Intentions and Side Effects

Abstract

Judging that somebody does something intentionally is a key element in interpersonal relationships. If somebody does something good intentionally, they are more apt targets for praise and rewards. If somebody does something bad intentionally, they are more apt targets for blame and punishment. But what is involved in acting intentionally? Again, there is substantial philosophical disagreement about the right answer to that question. Important cases that have been used to test theories about intentional action involve side effects of actions. Research shows there is a tendency for good side effects to be judged as not intentionally brought about, yet bad side effects are judged to be intentionally brought about. This chapter documents that this judgment asymmetry is predicted by the global personality trait extraversion. Those who are extraverted tend to judge that bad side effects are intentionally brought about whereas introverts do not agree that bad side effects are intentional, even after controlling for a host of other factors including concepts of intuitional action, other personality traits, and cognitive skills. We also present evidence that we can predictably manipulate intentionality judgments of side effects based on what we know about judgment tendencies of extraverts.

“Suppose the vice-president of a company went to the chairman of the board and said, ‘We are thinking of starting a new program. It will help us increase profits for this year’s balance sheet, but in ten years it will start to harm the environment.’ The chairman answered, ‘I don’t care at all about harming the environment. I just want to make as much profit for this year’s balance sheet as I can. Let’s start the new program.’ They started the new program. Sure enough, the environment started to be harmed. Did the chairman intentionally harm the environment?”

Judgments of intentions and intentionality are fundamental elements of social cognition (Malle, Moses, & Baldwin, 2001). To see this, just imagine the different reactions you would have if somebody accidentally versus intentionally stepped on your foot in an elevator. You’d likely have a stronger, less forgiving attitude toward the intentional foot stepper. As you can see, judgments of intentionality are pervasive and important in everyday life. Intentionality judgments may also profoundly impact people’s lives. For example, in court cases, if somebody is thought to have intentionally brought about harm, greater amounts of punishment or remuneration are often warranted. Or, if you found out your mother intentionally killed somebody you would probably feel differently than you would if you learned she accidentally killed somebody. As such, judgments of intentionality can play important roles in how we relate, react, and interact with others.

Much like the free will debate, many theorists hold that there is, or should be, a tight connection between the philosophical and empirical investigation of intentions and intentionality. Some philosophers are explicit that they take folk intuitions seriously (Adams, 1986; McCann, 1998, 2005; Mele, 1992). As Mele writes, “a philosophical analysis of intentional action that is wholly unconstrained by that [folk] concept runs the risk of having nothing more than a philosophical fiction as its subject matter” (A. Mele, 2001, p. 27). Mele continues that “what it is to do something intentionally.....will be anchored in common-sense judgments about particular hypothetical or actual actions” (2001, p. 27).

Partially because of this tight connection between everyday intuitions and theorizing about intentional action, a sizable empirical literature has sprung up in the past few decades. This literature about intentional action has largely dealt with what we will call the Knobe effect or side-effect effect (Knobe, 2003a).¹ The Knobe effect is typified by people judging that bad side effects (e.g., harming the environment) are brought about intentionally whereas good side effects (e.g., hel** the environment) are not.

This chapter revolves around personality predicting the Knobe effect. However, some other theoretical side effects of these data are worth mentioning. We therefore start with a review of some of the major explanations of Knobe effect. We argue, similar to the arguments in Chap. 2, that no single account of the Knobe effect (or intentional action, for that matter) is likely to accurately capture all or most intuitions about intentional action. Instead, the Knobe effect appears to multiply determined by a number of factors including biases, concepts, personality, and judgment environments. As such, any single parsimonious account of the Knobe effect is not likely to succeed. The data about the Knobe effect again highlight the dangers of treating “the folk” as a monolithic entity. Intuitions are importantly diverse and predictable, and this diversity is philosophically and practically important.

The Intentional Action Side-Effect Effect

Defining “side effect” can be somewhat tricky. We will use the following definition of a side effect: “X is a side-effect action performed by an agent S if and only if S successfully seeks to perform an action A, E is an effect of his so doing, X is his bringing about E, and X has the following properties: S is not at the relevant time seeking to X either as an end or as a means to an end, and X is not in fact a means to an end that S is seeking at the relevant time” (Cushman & Mele, 2008, p. 179). Judgments of side effects have been thought to be important because they feature events that are foreseen but not intended (McCann, 2005; Nadelhoffer, 2006a). For example, one prominent view of intentional action—the Simple View—holds that side effects can never be brought about intentionally (Adams, 1986; McCann, 1998). The Simple View states that an action is intentionally brought about only if that action is intended. Side effects are by definition consequences of actions that are not intended, even if they are foreseen. For these reasons, on the Simple View, side effects of actions cannot be brought about intentionally. However, many people judge some side effects to be intentional (Knobe, 2003a, 2003b, 2004, 2006, 2010a, 2010b; Mele & Cushman, 2006; Nadelhoffer, 2006a, 2006b), which has been argued to count as a strike against the Simple View of intentional action (Nadelhoffer, 2006b).

Knobe (2003a) provides evidence that intentional action intuitions about side effects are responsive to the “goodness” or “badness” of the side effect. That is, when participants are presented with a good side effect, they tend not to judge that side effect is brought about intentionally. However, when people are presented with a bad side effect, they tend to judge that bad side effect as being brought about intentionally. This asymmetry is the Knobe effect or the side-effect effect. To illustrate, consider the following pair of cases:

Harmful/Helpful Chairman

The vice-president of a company went to the chairman of the board and said, “We are thinking of starting a new program. It will help us increase profits for this year’s balance sheet, but in ten years it will start to (harm/help) the environment.” The chairman answered, “I don’t care at all about (harming/hel**) the environment. I just want to make as much profit for this year’s balance sheet as I can. Let’s start the new program.” They started the new program. Sure enough, the environment started to be (harmed/helped). (Knobe, 2003a)

Theoretically, the main difference between Harmful and Helpful Chairman is the moral valence of the side effect. The helpful side effect is good and the harmful side effect is bad. When participants were asked to respond to the following statement “the chairman intentionally harmed/helped the environment,” their responses were qualitatively different (help is judged unintentional, harm is judged intentional). This effect has been replicated several times across cultures (Knobe & Burra, 2006), ages (Leslie, Knobe, & Cohen, 2006), and scenarios (Cushman & Mele, 2008; Knobe, 2003b, 2004, 2006; Mele & Cushman, 2006; Nichols & Ulatowski, 2007).

What could account for these differences between the helpful and harmful Chairman Cases Two prominent general types of explanations have been offered—core concept and non-concept explanations. Core concept explanations attempt to explain the difference as a function of the proper application of our concept of intentional action. That is, our concept of intentional action is somehow sensitive to the harmfulness or helpfulness of the action, and that helpfulness or harmfulness of the action is an appropriate element used in classifying whether an action is intentional. Non-concept explanations do not hold that our core concept of intentional action is appropriately sensitive to the moral valence of the action. Rather, the difference between the helpful and harmful Chairman Cases is the result of something other than the correct application of the concept such as making a mistake. As such, responses to either the helpful or harmful Chairman Cases do not necessarily reflect the correct application of a concept of intentional action and require additional mechanisms to account for the difference. We now turn to illustrative examples of each type of explanation.

Correctly Applied Concept Explanations

Knobe (2006) attempts to explain the asymmetry by different systems that are activated for the good and bad side effects. In the first process, one identifies the side effect as being either good or bad. For example, in this process, people identify the harmful chairman as bringing about the harmful side effect and identify the Helpful chairman as bringing about the helpful side effect. After this determination, one uses one’s concept of intentional action. If properties sufficient for intentionality judgments are found (e.g., foresight for bad side effects, desire for good side effects), then one judges that those good or bad side effects are intentionally brought about. Given this account, we can explain the asymmetry. Moreover, on this account the core concept of intentional action plays an important, and appropriate, role. It is part of the very concept of intentional action that it looks for moral features of the actions (and consequences of actions) that are to be judged intentionally. As such, this account gives central role to one’s concept of intentional action. As Knobe himself notes, “moral considerations are playing a helpful role in people’s underlying competence itself” (Knobe, 2006, p. 226).

Phelan and Sarkissian (2008) object to Knobe’s (2006) explanation. In one study, they provided participants slightly modified, but structurally similar, chairman scenarios (Knobe & Mendlow, 2004). These scenarios describe the president of a corporation who intends to increase sales in Massachusetts and foresees, but doesn’t care, that it will decrease sales in New Jersey. On a natural reading, decreasing sales in New Jersey is a side effect of increasing sales in Massachusetts. Most people judged that the president lowered sales in New Jersey intentionally. On Knobe’s original view, the judgment of intentionality should mean that the participants view the side effect as bad. But they don’t. Most people did not judge lowering sales tax in New Jersey as bad. The lack of judging the side effect as bad and the presence of the asymmetry puts pressure on Knobe’s (2006) account since the first system would not provide different verdicts for the two side effects. Hence, at least in some cases, the two-system account cannot completely explain the asymmetry typical of the Knobe effect.

Some mental states of the actor also influence ascriptions of intentionality to side effects. One mental state that reduces intentionality judgments of bad side effects is regretfully bringing about that side effect (Phelan & Sarkissian, 2008). Sverdlik (2004) set up an experiment where a person foresees that mowing the lawn will wake up his neighbors, but he does it anyway. In one condition, he is described as regretfully waking up his neighbors. In the other condition, any mention of regret is omitted. Participants were randomly assigned to one of the conditions. Those in the regret condition had significantly lower intentionality ratings for waking up the neighbors than those in the no-regret condition. Much like the results of the Phelan and Sarkissian study, these results put pressure on the two-process account since in both cases the side effect was bad. Hence, the badness of the side effect and activation of corresponding processes again do not completely explain intentionality judgments of side effects.

These types of results have forced a revision in Knobe’s (2006) view. Pettit and Knobe (2009) have offered the following revised concept-based account of the Knobe effect. In general, many people are sensitive to whether the person being evaluated acts with an appropriate or inappropriate attitude. Something similar may happen when people make intentionality judgments. In particular, we can think of the attitude that the agent has as a spectrum from having a favorable to unfavorable attitude toward a side effect. The agent’s attitude can fall anywhere on that spectrum. The moral judgments that one makes about the side effect and the agent’s attitude to that side effect set the “default” point on the spectrum to which intentionality judgments are compared and made. If the agent’s attitude falls on the favorable side of the default, the action is judged intentional. If the agent’s attitude falls on the unfavorable side of the default, the action is judged as unintentional. But the important thing is that the default position can change in part depending on the moral judgment one makes about what attitudes one should have toward an event. According to Pettit and Knobe (2009), this can explain the Knobe effect even though the chairman has exactly the same attitude (i.e., not caring) in both Helpful and Harmful Chairman. In Helpful Chairman, the chairman lacks a favorable attitude toward something that he normally should have a favorable attitude toward, so his attitude falls on the unfavorable side of the scale and is judged unintentional. In Harmful Chairman, the chairman lacks a negative attitude that he normally should have toward harming the environment, so his attitude falls on the favorable side of the default and is judged intentional. Similar reasoning can be applied to explain the Sverdlik and the Phelan and Sarkissian studies.

Nichols and Ulatowski (2007) have an alternative view of the side-effect effect. Their view centers on “interpretative diversity.” As you will see, this view significantly shaped our own views about the Knobe effect (and other effects reported in experimental philosophy). This account holds that people interpret the word “intentionally” differently. Some people have an interpretation that focuses on foresight while others have an interpretation that focuses on desire. Others oscillate between these two interpretations, thus accounting for the asymmetric pattern of judgments typical of the side-effect effect. To test their view, participants responded to both the Harmful Chairman and Helpful Chairman cases. Their results again confirmed the existence of a judgment asymmetry, and they, like previous studies, found that many people did not display the judgment asymmetry. However, and important for Nichols and Ulatowski’s view, in their studies roughly a third of all participants responded “no” to both Harmful and Helpful Chairman, a third responded “yes” to Harmful and Helpful Chairman, and a third responded “yes” to Harmful Chairman and “no” to Helpful Chairman. Participants were allowed to explain their answers. Two major explanations emerged. One explanation is that the chairman knew that the side effect was going to come about. These people tended to think that the chairman brought about the side effects intentionally. The other explanation involved the desire of the chairman. Since the chairman did not desire to bring about the side effect, participants using this explanation tended to think that the side effect was not brought about intentionally. Hence, Nichols and Ulatowski concluded that “considerations of outcome may influence which interpretation the term is given” (2007, p. 361). In terms of these explanations, it seemed plausible to think that people employ a knowledge-based and a motive-based concept of intentional action.

Cushman and Mele (2008; Mele & Cushman, 2006) provide additional evidence and extend Nichols and Ulatowski’s work. In their studies, in addition to the two Chairman Cases, participants were given a variety of scenarios focusing on two major differences. In one set of scenarios, the person is described to encourage the interpretation that the person believes that side effect will come about but does not desire that side effect. In the other set of scenarios, the person is described to encourage an interpretation that the person desires the side effect come about but does not believe that it will. Like Nichols and Ulatowski, Cushman and Mele found three general patterns of responses—some answer “yes” to both Harmful and Helpful Chairman, some answer “no” to both, and some answer “yes” to Harmful Chairman but “no” to Helpful Chairman. Because their vignettes systematically varied belief and desires, they discovered that nearly all people think that if a person desires a side effect to come about (and the side effect does come about), then they did it intentionally. However, there was diversity concerning whether belief was enough to judge that a side effect is brought about intentionally, with some people thinking that belief is enough whereas others thought that belief was not enough to judge that side effect as being brought about intentionally. For these reasons, Cushman and Mele think that there are at least two concepts of intentional action—one where belief is enough to bring about a side effect intentionally and one where desire is required to bring about a side effect intentionally. Cushman and Mele also allow for the possibility of a third concept where normally desire is required for an action being judged intentionally except for actions that are morally bad. In those morally bad instances, knowledge may be enough to judge the action intentional. This third concept can thereby explain the judgment asymmetry typical of the side-effect effect.

Sripada (2010) attempts to explain the Knobe effect with what he calls the Deep Self Model. The Deep Self is the stable set of psychological characteristics of a person. Contrast the Deep Self with the Acting Self. The Acting Self is the self that is the proximal cause for an action. Often the Deep and Acting Self correspond to one another, for example, when a vegetarian does not eat meat on a particular occasion. But sometimes the Deep and Acting self are discordant—for example, when the momentary desire to eat an expertly prepared piece of meat just overwhelms the vegetarian and she eats the meat. According to Sripada, people’s concept of intentional action involves the Concordance Criterion: when the event that one’s action brings about lines up with one’s deep self, then we are likely to judge the outcome intentional. As such, Sripada argues that normative considerations (goodness or badness of the side effect) are irrelevant to the Knobe asymmetry. The chairman in Harmful Chairman is more likely to be thought to have a deep and enduring disdain for the environment than the chairman in Helpful Chairman. Sripada presents evidence that the “Deep Self” accounts for the asymmetry because after controlling for Deep Self features, the normative features (goodness/badness, blame/praise) are not related to the asymmetry. Hence, the asymmetry is explained without reference to normative factors at all but rather by one’s concept of intention action.

But the Deep Self model is not completely satisfactory (see also Rose, Livengood, Sytsma, and Machery (2012)). Jason Shepard (2011) has argued that while the Deep Self model can explain some of the variance associated with the asymmetry, it cannot explain it all. In fact, non-normative factors may still play a role in people’s concept of intentional action. Shepard offers a number of examples where deep self-attitudes did not have a significant impact on intentionality judgments in accordance with the Deep Self model. To take one example, Shepard gave participants a scenario indicating that the chairman had a deep and abiding commitment to the environment. In such a case, the Deep Self model should predict that the chairman intentionally brought about the help to the environment because he has a Deep Self in concordance with the event that is brought about by his intended action. However, Shepard’s experimental results did not bear out this prediction. In fact, the results in the caring chairman case were not different from Helpful Chairman, calling into question the Deep Self model.

Not Correctly Applied Concept-Based Explanations

Unlike concept-based explanations that take the Knobe effect to reflect a central feature of competent application of the concept of intentional action, non-concept explanations try to explain the Knobe effect without reference to our concept of intentional action. Non-concept explanations can proceed in a number of ways. For example, as we will see, non-concept explanations can hold that people are somehow illegitimately biased when making intentional action ascriptions. Or perhaps people really mean something else when they use the word “intentional” in the contexts that generate the Knobe effect. If they mean something else with the word “intentional,” then again, they do not apply their core concept of intentional action in the Chairman Cases. Rather, people use some other concept and use the label “intentional” as a way to talk about that concept given the restrictions in the experiment (e.g., questions asking about intentionality). In these ways, people are not correctly applying their concept of intentional action.

Several non-concept accounts have been offered to explain the side-effect effect of which we will review just a few. One is offered by Edouard Machery (2008). On Machery’s account, the asymmetry can be explained by the notion of trade-offs. In the Harmful chairman case, the chairman can be seen as trading off a bad thing (i.e., harming the environment) for a good thing (i.e., increasing the bottom line). Because there is a bad thing that occurs with a good thing, many people may view that kind of exchange as being intentionally brought about. However, in the Helpful chairman case, there is no bad thing being traded for a good thing. Rather, the side effect is helpful and need not be traded for. So, in that case, many people may be inclined to think that the helpful side effect is not intentionally brought about. Hence, the trade-off hypothesis can explain the side-effect effect with having to reference to core concept of intentional action (however, see Machery (2008) for potential problems with this account).

On an alternative account, Adams and Steadman (2004a, 2004b) focus on potential conversational implicature that may take place in the Chairman Cases (Grice, 1975). Conversational implicature can occur when, given the appropriate context, one uses a word or a phrase to express something other than what is literally meant by that word or phrase. Take, for example, a basketball game. A player may take a horrible shot and a fan may say “nice shot” to express displeasure with the shot. The fan uses the expression “nice shot” to mean something else (in this case, conversationally implying “bad shot”). Adams and Steadman argue that something similar may happen in the Chairman Cases. Participants may judge that the Harmful Chairman is blameworthy for bringing about the bad side effect. Given the response options in the experiment, participants can only express this judgment of blame by indicating that the chairman brought the bad side effect about intentionally. If participants did not indicate the harm was brought about intentionally, then that would conversationally imply that the chairman is not blameworthy. For the Helpful chairman, participants do not want to conversationally imply that the chairman is praiseworthy for not caring about hel** the environment. Consequently, they judge hel** the environment is not intentionally brought about. In these ways, conversational implicature, and not a core concept, can explain the side-effect effect (see Malle (2006) and Knobe (2003b) for responses).

Nadelhoffer (2004) thinks that the side-effect effect is best explained by biasing. In this case, the bias is the result of differential affective reactions that people tend to have toward the Helpful and Harmful Chairman. In the Harmful Chairman case, there is negative affect that is generated by the chairman not caring about harming the environment. Participants thereby perceive the Harmful Chairman negatively and this negative impression (caused by the bias) results in participants tending to judge that the harm was brought about intentionally. Along a similar line, people are likely to have a negative affective reaction toward the Helpful Chairman because the Helpful Chairman does not care about something that he should (i.e., hel** the environment). Because of the bias, people do not want to praise the Helpful chairman for bringing about the help but they do want to blame the chairman for bringing about the bad side effect. When Nadelhoffer measured praise and blame ratings for the Helpful and Harmful chairman (respectively), praise ratings were much lower than blame ratings. Hence, these data are consistent with an affective biasing account that could explain the side-effect effect (see Nadelhoffer (2004, 2006a) for direct evidence for the affective biasing account).

Similar to Nadelhoffer (2004), Malle and Nelson (2003) propose an account where negative affect generated in the Harmful chairman case plays a role in judging that side effect intentional. They key on the tendency for people, when they are fighting, to judge the person who they are fighting with harshly. These judgments, including intentionality judgments, are biased by the negative affect that is generated during the fight. For example, if Jean and Robin are fighting and Jean bumps into a plate and breaks it, there is a greater tendency for Robin to think that Jean broke the plate intentionally.

Something similar may be happening in the Chairman Cases. Malle (2006) argues that as the Chairman Cases are described, the Harmful Chairman does not care at all about the bad side effect he brings about. Given that bit of information, participants may think that they are supposed to use that information in the judgment about the chairman. After all, it is natural for participants to think that critical information included in a scenario is there for a reason. If participants think that they are supposed to use that information, they may use that evaluative information rather than their core concept of intentional action when making judgments about the Chairman. That would mean that people have the tendency to judge the Harmful Chairman as bringing about the side effect intentionally. However, in the Helpful Chairman case, participants may be less likely to think that they are supposed to use the information about hel** the environment, and thereby judge that the Helpful Chairman did not bring about the side effect intentionally.

Personality Predicts Philosophical Disagreement in Intentional Action

At this point, we have documented two different general types of explanation for the side-effect effect (and admittedly not an exhaustive recounting of all possible explanations). We want to emphasize that these studies (and others) had substantial dissenting minorities. These dissenting minorities call for an explanation of why they respond differently from the majority responses. To begin to help offer a potential explanation, we will detail some of the experimental work concerning the relation of personality traits to the Knobe effect in section “The Knobe Effect and Extraversion.” In sections “The Knobe Effect, Extraversion, and Theoretical Accounts” and “Conclusion” we will then discuss why the empirical data could be important for theoretical accounts of the Knobe effect discussed in the previous sections.

The Knobe Effect and Extraversion

As already detailed in previous chapters, many personality traits are related to differences in cognition. These differences include one’s motivation, judgment tendencies, detection of cues in the environment, the perceived importance of those cues, among others (Funder, 1991, 1995; McCrae & Costa, 1990). Our focus here is again on the global personality trait extraversion. Extraverts are socially minded individuals with relatively less emotional regulation who are motivated to engage in social activities and cognition. It stands to reason, then, that when a person engages in a harmful, socially undesirable action, extraverts would be more likely than introverts to be sensitive to those undesirable events and more motivated to express their feelings. As such, the Knobe effect is likely to be positively related to extraversion since harming the environment is socially undesirable whereas hel** the environment is not.

The first study we review focuses on the classic Knobe effect Chairman Cases (A. Feltz & Cokely, 2009). A similar procedure used in the free will studies was used to establish the relation of extraversion with the Knobe effect. Participants were given the classic chairman scenarios and a general measure of global personality traits including extraversion (the Ten Item Personality Inventory). It was predicted that extraversion would be related to the side-effect effect. However, extraversion is likely not the only factor that contributes to the side-effect effect. Other individual differences are likely related as well, such as expertise or cognitive impulsivity. Because these other factors are known to influence judgments in other domains they may also contribute to the judgments typical of the side-effect effect (Cokely, Galesic, Schulz, Ghazal, & Garcia-Retamero, 2012; Ericsson, Prietula, & Cokely, 2007; Frederick, 2005; Stanovich, 1999). For example, those who have a reflective cognitive style are more likely to wait some time to receive a larger reward and cognitively reflective individuals may be less likely to be influenced by framing effects in gambles than less cognitively reflective individuals (Frederick, 2005). Individual differences in attentional control are related to differences in strategies such as double checking and are associated with longer deliberation about problems or tasks (Cokely & Kelley, 2009; E. T. Cokely, Kelley, & Gilchrist, 2006). Any of these factors could influence some people to think more carefully about the cases, and that could cause different patterns of judgments about side effects (e.g., being less biased).

Accordingly, cognitive impulsivity was measured and statistically controlled for in the first study we will review. Cognitive impulsivity refers to some individuals’ tendency to rely on intuitive or gut reactions versus those who tend to rely on more deliberate and effortful processing. We used the Cognitive Reflection Task to measure cognitive impulsivity (Frederick, 2005). We also controlled for other potential individual differences by measuring self-control (Tangney, Baumeister, & Boone, 2004); working memory capacity (as measured by the working memory operation-span measure (OSPAN)), (Turner & Engle, 1989); and self-reported scholastic aptitude test (SAT) scores which are known to correlate with general intelligence (Frederick, 2005). Consistent with what we have done in previous chapters, we used the Ten Item Personality Inventory to measure the Big Five personality traits (Gosling et al., 2003). Finally, participants responded to both the Harmful and Helpful Chairman Cases, counterbalanced for order on 7-point scale (from strongly disagree to strongly agree) (Mele & Cushman, 2006).

A mixed model analysis of variance (ANOVA) with order (Help first, Help second) as a between-subjects variable and side effect (Harm, Help) as a repeated measure revealed a large difference in side-effect intentionality judgments (i.e., asymmetry). Harmful side effects were judged intentional (M = 5.0, SD = 1.9) whereas helpful side effects were judged unintentional (M = 2.1, SD = 1.5).² To provide evidence that extraversion had unique predictive power when controlling for other individual differences, a stepwise multiple linear regression with the side-effect asymmetry as the dependent variable and personality, brief self-control, OSPAN, SAT, CRT, and sex as independent variables was conducted. The analysis revealed that only extraversion was reliably related to the judgment asymmetry.³

To further illustrate the observed relations between the Knobe effect and extraversion, were created extraversion quartiles. To create the extraversion quartiles, we looked at the overall distribution of extraversion scores for the participants. Then we identified the top 25% and the bottom 25% of the extraversion scores. We used those extraversion quartiles as independent variables in an ANOVA. This analysis revealed a large overall interaction of extraversion with judgments about the Chairman Cases. In short, those who were extraverted displayed a large difference in judgments typical of the side-effect effect whereas those who were introverted had a much more muted asymmetry (see Fig. 3.1).

Fig. 3.1

Introverts and Extraverts (bottom versus top quartile) by side-effect intentionality ratings (harm, help). Positive numbers indicate agreement, negative numbers disagreement, and error bars represent the standard error of the mean

One of the take-home messages of this study was that extraversion was related to the side-effect effect. Moreover, even after statistically controlling for several individual differences, extraversion continued to predict the judgment asymmetry. This provides some evidence that extraversion is perhaps the most important general individual difference that predicts the side-effect effect (at least in some paradigmatic cases).

Replications

If the relation of extraversion with the Knobe effect is real, then it should be found in subsequent studies that attempt to estimate that relation. We report three new studies to replicate the effect of extraversion with the Knobe effect.

In the first study, we used a probabilistically representative national sample of the United States recruited from the company Knowledge Networks (N = 295), which conducted a survey with a diverse range of people living in the United States that was essentially proportional to the actual diversity of people living in the United States (i.e., probabilistically representative). The mean age was 46, SD = 16.15 and 50% identified as female. Just as in the previous study, participants completed the two Chairman Cases presented in the harm-help order to help reduce the known order effect. Participants responded on a 7-point Likert scale (1 = strongly disagree, 7 = strongly agree). Participants also completed the Ten Item Personality Inventory but none of the other covariates from the previous study were collected because they had already been found to be unrelated to the Knobe effect (e.g., OSPAN, SAT, CRT). We replicated the relation of extraversion with the Knobe effect in this national sample r (294) = .20, p < .001, observing a pattern of findings that was similar to the result from Cokely and Feltz (2009b) that was discussed above.

Theoretically, if extraversion is systematically and pervasively related to the Knobe effect, then extraversion should also predict the judgment asymmetry in structurally similar but different scenarios. If extraversion fails to predict in structurally similar but different scenarios, there may be something idiosyncratic about the Chairman Cases that is important for the relation to extraversion. In that case, the relation of extraversion to judgments in the Chairman Cases would not provide compelling evidence for general patterns of intentionally judgments. Does the predictive power of personality generalize across different Knobe-style cases?

To provide some additional evidence for the relation between extraversion and the Knobe effect, consider the second new study we conducted designed to test the extent to which extraversion remains robust across structurally identical but different Knobe-type assessments. The structurally similar scenarios described a Dean who harmed or helped qualified applicants as a side effect. One hundred and forty-seven participants were recruited from Amazon’s Mechanical Turk panel of online workers. Thirty-two participants were excluded for not completing the survey or for requesting that their answers not be used. Fifty-two percent (N = 60) were male. Ages ranged from 18–76, M = 30.91, SD = 11.

Participants were randomly assigned to only one of two conditions. In the Harm condition, participants were given the Harmful chairman and a structurally similar Harmful Dean scenarios, counterbalanced for order. In the Help condition, participants received Helpful chairman and a structurally similar Helpful Dean scenarios, counterbalanced for order. The Harmful and Helpful Dean scenarios were as follows.

Harmful [Helpful] Dean

A professor at a university went to the Dean of the university and said, ‘I want to start a new set of criteria for admissions into the university. It will help draw more attention to the university, but it will also harm [help] a lot of qualified and deserving applicants who do not have the funds for admissions.’ The dean of the university answered, ‘I do not care at all about harming [hel**] qualified and deserving applicants from being admitted due to a lack of funds. I just want to make this university as recognized as possible. Let’s start the new program.’ They started the program and sure enough, qualified and deserving applicants without the funds were harmed [helped].

Participants were then asked on a 7-point scale (1 = strongly disagree, 7 = strongly agree) to what extent they agreed with the following statement: “The dean intentionally harmed [helped] the applicants.” After responding to the two scenarios, participants completed the Ten Item Personality Inventory and basic demographic information was gathered. In this study, we did not collect any additional covariates (e.g., CRT, OSPAN, SAT).

The Knobe effect was replicated for both the Chairman and Dean cases. Ratings of intentionality for Harmful Chairman (N = 64, M = 5.59, SD = 1.71) were significantly higher than intentionality ratings for Helpful Chairman (N = 51, M = 2.04, SD = 1.59) F (1, 111) = 129.37, p < .001, η_p² = .54. There was no main effect of order and order did not interact with judgments Fs < 1. Intentionality ratings for Harmful Dean (N = 64, M = 5.61, SD = 1.56) were significantly higher than intentionality ratings for Helpful Dean (N = 51, M = 4.6, SD = 2.31), F (1, 111) = 6.31, p = .01, η_p² = .05. However, this effect was qualified by an interaction of order. Judgments of Harmful Dean remained relatively stable when presented first (M = 5.59, SD = 1.38) compared to when it was presented second (M = 5.63, SD = 1.72), but Helpful Dean was markedly different when presented first (M = 3.65, SD = 2.41) compared to second (M = 5.57, SD = 1.49), F (1, 111) = 10.39, p = .002, η_p² = .09.

Correlations among the dependent variables for the helpful and harmful conditions are reported in Tables 3.1 and 3.2.

Table 3.1 Correlations for Helpful condition

Table 3.2 Correlations for Harmful condition

The relation of extraversion with Harmful Chairman was replicated and was the only personality trait reliably related to intentionality judgments. Neither sex nor age was related to intentionality judgments for Harmful Chairman. Largely the same pattern emerged for Harmful Dean. Extraversion trended toward significance and the size of the relation is consistent with previous studies. In the Helpful cases, extraversion was only related to Helpful Dean. This unexpected result may be partially due to the order effect present in the scenarios. Extraversion was not reliably related to Helpful Chairman.

To illustrate the relations between extraversion and intentionality judgments in the Harmful cases, rough extraversion quartiles were calculated. There were significant differences between extraverts (M = 6.69, SD = 0.48) and introverts (M = 5.14, SD = 2.07) in Harmful Chairman F (1, 23) = 5.1, p = .03, η_p² = .18. A similar difference was found between extraverts (M = 6.54, SD = 0.78) and introverts (M = 5.71, SD = 1.33) in Harmful Dean F (1, 23) = 6.06, p = .02, η_p² = .18 (see Fig. 3.2). Thus, these results suggest that the relations of extraversion to judgments about the side-effect effect are robust across different scenarios that employ harmful and helpful side effects.

Fig. 3.2

Extraversion quartiles for Harmful Chairman and Dean

We have emphasized the importance of replication in establishing relations among philosophically relevant intuitions and individual differences. To illustrate the importance of replication, some of our early work suggested that the order effect present with the Knobe effect was primarily attributable to responses of women (A. Feltz & Cokely, 2007). The somewhat surprising relation of the Knobe effect order effect with sex provides an illustrative example of some of the dangers of individual differences research (and empirical research in general), and serves as a reminder of how replications can help protect against common risks. In a third study, we attempted to replicate the finding that sex was associated with the order effect. This study also provided an additional attempt to replicate extraversions’ relation to the Knobe effect (A. Feltz & E. T. Cokely, 2011).

Again, participants responded to both the Harmful and Helpful Chairman Cases, counterbalanced (Cushman & Mele, 2008). Participants then rated their intentionality judgments about the side effects in each of those scenarios on a scale from 1–7 (disagree to agree). The results are reported in Table 3.3.

Table 3.3 Harm and Help responses by order

As the pattern of results above indicates, we observed the typical asymmetry in judgments.⁴ When we conduced additional analyses, we replicated the order effect such that intentionally judgments for the Harmful Chairman were lower when Helpful Chairman was presented first compared to when Helpful Chairman was presented second.⁵ But importantly, we failed to replicate the relation of the order effect with sex and sex was not otherwise related to the judgment asymmetry (F’s < 1).

The experiment also allowed for a replication of the relation of extraversion. In this analysis, we used a regression framework and included all of the Big Five personality traits, the CRT score, and sex as predictors and the judgment asymmetry as the outcome variable. The regression indicated a marginally significant relation of extraversion with the side-effect effect, presenting an effect size was consistent with previous studies.⁶ Again, for illustrative purposes, we divided the participants in upper and lower extraversion quartiles. Using the quartiles as the independent variable indicated a near significant interaction with the judgment asymmetry.⁷ In particular, those in the bottom quartile had lower intentionality judgments for the Harmful Chairman (M = 4.57, SD = 2.3) compared to those who were in the upper quartile (M = 6.0, SD = 2.0). Despite a reduction in power (due to reducing the number of participants in the analysis), extraversion was a significant predictor of harm judgments.⁸ Extraversion did not reliably predict Help judgments (F < 1).

Although the experiment replicated the order effect, we did not replicate the relation of the order effect with sex. This may mean that sex is not normally an influential factor in the order effect or it could mean that sex is otherwise mediated by other more proximal factors (e.g., different distributions of personality between the sexes, etc.). However, the experiment replicated extraversion’s relation to the Knobe effect. As such, this experiment serves as both converging evidence and corrective evidence, providing yet another cautionary tale for those conducting research focusing on individual differences or experimental manipulations.

The Knobe Effect, Extraversion, and Theoretical Accounts

The previous section suggests that extraversion may generally be tightly linked with the Knobe effect. But given the host of explanations offered in section “The Intentional Side-Effect Effect” of the Knobe effect, how does extraversion factor into these possible explanations? To illustrate one possible set of relations, extraversion could be related to the Knobe effect because extraversion (a) is mediated by some other concepts (e.g., extraverts may have a specific concept of intentional action where harmful, but not helpful, side effects are judged to be intentional or perhaps introverts have a different concept or set different thresholds for intentionality judgments compared to extraverts); (b) reflected an affective bias (e.g., extraverts may have an increased tendency to blame the Harmful Chairman more than the Helpful Chairman compared to introverts); or (c) both a and b. We conducted an experiment to help clarify the relations among a-c (Cokely & Feltz, 2009b). We assessed potential differences in concepts using a technique others have used where they presented some scenarios that did not generate or generated very low affective responses (Cushman & Mele, 2008; Nadelhoffer, 2006a). We attempted to test the affective biasing account by manipulation the order the scenarios were presented in. The reasoning goes like this. If non-affective scenarios were presented before potential affective scenarios, then one’s core concept of intentional action might be activated and carry over from judgments about the non-affective scenarios to the affective scenarios, thereby reducing the overall intentionally judgments in those affective cases. However, if the affective scenarios were presented before the non-affective scenarios and the affective biasing account is right, then we should see higher intentionality judgments in the affective cases compared to the alternate order. If that pattern of results is seen, then that would support, or at least be consistent with, an affective biasing account.

We conducted a study to test these different explanations of the side-effect effect. More importantly, we wanted to see if extraversion continued to predict the side-effect effect even after accounting for the multiple ways in which the effect could be produced. Specific individual differences in folk intuitions related to intentional action concepts were directly measured and manipulated (Cushman & Mele, 2008). In particular, the belief-is-sufficient concept was assessed with two scenarios. The first scenario we call Deer features a protagonist who accidentally kills another hunter as a side effect. Importantly, the protagonist did not believe he would kill the hunter and the protagonist also did not have a desire to kill the hunter.

Deer: Imagine that there is a man out in the woods who is participating in a hunting competition. After spending hours waiting for a deer to cross his path, the hunter suddenly sees the largest deer he has ever seen. If he can only kill this deer, he will surely win the competition. So, the hunter gets the deer in his sights and pulls the trigger—thereby killing the deer. Unfortunately, the bullet exited the deer’s body and struck a hunter who was hiding nearby. (Nadelhoffer, 2006a)

Participants then responded to the following prompt: “The man intentionally shot the hunter” on a 7-point scale that was used in the other studies (from disagree to agree; participants responded to the same Likert scale in all of the scenarios in this study). The purpose of this study was to highlight that sometimes there can be side effects that were not intended and not intentionally brought about.

The next scenario we called Eagle. The key features of this case were that the protagonist believes that the side effect will happen but does not want the side effect to happen.

Eagle: Imagine that there is a man out in the woods who is participating in a hunting competition. After spending hours waiting for a deer to cross his path, the hunter suddenly sees the largest deer he has ever seen. If he can only kill this deer, he will surely win the competition. So, the hunter gets the deer in his sights—but at the last second, he notices that there is a beautiful eagle perched in a tree nearby. The hunter realizes that if he shoots the deer, the sound of the gunfire will definitely cause the eagle to fly away. But he does not care at all about the eagle—he just wants to win the competition. So, he shoots and kills the deer. And as expected, the sound of the gunfire causes the eagle to fly away. (Nadelhoffer, 2006a)

After reading the scenario, participants responded to the following prompt “The hunter intentionally scared away the eagle.” Eagle allowed us to measure two different concepts of intention action. If participants responded that scaring away the eagle was intentional, then we classified those participants as having a belief-is-sufficient concept. If participants responded that scaring away the eagle was not intentional, then we classified those people as having a belief-is-insufficient concept.

Given our general theoretical perspective that most effects found in experimental philosophy (and behavioral science more broadly) are the predicable products of the interplay of person, process, and environmental factors, we predicted that many (if not all) of the factors identified above would contribute to the side-effect effect. In particular, we predicted that those who had the belief-is-sufficient concept would have higher intentionality judgments than those who had the belief-is-insufficient concept because in both the Harmful and Helpful Chairman Cases, the chairman knew that the program would influence the environment. We would also expect that extraversion would predict the asymmetry. Finally, given the observed order effects in previous studies, we expected that there would be an order effect in this study that might support the affective biasing account (i.e., when one makes a judgment about a non-affective case, one may be more likely to use one’s core concept and not be biased by other affective features of subsequent cases). The order effect should reduce the overall asymmetry when the non-affective cases are presented before the affective cases.

Since the predictions of this follow-up experiment were fairly complicated, they are summarized in the following points:

• 1. Priming the belief condition will result in an overall reduction of the intentional action side-effect asymmetry.

• 2. Those who are identified as having a belief-is-sufficient concept will judge both of the chairman’s side effects as more intentional, as compared to those who have a belief-is-insufficient concept.

• 3. Extraversion will continue to account for unique judgment variance in the intentional action side-effect asymmetry after controlling for 1 and 2.

Executing the general strategy to link intuitions to personality traits, participants completed the Ten Item Personality Inventory and responded to Harmful Chairman, Helpful Chairman, Deer, Eagle. Here, we used two “blocks” to counterbalance the order. Participants received only one of the following two orders: [Deer, Eagle, Harmful Chairman, Helpful Chairman] or [Harmful Chairman, Helpful Chairman, Deer, Eagle]. Because we knew of the order effect that obtained between Harmful Chairman and Helpful Chairman, and to avoid a potential confound in the design, all participants received Harmful Chairman before Helpful Chairman.

Results concerning the Helpful and Harmful Chairman revealed the typical overall Knobe effect (Harmful Chairman M = 5.0, SD = 2.0; Helpful Chairman M = 3.2, SD = 2.0).⁹ But things were more complicated after that basic finding. There were two higher-order interactions. The first involved those who were primed versus those who were not primed. Those who responded to the non-affective cases before the affective cases (Harmful Chairman M = 4.8, SD = 2.0; Helpful Chairman M = 3.4, SD = 2.0) had a significantly smaller judgment asymmetry between the Helpful and Harmful Chairman Cases than those who received the affective cases first (Harmful Chairman M = 5.5, SD = 1.9; Helpful Chairman M = 3.0, SD = 2.0, see Table 3.4).¹⁰ The second interaction involved the two concepts of intentional action measured in this study. Those who had the belief-is-sufficient concept had a larger judgment asymmetry (Harmful Chairman M = 6.0, SD = 1.6; Helpful Chairman M = 3.6, SD = 2.2) than those who had the belief-is-insufficient concept (Harmful Chairman M = 4.2, SD = 2.0; Helpful Chairman M = 2.8, SD = 1.7) (Fig. 3.3).¹¹ Additionally, those who had the belief-is-sufficient concept had higher ratings of intentionality for the side effects on both Harmful and Helpful Chairman.¹² We did not find evidence for a three-way interaction between concepts and priming conditions (F < 1).

Table 3.4 Intercorrelations for main variables

Fig. 3.3

Mean responses as a function of primed or not primed orders

Important for our purposes, we tested whether extraversion predicted the judgment asymmetry typical of the side-effect effect. To do so, we constructed a set of hierarchical linear regressions. Hierarchical regressions proceed by including predictor variables in a specific order and then determining if subsequent predictor variables improve model fit. We included each of the relevant independent variables as predictors in three different models. The final model had three independent variables including (1) priming, (2) extraversion, and (3) concepts as predictors of harm judgments (Table 3.5). This full model accounted for a large amount of variance in judgment.¹³ After controlling for priming, extraversion remained a reliable predictor of the harm judgments. Thus, part of the predictive power of extraversion appears to result from extraversion’s positive association with the belief-is-sufficient concept. However, extraversion’s effect was also mediated by the large effect of concepts (Table 2.2).

Table 3.5 Hierarchical linear regression analysis explaining intentional action judgments in the Harm condition

Subsequent analyses were performed to evaluate hypothesis 3. Hierarchical regression models examined extraversion within each of the two concept groups. Regression analysis indicated that for the belief-is-sufficient group, extraversion was unrelated to judgment asymmetry (F < 1). Regression analysis next assessed the belief-is-insufficient group using (1) priming and (2) extraversion as predictors. This model was a reliable predictor of the judgment asymmetry.¹⁴ Consistent with Hypothesis 3, after controlling for the effect of priming, extraversion continued to account for unique variance for those individuals who had the belief-is-insufficient concept.¹⁵ Additionally, the unique effect of extraversion was found to primarily reflect a relationship with harm judgments (Figs. 3.4 and 3.5).¹⁶

Fig. 3.4

Large, qualitative differences in the judgment asymmetry are predicted by individual differences in specific concepts (belief-is-sufficient, belief-is-insufficient). Positive numbers indicate agreement, negative numbers indicate disagreement, and error bars represent the standard error of the mean

Fig. 3.5

The side-effect judgment asymmetry (Help versus Harm) for belief-is-insufficient, which is predicted by extraversion (Introvert versus Extravert). Positive numbers indicate agreement, negative numbers disagreement, and error bars represent the standard error of the mean

The results of this experiment provide support for Hypotheses 1–3. As predicted by Hypothesis 1, the judgment asymmetry typical of the Knobe effect was smaller following priming. This provides causal evidence of the presence of judgment bias in both Harmful and Helpful Chairman (see also Nadelhoffer (2004)). However, the priming effect was relatively small compared to the large, qualitative judgment differences associated with different concepts. Consistent with Hypothesis 2, we observed that people with a belief-is-sufficient concept tended to judge actions as more intentional whereas people with a belief-is-insufficient tended to judge those actions as unintentional. Moreover, although it was not predicted, extraversion was a significant predictor of differences in concepts. Extraverts were more likely to belong to the belief-is-sufficient group whereas introverts were more likely to belong to the belief-is-insufficient group. Finally, extraversion predicted unique variance even when controlling for different concepts.

These experiments suggest that a variety of factors play theoretically important roles in intentional action judgments. The intentional action side-effect asymmetry does not appear to result from a single mechanism but instead reflects robust influences of both individual differences and judgment processes (e.g., affective biases). Previous experiments provided evidence of individual differences because extraversion was strongly related to the side-effect asymmetry even after controlling for other potentially influential individual differences (e.g., cognitive abilities, sex). The current experiment provided converging evidence of a judgment bias demonstrating that priming intentional action concepts causally reduced the judgment asymmetry. The relation between extraversion and the judgment asymmetry was replicated in the current study; however, extraversion’s effect was explained in part by its association with specific concepts. Extraverts tended toward a belief-is-sufficient concept whereas introverts tended toward a belief-is-insufficient concept. These specific individual differences in concepts were in turn associated with large, qualitative and theoretically important differences in judgment. Those who had the belief-is-sufficient concept tended to judge that all side effects were more intentional whereas those who had a belief-is-insufficient concept tended to judge that all side effects were less intentional. Finally, when individual differences in concepts were taken into account, extraversion continued to predict unique variance for harm judgments for some people (i.e., belief-is-insufficient group). In summary, the Knobe effect appears to be generally a function of the interplay of the main factors: (1) affective judgment biases, (2) specific concepts, and (3) personality.

Framing Intentional Action Intuitions

Extraversion predicts philosophical bias for some paradigmatic and theoretically important intentional action intuitions—namely, the Knobe effect. One explanation of the relation between extraversion and the Knobe effect is that extraverts may have different sensitivities or motivations than introverts. Extraverts may be more sensitive than introverts to the socially undesirable elements in the Harmful Chairman case and may be more motivated to express this. If extraverts are differentially sensitive than introverts to some of the elements in the Chairman Cases, then it stands to reason that one should be able to change the nature of the relation between extraversion and intuitions reflecting the Knobe effect. This can be done by either changing elements of the scenarios or by changing the environment in which people are asked to make decisions.

To show that intentional action intuitions can be predictably manipulated by changing key elements of the scenarios, we gave participants revised versions of Harmful and Helpful Chairman. These revised versions took inspiration from classic framing cases (Tversky & Kahneman, 1981). These classic cases have been used to demonstrate that the way that options are described can alter the choices that many individuals end up making, even if the options are apparently logically identical. In these cases, a new disease puts 600 people at risk of dying. In one condition (the negative frame), people can choose between adopting Program A where 400 people will die for sure or Program B where there is a 1/3 chance of nobody dying and a 2/3 chance of everybody dying. When described in terms of dying, around 20% of people chose program A. In a separate condition (the positive frame), people make a similar choice except the wording is changed. Instead of 400 people dying in Program A, they are told that Program C will save 200 people. Program D is described as having 1/3 chance of saving everybody and a 2/3 chance of saving nobody. Remarkably, in the saving condition, 70% of people chose Program C. Critically the expected utility of all the programs is the same. One could expect that on average, 400 people would die and 200 would live in each of the choice tasks. In terms of expected utility, the two choice tasks are identical. However, the “frame” of the choice influenced judgments of some people.

In this light, the Harmful Chairman was revised so that the side effect was framed positively (saving rain forest) even though the side effect harms the environment. Helpful Chairman was likewise altered so that the helpful side effect is framed negatively (destroying rain forest) but the environment is helped. One hundred and sixteen undergraduates at a large state university were recruited to participate in the study. One group of participants (N = 58) received the following scenario.

Chairman Help Negative

The vice-president of a company went to the chairman of the board and said, “We are thinking of starting one of 2 new programs. Either one will help us increase profits for this year’s balance sheet, and in ten years they will help the environment by affecting up to 6000 acres of a rain forest. If we choose program A, 4000 acres of the forest will be destroyed with certainty. If we choose program B, there is a 67% chance that 6000 acres will be destroyed.” The chairman answered, “I don’t care at all about the trees of the rain forest. I just want to make as much profit for this year’s balance sheet as I can. Let’s start program A.” They started the new program. Sure enough, in ten years the rain forest was helped.

A separate group of participants (N = 58) received the following scenario.

Chairman Harm Positive

The vice-president of a company went to the chairman of the board and said, “We are thinking of starting one of 2 new programs. Either one will help us increase profits for this year’s balance sheet, but in ten years they will harm the environment by affecting up to 6000 acres of a rain forest. If we choose program A, 2000 acres will be saved with certainty. If we choose program B, there is a 33% chance that 6000 acres will be saved.” The chairman answered, “I don’t care at all about the size of the rain forest. I just want to make as much profit for this year’s balance sheet as I can. Let’s start program A.” They started the new program. Sure enough, in ten years and the rain forest was harmed.

Participants were then asked to rate their agreement with the appropriate version of the following statement (on a 7-point scale, 1 = strongly disagree, 4 = neutral, 7 = strongly agree): “The chairman intentionally harmed/helped the rain forest.” The additional information about risk describes logically identical options—on average there would be 4000 fewer acres of rain forest. However, the description of “saving” and “destroying” theoretically should alter some people’s judgments about the intentionality of the side effect. Specifically, if extraverts are more sensitive to socially unacceptable consequences or are motivated to act on those consequences, destroying the rain forest should impact their intentionality judgments more than introverts even if the side effect ultimately helps. Similarly, in the Harm case, when the action is described as saving the rain forest, extraverts should be more likely than introverts to judge the action as less intentional even if the side effect is harmful. However, introverts should be relatively less affected by this additional information and give responses similar to pervious experiments (i.e., display a muted Knobe effect). After reading one of the scenarios, participants filled out the Ten Item Personality Inventory.

An ANOVA indicated a statistically significant, but markedly reduced, Knobe effect between the Chairman Harm Positive (M = 3.88, SD = 1.73) and Chairman Help Negative (M = 2.29, SD = 1.57).¹⁷ This muted Knobe effect appears to be the result of the framing. We observed the predicted reversal in extraversion’s relation to judgments about the harmful and helpful side effects. When the harm was framed positively (i.e., saving parts of the rain forest), extraversion was negatively related to intentionality judgments, r (58) = −.27, p = .04. Likewise, when the help brought about was framed negatively (destroying parts of the rain forest), extraverts were more likely to judge the help to have been brought about intentionally r (58) = .33, p = .01. To illustrate further, rough quartiles were constructed for extraverts and introverts. Extraversion had a statistically significant interaction effect with judgments about Chairman Harm Positive (Extraverts: N = 13, M = 3.69, SD = 1.84, Introverts: N = 13, M = 4.69, SD = 1.65) and Chairman Help Negative (Extraverts: N = 14, M = 3.21, SD = 2.01, Introverts: N = 15, M = 1.73, SD = 1.39)¹⁸ (see Fig. 3.6).

Fig. 3.6

Intentional action framing by extraversion

As this study suggests, the relation of extraversion to judgments about side effects can be predictably manipulated. In this case, there is something about losses that extraverts seem to be especially sensitive to. When the emphasis on losses is minimized, extraverts have a different set of intuitions. As such, intuitions about the Knobe effect are partially a result of the decision task along with personality. In the next section, we detail an experiment that indicates that intuitions about the Knobe effect are also partially a result of the affordances in the decision making environment—i.e., the choice architecture.

Perspective in Intentional Action Attributions

To take stock, some people judge the intentionality of some bad side effects differently from some good side effects. Extraversion predicts some of these differences. Moreover, intuitions about the Knobe effect can be manipulated, and the manipulation can be predicted based on some of extraverts’ sensitivities. One common feature of the studies reviewed so far is that participants were asked to imagine some scenarios and then make judgments about them. However, it seems likely a different pattern of judgments would be found if participants were actually performing actions with side effects. Indeed, there is a large literature documenting differences in people’s judgments and behaviors as a function of whether one is performing an action (i.e., an actor) versus observing the behavior of another (i.e., an observer) (Jones & Nisbett, 1972; Malle, 2006; Malle & Knobe, 1997; Malle, Knobe, & Nelson, 2007). Just imagine driving a car when somebody cuts you off. What do you think of that person? Now imagine that you do exactly the same thing and cut someone off. Are your thoughts about your own versus the other person’s behaviors different? The literature suggests that they would be. For example, you would likely have a much harsher judgment of the person who cut you off (that jerk!) compared to your own behavior (I was just in a hurry!).

The same basic principles have been applied in experimental philosophy. Consider the following cases where one is asked to make a judgment as a third-party observer or as a first-person actor:

A trolley is hurtling down the tracks. There are five workers on the track ahead of the trolley, and they will definitely be killed if the trolley continues going straight ahead since they won’t have enough time to get out of harm’s way. There is a spur of track leading off to the side where another person is working. The brakes of the trolley have failed and there is a switch which can be thrown to cause the trolley to go to the side track. Imagine that you are an innocent bystander who happens to be standing next to the switch. You realize that if you do nothing, five people will definitely die. On the other hand, you realize that if you throw the switch, you will definitely save the five workers. However, you are also aware that in doing so the worker on the side track will definitely be killed as the result of your actions. (Nadelhoffer and Feltz (2008) see also Petrinovich and ONeill (1996))

As the cases illustrate, the only major difference between the scenarios was whether judgments were made about John or about “you.” The key dependent variables in this scenario were whether it was permissible to throw the switch and how much control one had over the situation. Consistent with the actor-observer literature, people tended to think it more permissible for John to throw the switch (90% said permissible) and that John had more control (M = 3.72) than when they were given the “you” version (65% and M = 2.88). These results indicate that one’s viewpoint can influence judgments about moral permissibility and control (Nadelhoffer & Feltz, 2008). This finding is consistent with a large literature that differences in perspective can alter some judgments including reasons for performance on tests (Nisbett & Wilson, 1977), reasons for choices of academic major or girlfriends, and reasons for volunteering (Nisbett, Legant, & Marecek, 1973) (see for reviews, Baron and Branscombe (2012); Malle (2006)).

Given the pervasive impact of actor-observer differences, these differences would also likely exist for some intentional action intuitions. One way to illustrate the actor-observer difference with respect to the Knobe effect is to take a similar strategy used in the Trolley case above—simply ask one group to imagine they are the chairman and compare response to another group that responds to the normal Chairman scenarios. However, when participants were asked to imagine being the chairman, the same patterns of results typical of the Knobe effect were found and not reliably different from the original cases where one was an observer. A stronger manipulation than asking people to imagine being an actor or an observer is likely needed. Feltz, Harris, and Perez (2012a) thought that one way to generate these effects was not just to ask people to imagine being the chairman but to put them in a situation where they actually were (like) the chairman. That is, they thought that participants needed to take some real action and make a decision that had some good or bad outcome to be like the chairman.

To allow for participants to actually become actors, Feltz, Harris, et al. (2012a) had participants play a “game.” The game was the kind of game that is typical for an area of research called Behavioral Economics where people made decisions about allocating resources against other people in the same game. In this game, the decisions that the players make impact other players. How well a player does in the game depends both on the individual decisions of the player along with the decisions of all the other players in the game. In Behavioral Economics, “how well” a player does is often determined by how many “experimental currency unites” (or ECUs) a player accrues in the game. Then, at the end of the game, the players trade in their ECUs for real money. So, there is no lip service in this game since the rewards (and punishments) were real and not hypothetical.

The game that the players played was as follows. Participants came to a computer lab in 4 different groups of between 4 and 12 people (total number of participants was 45). Participants were instructed about the rules of the game before starting to play. They were told that they would earn $10 for showing up to the study (a standard at the time in Behavioral Economics studies) and they would have the chance to earn more money depending on how they and others played (as a matter of fact, final payoffs were between $16 and $20). After getting these instructions, participants were assigned to one of four conditions. Participants were assigned to a Help and a Harm condition. In the Harm condition, participants were instructed that they had ten tokens that they had to decide what to do with. They only had two options: they could keep all their tokens, or they could contribute any percentage of those tokens (i.e., 1–10) to a “group account.” For each token they kept, they would get 10 ECUs toward their final reward at the end of the experiment. However, here’s the rub. If the player invests a token in the “group account,” that player would earn 12 ECUs but all the other players in the game would receive a 3 ECU penalty. The other condition was the Help condition. In the Help condition, the instructions were exactly the same as they were in the Harm condition except that contributing to the group account would generate a 3 ECU bonus to others in the game. The final condition was the Observer condition. In this condition, participants did not have to take any action but read about the action another person took that impacted them (i.e., they observed the game playing action of another person). In this case, either the other person’s action generated a bonus or penalty in line with the Harm and Help conditions. In this case, Feltz et al. (2012a) simply set things up so that somebody was described as contributing ten tokens to the group account thereby generated a 30 ECU bonus or penalty. Hence, given the setup of the game, actor-observer differences might be observed not for hypothetical but for real cases.

The four conditions that participants were exposed to were determined as follows. People were both actors and observes but only in one of the Help or Harm conditions. The four conditions were generated by counterbalancing the orders of being an actor or an observer. After each instance of observing or acting, participants responded to the key 3 statements in this study (the actual contribution to the accounts was not the main dependent variable in the study):

• 1. “You/the other participant intended to generate the penalty/bonus.

• 2. You/the other participant intentionally generated the penalty/bonus.

• 3. You/the other participant are/is blameworthy/praiseworthy for generating the penalty/bonus.”

Participants responded on a 7-point Likert scale (1 = disagree, 7 = agree).

Feltz, Harris, et al. (2012a) were only interested in looking at responses to 1–3 from people who performed some action. It was simply an unavoidable element of the experiment that some people might decide not to contribute to the group account in which case they did not perform an action to contribute. So, they only looked at participants who contributed to the group account. Overall, 18 people made a decision to contribute to the group account in the Harm condition and 20 people made a decision to contribute to the group account in the Help condition. The responses to prompts 1–3 indicated that there was indeed an actor-observer asymmetry in judgments (see Table 3.6).

Table 3.6 Means and standard deviations for actors’ and observers’ intention and intentionality judgments

So far, these results suggest that there can be actor-observer differences with intentional action intuitions. But these overall intuitions do not necessarily inform actor-observer differences associated with the Knobe effect. The harm/help to the other participants may not be side effects of the intended action to make greater profits. To find those who viewed the harmful or helpful consequence as a side effect, those who reported that the harm or help was not intended were identified (those answering 1–4 on the intention question above). After these exclusions, 16 participants did not intend the harmful outcome of investing and 10 participants did not intend the helpful outcome. As predicted, Feltz, Harris, et al. (2012a) found an actor-observer difference with the side effects. In the Actor condition, the harmful outcomes were thought to be less intentional (M = 2.31) than the helpful outcomes (M = 3.1). However, there were no reliable differences in the Observer condition

A question that was centrally important for the purposes of this chapter is whether extraversion was related to the asymmetry in judgments. A similar strategy was taken in this study and those reviewed above. Participants also completed the Ten Item Personality Inventory. Here, extraversion was strongly correlated with the harm judgments for those in the Actor condition, r = −.55, p = .03, and was not significantly correlated with actors’ intentional help judgments r = .40, p = .25. To help further understand those relations, Feltz, Harris, et al. (2012a) created two groups based on extraversion scores. Those two groups were identified by those who were in the top half of extraversion versus bottom half of extraversion (i.e., a median split). Using the median split as an independent variable along with the Harm/Help condition resulted in the expected interaction (see Fig. 3.7).

Fig. 3.7

Extraversion median intentionally means. Error bars represent 1 standard error

Meta-Analysis

Given the number of studies reported in this chapter, it is useful to have an overall summary of the relation between extraversion and the Knobe effect. So, we conducted a meta-analysis to combine the results reported in this chapter.

Some of the effects reported in this chapter were predicted to have the opposite sign. For example, it was expected in the framing cases (i.e., framing the Harmful chairman positively), the effect of extraversion would be opposite of the normal direction of the relation of extraversion with the Harmful chairman. Because these differences were predicted, we accounted for this difference by changing the direction of the signs so that they would be consistent with the predictions of extraversion’s relation to the Knobe effect (i.e., we made all the predicted relations positive).

Given these stipulations, the overall mean effect size between extraversion and judgments typical of the Knobe effect was estimated to be .24, (95% CI 0.17–0.31), p < .001 (see the forest plot in Fig. 3.8). No heterogeneity was observed in the effect sizes Q (7) = 3.72, p = .82, suggesting that the relation remained stable regardless of the experiment. A test of the funnel plot did not reveal evidence of publication bias, z = 1.64 p = .1, but this result should be interpreted with caution since we did not do a complete, full search for any unpublished studies. Hence, there is reason to be quite confident that the relation between extraversion and the Knobe effect is robust across several different testing environments and materials.

Fig. 3.8

Forest plot of the relation between extraversion and the Knobe effect

Conclusion

In this chapter, we have surveyed a number of different studies suggesting that extraversion is systematically related to an important class of intentional action intuitions about side effects. The variety of experiments suggests that there is not something idiosyncratic about the classic Chairman Cases that accounts for the relation between extraversion and intuitions about side effects. Changing the nature of the scenarios predictably changed the relation of extraversion with intentionality judgments. Similarly, changing the decision making environment predictably altered the relation of extraversion to intentional action intuitions. The meta-analysis suggested that the relation remained consistent across these different experiments. In summary, experimental data and statistical modeling indicate that many intuitions relevant to the Knobe effect are systematically related to some general, heritable personality traits.

The persistent relation of individual differences to the Knobe effect challenges many explanations of the Knobe effect. Almost all the models concerning the Knobe effect are based on mean responses and do not take into account individual differences. Additionally, most of the theories at least tacitly hold that there is one factor that can account for the Knobe effect. Even at the broadest level of explanation (i.e., core concept v. non-concept explanations), the notion that a single explanation for the side-effect effect can account for the varied responses is just wrong.

The data we have reviewed present a substantially different picture of what would be required for an adequate account of the Knobe effect. There is not necessarily any intentional action judgment asymmetry as there may not typically be any complete judgment reversal. Rather than any single general “bias” or judgment process that causes participants to switch from intentional (harm) to unintentional (help) judgments, the change in judgment appears to be more modest, changing for example from neutral judgments to intentional or from unintentional to neutral. The only identifiable and somewhat complete judgment reversal involved the group of extraverts who behaved as if they held a belief-is-insufficient concept. Additionally, we found that by changing the task (framing) or the judgment environment (making people actual actors) we could predictably change the overall pattern of intuitions. In these ways, the observed judgment asymmetry, which seems to be the product of the interplay of several distinct mechanisms, may be more accurately characterized as a collection of intentional action biases or multiple judgment asymmetries.

If all of this is right, then our results indicate that a parsimonious account of folk intentional action intuitions is not likely forthcoming. Intentional action intuitions appear to be related to a number of independent factors including one’s personality, biases, the task, and the task environment. Because there is no single judgment asymmetry, the intentional action intuitions generated typical of the Knobe effect seem to indicate that there is no the folk concept of intentional action. Once again, the evidence suggests there is no single, monolithic folk concept of intentional action. Rather, there appear to be several. We expect that there will be “groups of folk” who express different, stable, predictable, and philosophically interesting intuitions (Cushman & Mele, 2008). Hence, proposing a theory or conceptual analyses based on the presupposition that there is only one folk concept, set of intuitions, or mechanism will be incomplete or inaccurate.

It is still possible to propose a single theory or conceptual analysis in light of the different factors that are related to intentional action intuitions. But possibility is cheap. The mounting evidence suggests that there is stable diversity in people’s intentional action intuitions. If there is predictable and stable variation of intentional action intuitions, then a theory or conceptual analysis concerning intentional action that takes those intuitions as evidence must account for that variation in some way. For example, theorists may treat the variation as confirming evidence (e.g., relativism, conceptual diversity), or the theorist may explain why at least some of the intuitions are wrong (e.g., an error theory). In any event, an account of those intuitions is required for most theories or conceptual analyses about intentional action. Simply holding that it is possible to give such an account is insufficient to deflect the worries presented by stable individual differences in intuitions. Therefore, the philosophical use of intentional action intuitions requires a comprehensive understanding of the extent of variation in those intuitions.

In conclusion, the evidence presented in this chapter provides support that an important class of intentional action intuitions is related to a global personality trait. In the previous chapter, we have seen that some free will intuitions are related to one’s personality. Consequently, there is gathering evidence that personality is meaningfully related to many philosophically relevant intuitions and that the relation is not limited to one kind of intuition (e.g., intuitions about free will). In the next chapter, we review another set of philosophically relevant intuitions related to personality—ethical intuitions.