behavior (Vol. 2, pp. 482-526), New York: Guilford Press, Kahneman, D. (1995). Varieties of counterfactual thinking. In N. J. Roese & J. M. Olson (Eds.), What might have been: The social psychology ~?["counte~wtual thinking (pp. 375-396). HiUsdate, NJ: Ertbaum. Kahneman, D., & Miller, D. T. (1986). Norm theory: Comparing reality to its alternatives. PsychologyReview 93, 136-153. Kahneman, D., & Tversky, A. (1982). The simulation heuristic. In D. Kahneman, P, Slovic, & A. Tversky (Eds.), Judgment under uncertainty: Heuristi~wand biases (pp. 201-208). New York: Cambridge University Press, Kahneman, D., & Varey, C. A. (1990). Propensities and counterfactuals: The loser that almost won. Journal of Personality and Social Psycholog.v,, 59, 1101-1110. Kanouse, D. E. (1972). Language, labelling, and attribution, in E. E. Jones, D. E. Kanouse, H. H. Kelley; R. E. Nisbett, S. Valins, & B. Weiner (Eds.), Attribution: Perceiving the causes of behavior (pp. 1 2 1 - 135 ). Morristown, NJ: General Learning Press. Kelley, H. H. (1971a). Attribution in social interaction. In E. E. Jones, D. E. Kanouse. H. H. Kelley, R. E Nisbett, S. Valins, & B. Weiner (Eds.), Attribution: Perceiving the causes of behavior (pp. 1-26). Morristown, N J: General Learning Press. Kelley, H. H. ( 1971 b). Causal schemata and the attribution process. In E. E. Jones, D, E. Kanouse, H. H. Kelley, R. E. Nisbett, S. Valins, & B. Weiner (Eds,), Attribution. Perceiving the causes of behavior (pp. 1 5 1 - 174 ). Morristown, N J: General Learning Press. Kelley, H. H. ( 1973 ), The process of causal attribution. American Psychologist. 28, 107-128.

COUNTERFACTUALS, CAUSES, AND PREVENTABILITY Klayman, J., & Ha, Y.-W. ( 1987 ). Confirmation, disconfirmation, and information in hypothesis testing. Ps~'hological Review, 94, 211228. Landman, J. ( 1987 ). Regret and elation following action and inaction: Affective responses to positive versus negative outcomes. Personality and Social Psychology Bulletin, 13, 524-536. Lehman, D. R., Lempert, R. O., & Nisbett, R. E. (1988). The effects of graduate training on reasoning: Formal discipline and thinking about everyday-life events. American Psychologist. 43, 431-442. Lewin, K. (1936). Principles of topological psychology ( E Heider & G. M. Heider, Trans.). New York: McGraw-Hill. Lipe, M. G. ( 1991 ). Counterfactual reasoning as a framework for attribution theories. PsychologicalBulletin. 109, 456-471. Mackie, J. L. (1974). The cement qf the universe:A study of causation. Oxford, England: Oxford University Press. Mandel, D. R. (1995). IA postexperimental survey of participants' conceptions of causes and preventors ]. Unpublished raw data. Markman, K. D., Gavanski, I., Sherman, S. J., & McMuUen, M. N. ( 1993 ). The mental simulation of better and worse possible worlds. Journal qIExperimental Social Psychology, 29, 87-109. Markman, K, D., Gavanski, 1., Sherman, S. J., & McMullen, M. N. ( 1995 ). The impact of perceived control on the imagination of better and Worse possible worlds. Personality and Social Ps.vchologyBzdletin, 21, 588-595. McGill, A. L. (1989). Context effects in judgments of causation. Jozzrnal of Persona6ty and Social Psychology, 57, 189-200. McGilI, A. L. (1990). Conjunctive explanations: The effect of comparison of the target episode to a contrasting background instance. Social Cognition. 8. 362-382. McGilI, A. L., & Klein, J. G. (1993). Contrastive and counterfactual thinking in causal judgment. Journal of Personality and Social Psychology 64, 897-905. McGill, A. L., & Klein, J. G. ( 1995): Counterfactual and contrastive reasoning in explanations for performance: Implications for gender bias. In N. J. Roese & J. M. Olson (Eds.), What might have been: The social ps)~'hology of cozmterfactual thinking (pp. 333-352 ). Hillsdale, N J: Erlbaum. Mill, J. S. (1973). A system of logic ratiocinative and inductive. In J. M. Robson (Ed.), Collected works of John Stuart Mill ( Vols. 7 and 8 ). Toronto, Ontario, Canada: University of Toronto Press. (Original work published 1843) N'gbala, A., & Branscombe, N. R. (1995). Mental simulation and causal attribution: When simulating an event does not affect fault assignment. Journal of Experimental Social Psychology, 31, 139162. Nisbett, R. E.~ Fong, G. T., Lehman, D. R., & Cheng, P. W. (1987, October). Teaching reasoning. Science, 238, 625-631. Nisbett, R. E., & Ross, L. (1980). Human in/~,rence: Strategies and shortcomings of social judgment. Englewood Cliffs, N J: PrenticeHall. Olson, J. M., Roese, N. J., & Zanna, M. P. (in press). Expectancies. In E. T. Higgins & A. W. Kruglanski (Eds.), Social psychology: Handbook ¢fbasicprinciples. New York: Guilford Press. Peeters, G., & Czapinski, J. (1990). Positive-negative asymmetry in evaluations: The distinction between affective and informational negativity effects. European Review of Social Psychology, 1, 33-60.

463

Read, S. J. ( 1987 ). Constructing causal scenarios: A knowledge structure approach to causal reasoning. Journal ~fPersonality and Social Psychology,, 52, 288-302. Read, S. J. (1992). Constructing accounts: The role of explanatory coherence. In M. L. McLaughlin, M. J. Cody, & S. J. Read (Eds.), Explaining ~mesel['to others: Reason-giving in a social context (pp. 220). Hillsdale, N J: Erlbaum. Read, S. J., & Marcus-Newhall, A. (1993). Explanatory coherence in social explanations: A parallel distributed processing account. Journal rf Personality and Social Ps~'hology, 65, 429-447. Roese, N. J. (1994). The functional basis of counterfactual thinking. Journal ~[Personalityand Social Psychology, 66, 805-818. Roese, N. J., & Olson, J. M. (1993). The structure of counterfactual thought. Personalityand Social PsychologyBulletin, 19, 312-319. Roese, N. J., & Olson, J. M. (1995a). Counterfactual thinking: A critical overview. In N. J. Roese & J. M. Olson (Eds.), What might have been: The social psychology o['cozmterfactual thinking (pp. 1-56). Hillsdale, N J: Erlbaum. Roese, N. J., & Olson, J . M . (1995b). Outcome controllability and counterfactual thinking. Persona#ty and Social Psychology Bzdletin,

21,620-628. Shaver, K. G. ( 1985 ). The attribzztion ~fblame: Causality, responsibilit3¢ and blameworthiness. New York: Springer-Verlag. Shaver, K. G., & Drown, D. (1986). On causality, responsibility, and self-blame: A theoretical note. Journal of Personality and Social Psychology, 50. 697-702. Sherman, S. J., & McConnell, A. R. (1995). Dysfunctional implications of counterfactual thinking: When alternatives to reality fail us. In N. J. Roese & J. M. Olson (Eds.), What might have been: The social ps)~'hology ~?["cozmter/actzzal thinking (pp. 199-232 ). Hillsdale, N J: Erlbaum. Steiner, G. ( 1975 ). After Babek Aspects ~f language and translation. New York: Oxford University Press. Taylor, S. E. (1991). Asymmetrical effects of positive and negative events: The mobilization-minimization hypothesis. Psychological Bulletin, 110, 67-85. Taylor, S. E., & Pham, L. B. (in press). Mental simulation, motivation, and action. In E M. Gollwitzer & J. A. Bargh (Eds.), Action science: Linking cognition and motivation to behavior. New York: Guilford Press. Taylor, S. E., & Schneider, S. K. (1989). Coping and the simulation of events. Social Cognition, 7, 174-194. Thagard, P. (1989). Explanatory coherence. Behavioral and Brain Sciences, 12, 435-467. Tversky, A., & Kahneman, D. (1980). Causal schemas in judgments under uncertainty. In M. Fishbein (Ed.), Progress in socialpsychoL ogy(Vol. 1, pp. 49-72). Hillsdale, NJ: Erlbaum. Wason, E C. (1966). Reasoning. In B. M. Foss (Ed.), No' horizons in ps~'hology (pp. 135-151 ). Harmondsworth, England: Penguin. Weiner, B. (1985). "'Spontaneous" causal thinking. PsychologicalBulletin, 97. 74-84. Wells, G. L., & Gavanski, I. (1989). Mental simulation of causality.

Journal ¢f Personality and Social Ps~'hology, 56.161-169. Received July 27, 1995 Revision received F e b r u a r y 23, 1996 Accepted F e b r u a r y 28, 1996 •