The influence of desire and knowledge on perception of each other and related mental states, and different mechanisms for blame

Journal of Experimental Social Psychology 60 (2015) 27–38

Contents lists available at ScienceDirect

Journal of Experimental Social Psychology journal homepage: www.elsevier.com/locate/jesp

The influence of desire and knowledge on perception of each other and related mental states, and different mechanisms for blame Sean M. Laurent a,⁎, Narina L. Nuñez b, Kimberly A. Schweitzer b a b

University of Oregon, 1227 University of Oregon, Eugene, OR 97403, USA University of Wyoming, 1000 E. University Drive, Laramie, WY 82071, USA

H I G H L I G H T S • • • •

Knowledge and desire affect blame through different routes. Knowledge and desire affect perception of each other, perception of related mental states, and immorality. Knowledge affects perceived awareness through blame. Desire affects blame through judgments of agent immorality and perceived awareness.

a r t i c l e

i n f o

Article history: Received 31 May 2014 Revised 11 April 2015 Accepted 28 April 2015 Available online 30 April 2015 Keywords: Knowledge Desire Blame Mental states Negligence Morality Moral psychology

a b s t r a c t Two experiments (Experiment 1 N = 350; Experiment 2 N = 153), used relatively simple (Experiment 1) and complex (Experiment 2) vignettes to investigate whether two ostensibly distinct mental states that underlie intentionality judgments influence each other, related mental states, and agent morality, and also whether they affect blame through different routes. Knowledge (that a particular action can lead to a particular outcome) affected perceptions of an agent's desire by first increasing blame, which increased perceptions that the agent was aware of acting, while acting. Desire (for a particular outcome) affected blame and perceptions of agent knowledge by increasing perceptions that the agents were immoral (measured after knowledge and desire were described, but before the agents' action and the harmful outcomes were described), which influenced perceptions of the agents' awareness. The importance of these findings for mental state perception research, including the relationship of mental states to blame, is discussed. © 2015 Elsevier Inc. All rights reserved.

1. Introduction Understanding others' mental states is a vitally important aspect of human social interactions. Not only does perception of others' mental states aid in understanding their motives, actions, and character (e.g., Reeder, 2009a, 2009b), it helps perceivers select appropriate responses to what others say and do. For example, knowing that someone wishes you harm, you may choose to avoid or confront that person. Surmising that a friend is blue, you might console or offer help. Simply put, understanding of or beliefs about the contents of others' minds shapes perceivers' reactions to and behavior toward them (Malle & Hodges, 2005). Understanding mental states is particularly important when trying to evaluate how blameworthy an agent is for bringing about a harmful ⁎ Corresponding author at: Department of Psychology, University of Oregon, 1227 University of Oregon, Eugene, OR 97403, USA. E-mail addresses: [email protected], [email protected] (S.M. Laurent).

http://dx.doi.org/10.1016/j.jesp.2015.04.009 0022-1031/© 2015 Elsevier Inc. All rights reserved.

outcome, and most theoretical models describing how people blame take this into account (e.g., Alicke, 2000, 2008; Heider, 1958; Shaver, 1985). That is, reasoning about mental states such as whether an agent wanted to cause harm, foresaw the potential for harm in an action, or acted intentionally to fulfill a goal of harming aids perceivers in determining how responsible, blameworthy, and deserving of punishment the agent is. For example, Lagnado and Channon (2008) found that relative to difficult-to-foresee harms or harms that resulted from unintentional actions, foreseeable and intentional harms were rated as more blameworthy and caused by agents' actions. Similarly, relative to negligent or reckless actions, intentional actions that lead to harm are seen as more blameworthy and deserving of punishment (Darley & Pittman, 2003; Malle & Nelson, 2003; Nobes, Panagiotaki, & Pawson, 2009; Shultz & Wright, 1985). Intentionality judgments are complex but can be made quite quickly (e.g., Malle & Holbrook, 2012), even though many theoretical models suggest that judgments regarding several distinct mental states are required prior to an agent's action being considered intentional. For

28

S.M. Laurent et al. / Journal of Experimental Social Psychology 60 (2015) 27–38

example, along with requisite elements such as ability/skill or effort expended, most models of intentionality include or specifically require the presence of mental states such as belief/knowledge, desire, awareness, and intent (e.g., Adams, 1986; Heider, 1958; Jones & Davis, 1965; Malle & Knobe, 1997; Shaver, 1985). Although research has begun to investigate the relative contributions to blame of different mental states that underlie intentionality judgments (e.g., Cushman, 2008; Nuñez, Laurent, & Gray, 2014), little is known about whether perception of certain mental states influences perception of other presumably distinct mental states. Moreover, little is known about the routes through which different mental state inferences affect blame. The current paper considers how desire and knowledge may reciprocally influence each other, related mental states, and agent morality, and the different routes by which these mental states impact blame.

2. The role of desire in blame Desire plays an important role in intentionality judgments because intentional action typically serves agents' hopes of bringing about the outcomes they desire (e.g., Malle & Knobe, 1997). However, without the conjoint presence of other mental states such as knowledge or awareness, desire should not be particularly blameworthy. That is, desiring that harm befall someone – a feeling that many people may, in their less proud moments, admit to – does not imply an intention to act in a way that will bring it about. Although it seems obvious that wishing someone harm should not, on its own, rationally lead to blame, there are reasons to believe that it might, particularly when the desired outcome occurs and the desiring agent plays a causal role in bringing about the outcome (e.g., Cushman, 2008). One reason for this is that desiring (or admitting to desiring) harm is non-normative (e.g., Guglielmo & Malle, 2010; Uttich & Lombrozo, 2010) and socially informative. With other factors held constant, moral people should typically not want to see others harmed, suggesting that people who wish others harm might be seen as immoral. Because social perceivers like it when bad people get their just desserts (Darley & Pittman, 2003), it becomes easier to understand why perceivers irrationally blame and punish moral transgressions (e.g., hypocrisy; Laurent, Clark, Walker, & Wiseman, 2014). In addition, because desire is implicated in antisocial motives for action and informs character judgments (Reeder, 2009a, 2009b), when an agent wants to see someone harmed and does something that leads to the desired outcome occurring, perceivers may label the person as “bad” and assume the agent somehow engineered the outcome. The idea of a route from desire to blame is also consistent with Alicke's (2000) culpable control model. For example, the blamevalidation hypothesis posits that when harm occurs, blame may be the “default attribution” (p. 558), leading people to process information in a biased way that results in holding a causal agent accountable. According to the culpable control model, biases in how people blame can affect their evaluation of links anywhere in the mind-to-behavior-toconsequence sequence. As one example, perceivers may have spontaneous negative reactions to a harmful action–outcome sequence based on a dislike of agents' motives or values. One way spontaneous reactions may increase blame is by altering perceptions of agents' mental states and the relation of these states to the behavior that caused the outcome. Thus, perceivers' reactions to an agent's unsavory desires, which provides a cue to motives and suggests values, might first negatively impact evaluations of the agent's character. Subsequently, character evaluations could influence perceptions of blame, leading to changes in the way people reason about an agent's volitional control (Alicke, 2000), such as by increasing the perception that the agent foresaw the risk of harm, particularly when the desired harm actually occurs. This suggests that when a desired harm occurs as a result of an agent's action, the presence of desire could influence blame through evaluations of moral character, subsequently affecting beliefs that the agent was aware of

acting, and thus possessed knowledge linking their action to the outcome, validating the impulse to blame. 2.1. The role of knowledge (and awareness) in blame Depending on interpretation, knowledge is often loosely equated with foresight and defined as a belief that one's action will have a particular consequence (e.g., Malle & Knobe, 1997; Shaver, 1985). This definition appears to imply either a present-moment action or an intention to act and bring about a desired outcome. However, knowledge can also be conceptualized more abstractly as a simple understanding of potential causal relations between actions and outcomes. That is, to know that one's action will have a particular consequence, one must first know that certain types of actions or classes of actions have the potential to bring about certain types of outcomes or classes of outcomes. Using this definition, knowledge informs subjective foreseeability (Lagnado & Channon, 2008) at a basic level (i.e., whether an agent could plausibly foresee or have foreseen the consequences of an action); however, it does not imply that any action has been or will be taken or that any outcome has occurred or will occur. That is, although most people know that certain actions can cause harm, their possession of knowledge does not imply their performance of these actions or plans to perform them. Whether a person should reasonably foresee or should have foreseen the potential for harm (e.g., reasonable foreseeability; Lagnado & Channon, 2008) can be shown to depend on the type of knowledge being considered, as well as on the person's level of awareness of performing a particular action linked to that knowledge. General or common knowledge subsumes knowledge that most people possess or are reasonably expected to possess. For example, a person who hates guns, has never fired one, and intends never to do so probably knows (and would be reasonably expected to know) that pulling the trigger on a pistol, when the pistol is loaded and aimed at someone, will probably lead to the person being injured or killed. In addition to common knowledge, some knowledge is privileged or specialized, in that not all people would be expected to possess it, diminishing reasonable expectation for some people but increasing it for others. For example, while a friend of Person P might know that P is allergic to strawberries, and would therefore have knowledge regarding the effect of feeding P strawberries, not all people would be expected to have this knowledge. Although knowledge in this sense is not noteworthy, it strongly suggests the presence of foresight when combined with awareness of performing the action one knows is linked to the possibility of harm. However, awareness itself is a complex mental state, because actions themselves can be construed or identified at different levels (e.g., Vallacher & Wegner, 2012), and agents may or may not be aware of the full scope of their actions. Continuing with the gun example, a person might pull the trigger on a gun they think is unloaded and end up harming someone because a bullet was in the chamber. In this case, the agent may have awareness while acting of pulling a trigger on a gun, but not of pulling the trigger on a loaded gun, so no foresight should be attributed. However, it could be argued that the agent should have been aware of the full scope of his action (i.e., he should have taken care that the gun was not loaded before pulling the trigger if there was no intent to harm). Similarly, P's friend might be aware of innocently offering her a sip of juice purchased at a health store without being aware that the juice contains strawberries. But if the friend is aware the juice contains strawberries, and knowledge is also present, her awareness of acting suggests foresight and perhaps the intent to harm. Thus, only when knowledge is combined with full awareness should foresight be rationally and fully attributed. Following this line of reasoning, expectations for foreseeability can arise not only from expectations for knowledge when awareness is present (e.g., that firing guns can cause harm or that P is allergic to strawberries), but also from expectations for awareness when knowledge is present (e.g., that one is pulling the trigger on a loaded gun or that one is offering a drink containing strawberries).

S.M. Laurent et al. / Journal of Experimental Social Psychology 60 (2015) 27–38

In our view, when considering how blameworthy an agent is for an ostensibly unintended outcome, the task of rational social perceivers is (among other things) to determine whether knowledge and awareness were both present in full, and if not, to what extent each state should have been present. However, when it comes to blame, social perceivers are not always (or perhaps, not even usually) rational. Consider again Alicke's (2000) blame-validation hypothesis. If perceivers are motivated to assign blame and they reason about an agent who possesses knowledge and whose action (linked to the knowledge) causes harm, their first impulse might be to blame the causal agent, and then in a process of blame validation, inflate their perceptions of the agent's awareness of acting (i.e., alter their perceptions of the agent's foresight). That is, when an agent lacks awareness that she or he reasonably should possess, the presence of knowledge might bias perceivers' judgments that awareness was actually present because of a desire to blame the agent. And even when awareness is reasonably absent (i.e., the agent could not be reasonably aware of the full scope of her or his action), to the extent that perceivers think the agent had knowledge, they may again exaggerate awareness to validate their blaming. It is only a short jump then to hypothesize how knowledge could indirectly effect perceptions of an agent's desire to harm through increases in blame and awareness: If an agent knows some action can cause harm and the outcome occurs, prompting attributions of blame, this might subsequently affect perceptions that the agent in fact wanted the outcome to occur.

2.2. The present experiments The analysis presented above suggests that routes from knowledge and desire to blame may differ. When awareness is claimed as absent but an agent's actions cause harm, knowledge or perceptions of knowledge should affect perceptions that the agent deserves blame and was aware of acting. It is also possible that judgments of blame will mediate perceptions of awareness. In turn, the knowledge to blame to awareness path should further mediate perceptions regarding the agent's desire, because increased blame should prompt perceivers to attribute awareness (or the reverse), and awareness of action (i.e., when the agent knows the action could cause harm and acts, leading to the outcome) suggests the agent desired the harm. When an agent's actions lead to harm, desire should primarily increase beliefs that the agent is immoral, but might also have total effects on blaming the agent and perceptions about the agent's possession of awareness. Immorality should mediate perceptions of blame and awareness. Finally, the desire to immorality to awareness and blame links should also mediate perceived knowledge. To test these hypotheses, in Experiment 1 we adapted a simple scenario used by Nuñez et al. (2014), where a woman fries chicken for a liked or disliked friend using cooking oil containing peanut oil. The friend, severely allergic to peanuts, gets sick. The agent, while aware of acting in the general sense, had no awareness of the scope of her action, because she never read the label on the oil (i.e., although aware of cooking with oil, she was unaware of cooking with oil that contained peanut oil). Desire (i.e., to see the friend harmed) and knowledge (i.e., of the friend's allergy, and thus of the potential link between an action and an outcome) were independently manipulated. In Experiment 2, we created a new, more complex scenario, where (among a variety of extraneous background details) a man (the agent) tells his housemate's girlfriend that the housemate is at a female friend's house. The girlfriend subsequently breaks up with the housemate. Among other details, desire was manipulated by describing the agent as romantically interested or not in the housemate's girlfriend. Knowledge was manipulated by describing that the agent knew or did not know about a conflict between the housemate and his girlfriend that involved an unknown (to the agent) ex-girlfriend. In all cases, the agent's awareness of the scope of his actions was deliberately limited so that he could not have

29

rationally foreseen the potential for harm in his action. The full scenarios for both experiments are available in Appendix A. In both experiments, to provide a strong causal test of the desire to immorality to blame (and other mental states) hypothesis, participants were first presented with information detailing the agents' mental states (i.e., manipulations of desire and knowledge). At this point, moral judgments were assessed. Following this, the agents' actions were described along with the harmful outcomes. Last, we collected data on the remaining dependent variables.

3. Experiment 1 3.1. Method 3.1.1. Participants Two samples were collected. The first was 160 U.S. citizens aged 18 or above, recruited through Amazon's Mechanical Turk website and paid $0.20 (69% male, MAge = 29.01, SD = 9.72, range = 18–67). The second was 201 undergraduate students (32.8% male, MAge = 19.80, SD = 2.27) who participated in partial fulfillment of course requirements. After deleting cases for multiple responding or incorrect responses to a simple (arithmetic) attention check question (6.9% of the Internet sample only), the final sample was N = 350.

3.1.2. Procedure After providing consent, participants read a short vignette (Nuñez et al., 2014; see Appendix A). In it, “Annie” invites “Julia” to dinner. Annie is described as behind on time as she hastily prepares fried chicken using cooking oil containing a small amount of peanut oil. However, Annie never read the label, so she lacks awareness of the full scope of her action (i.e., she is not aware, while using the oil, of the oil's composition). Julia, who is severely allergic to peanuts, gets very sick after eating the chicken. The crossed manipulation of knowledge and desire (described below) came first, followed by questions about Annie's immorality. Participants then read about Annie's action (cooking the chicken in the oil) and the outcome (the friend getting ill) before responding to the remaining dependent variables. Last, participants answered an attention check question and provided basic demographic information.

3.1.3. Knowledge and desire manipulations Knowledge was manipulated by describing that Annie knows (does not know) Julia has a severe peanut allergy, and therefore knows (could not know) the risk of exposing her to peanuts. Desire was manipulated by describing that Annie cares deeply for (secretly resents) Julia and would hate (wishes) to see her suffer. 3.1.4. Dependent variables1 Participants rated their agreement with statements measured on 10-point scales (1 = completely disagree, 10 = completely agree). Responses were averaged to form composite measures (items were reverse-scored where appropriate) and higher scores indicate greater endorsement of each construct. Reliability coefficients or correlations are given in parentheses. Participants completed variables in the order presented below. 1 Three items also measured the “wrongness” of Annie's actions. This variable was very strongly correlated with the blame measure (r = .94) and analyses did not substantively differ when using wrongness versus blame. We report the blame variable here, which most closely represents the construct of interest. Additionally, three items measured whether Annie “should have been aware” of the scope of her actions and the risk of harm, and two variables measured Annie's awareness of the risk of her actions. Because these variables did not clearly represent simple measures of awareness, but instead measured aspects of foresight or expectations for foresight, no analysis of these variables is reported.

30

S.M. Laurent et al. / Journal of Experimental Social Psychology 60 (2015) 27–38

3.1.5. Immorality (α = .97) Five items measured moral judgments: “Annie is…” “a good person,” “a bad person,” “a decent person,” “moral,” “immoral.” 3.1.6. Awareness (r = .64) Two items measured Annie's awareness of acting: “Annie was aware, while preparing dinner, that the oil she was using contained some peanut oil” and “Using that cooking oil that contained peanut oil, Annie was aware of what she was doing.” 3.1.7. Blame (α = .90) Three items measured blame/responsibility: “Annie…” “should be held responsible for the harm to Julia,” “deserves to be blamed for what happened to her friend Julia.” “It is Annie's fault that Julia got sick.” 3.1.8. Perceived knowledge (α = .99) Four items measured perceived knowledge: “Annie knew (did not know)…” “her friend Julia was allergic to peanuts,” “that if Julia ate peanuts she would get sick.” 3.1.9. Perceived desire (α = .98) Five items measured perceived desire: “Annie wanted (did not want) …” “her friend Julia to suffer.” “Annie…” “hoped something bad would happen to her friend Julia,” “didn't want anything bad to happen to her friend Julia,” “wished harm would befall Julia.” 4. Results and discussion 4.1. Total effects All measures were examined using 2 (knowledge present vs. absent) × 2 (desire present vs. absent) ANOVAs2 (df were 1, 346 in all analyses; see Table 1 for M, SD, d, and p of all measured variables as a function of knowledge and desire and for correlations among the variables). As expected, knowledge had a total effect on perceived knowledge (F = 1599.51, p b .001), blame (F = 125.37, p b .001) and awareness (F = 16.88, p b .001), but not on perceived desire (p = .55). The total effect of knowledge on immorality was marginally significant (F = 2.99, p = .085). Desire had a total effect on perceived desire (F = 1138.77, p b .001), immorality (F = 606.22, p b .001), blame (F = 9.94, p = .002), and awareness (F = 26.22, p b .001), but had no total effect on perceived knowledge (p = .49). A significant interaction between the manipulations emerged for immorality (F = 5.23, p = .02). Followup analyses showed that knowledge significantly influenced immorality when desire was present (p = .02) but not absent (p = .63). Desire significantly impacted immorality at both levels of knowledge (ps b .001). Strong support was found for the effects of both variables on blame and awareness. Desire also had a strong effect on immorality, even though this variable was measured before participants knew of the agent's action or the harm the action caused, suggesting that simply knowing a person desires harm is enough to inform judgments of the agent's moral character. Similarly, when desire was present, knowledge had a significant total effect on immorality. This demonstrates the potential for a strange (but not hypothesized) bias, whereby people who already judge a person as immoral (e.g., because of their unsavory desire to harm someone) see their simple possession of knowledge about how an action might harm as indicative of their character, even though the person has not acted and no one has been harmed. 2

Initially, sample (Internet vs. undergraduate) was included as a factor in all analyses. No main effects involving sample emerged, but one interaction between sample and knowledge was found (on awareness, p = .02) that did not affect interpretation of knowledge (i.e., the effect of knowledge on awareness was larger in the undergraduate sample than in the Internet sample).

4.2. Path analyses: the effects of knowledge on blame, awareness, and desire Path analyses with bootstrapping (5000 replications; N = 350) were used to examine mediation hypotheses. The first model examined whether the manipulation of knowledge would predict awareness through blame. Direct paths were estimated from the manipulation of knowledge to blame and awareness and from blame to awareness. Direct paths from knowledge to blame and from blame to awareness were significant, as was the indirect path (ps b .001). The path from the knowledge manipulation to awareness was not significant (p = .62). Removing this nonsignificant path left a model with good fit and where all paths were significant (χ2(1) = 0.24, p = .63, RMSEA = 0.00; ps b .001; see Fig. 1, top, which also provides all direct and indirect path coefficients, R2 for all variables, and bias-corrected 95% confidence intervals). When the causal order of awareness and blame (i.e., where awareness predicts blame) were reversed, knowledge significantly predicted awareness (p b .001), and awareness significantly predicted blame (p b .001), but knowledge continued to directly predict blame (p b .001). The indirect effect of knowledge on blame was also significant (p b .001). Removing the direct path from knowledge to blame left a model that did not fit the data well, χ2(1) = 89.62, p b .001, RMSEA = .50. Thus, although a model that reversed causality was also plausible, making the temporal ordering between blame and awareness difficult to determine, the most parsimonious model was one where knowledge worked through blame to affect perceptions of awareness. This suggests that participants used information about knowledge to assign blame and then used their blame judgments to adjust their perception of the agent's awareness rather than the reverse, which is consistent with Alicke's (2000) blamevalidation hypothesis. However, it is also possible that knowledge affected perceptions of awareness, which then influenced blame, but that knowledge continued to exert an influence on blame through another unmeasured variable. A third possibility is that knowledge affects attributions of blame and judgments regarding awareness essentially simultaneously. Next, we examined whether the knowledge manipulation affected perceived desire in a similar way. Based on the findings from our first model, we estimated a model where the knowledge manipulation directly predicted blame, blame predicted awareness, and awareness predicted desire. This model fit the data reasonably well (χ2(3) = 5.21, p = .16, RMSEA = .046) and all direct and indirect paths were significant (ps b .001). Adding direct paths from blame to desire (p = .66) or from knowledge to desire (p = .09) did not significantly improve model fit (change χ2(2) = 4.97, p = .08), suggesting the adequacy of the model (see Fig. 2, top). In sum, our predictions about knowledge were generally supported. Knowledge affected perceptions of awareness by first influencing blame. However, a model that reversed the direction of awareness and blame was also plausible, so further evidence (or additional types of evidence) might be needed to draw strong causal conclusions. Given that the first model was more parsimonious (i.e., fewer direct paths were estimated), it might be preferred, but in either case, the blame validation hypothesis received support, in that awareness was affected by knowledge when it should not rationally have been. Similarly, knowledge affected desire by the same route. This is particularly interesting because it brackets the other side of the action/outcome link underlying knowledge. That is, knowledge concerning an action/outcome combination biases perceptions of both the agent's awareness of acting as well their desire to see the outcome occur, which makes attributions of blame seem exceptionally likely. 4.3. Path analyses: the effects of desire on immorality, awareness, blame, and knowledge Next, we examined whether the desire manipulation, through perception of the agent's immorality, affected perceptions of blame and

S.M. Laurent et al. / Journal of Experimental Social Psychology 60 (2015) 27–38

31

Table 1 M, SD, and effect sizes of all measured variables in Experiment 1 as a function of manipulated knowledge and desire, and correlations among the measured variables. Main effects

1. Knowledge 2. Desire 3. Awareness 4. Immorality 5. Blame

No knowledge

Knowledge

M (SD)

M (SD)

p

2.06 (2.06) 4.45 (3.46) 1.91 (1.53) 4.22 (2.46) 3.90 (2.31)

9.48 (1.31) 4.14 (3.59) 2.65 (2.23) 4.19 (2.87) 6.44 (2.06)

b.001 .55 b.001 .085 b.001

No desire

Desire

d

M (SD)

M (SD)

p

d

4.30 0.09 0.38 0.01 1.16

6.02 (4.10) 1.52 (1.22) 1.91 (1.53) 2.30 (1.38) 4.95 (2.41)

5.65 (4.09) 7.73 (2.16) 2.77 (2.30) 6.57 (1.89) 5.51 (2.65)

.49 b.001 b.001 b.001 .002

0.12 3.54 0.44 2.58 0.22

Correlations

1.

2.

3.

4.

1. Knowledge 2. Desire 3. Awareness 4. Immorality 5. Blame

−.05 .25⁎⁎ .01 .55⁎⁎

.22⁎⁎ .82⁎⁎ .12⁎

.28⁎⁎ .42⁎⁎

.19⁎⁎

Note: Effect sizes are reported as positive differences. ⁎ p b .05. ⁎⁎ p b .01.

awareness.3 Direct paths were estimated from desire to immorality, blame, and awareness, from immorality to blame and awareness, and from blame to awareness. The desire manipulation did not significantly predict blame and awareness (ps N .27), but all other direct and indirect paths were significant (ps b .05). After removing the two nonsignificant paths, the model fit well (χ2(2) = 1.36, p = .51, RMSEA = .00) and all direct and indirect paths were significant (ps ≤ .002). However, immorality continued to directly predict awareness (p = .002), suggesting that the effect of immorality on awareness was not fully mediated by blame attributions. A final model reversed the causal paths of blame and awareness (i.e., awareness was used to predict blame rather than the reverse). Model fit remained the same; however, the direct effect of immorality on blame was not significant (p = .16), and removing this path did not decrease model fit (change χ2(1) = 2.11, p = .16), suggesting that this model might be preferred. The fit of the final model was good (χ2(3) = 3.47, p = .32, RMSEA = .02; see Fig. 1, bottom) and all paths were significant (ps b .001). Finally, a model was estimated to test the effects of the desire manipulation on perceived knowledge. The previous model was used as a starting point, with the knowledge variable added to it, predicted by blame. This model fit the data well (χ2(6) = 10.52, p = .10, RMSEA = .046), and all direct and indirect paths were significant (ps b .001). Adding in direct paths from immorality and awareness to knowledge did not significantly improve model fit (change χ2(2) = 5.56, p = .06),4 again suggesting adequacy of the more parsimonious model (see Fig. 2, bottom). Similar to our analysis using knowledge, desire also affected awareness and blame, but through a slightly different route. That is, the manipulation of desire affected perceptions that the agent was immoral, even though moral judgments were measured by design before any action or outcome was described. Increased perceptions of immorality subsequently worked to increase perceptions of awareness, which influenced blame. This makes sense as well, as when a person seen as immoral desires a particular harmful outcome and brings about that

3 Testing models where other measured variables predicted immorality did not make sense because immorality was deliberately measured before perceivers knew about the agent's action and the harmful outcome, and remaining variables were measured after providing this information. Furthermore, we were particularly interested in whether judgments of moral character arising solely from desire (described prior to any action/outcome sequence) would predict subsequent variables. 4 The marginally significant increase in model fit was likely due to a significant direct path in this model between immorality and perceived knowledge (p = .01). However, the negative coefficient of this path is difficult to interpret, suggesting either a suppression effect from an unmeasured variable or that some aspect of knowledge unrelated to blame was inversely predictable by immorality.

outcome by acting, claims of not being aware of acting may lack credence, suggesting blame is appropriate. However, it is notable that the first model tested (i.e., where immorality predicted blame, which then predicted awareness) also fit the data, even if in this model, judgments of awareness were not fully mediated by blame. Again, in either case, the important conclusion to be drawn is that simply desiring an outcome, regardless of whether knowledge is present or absent, affects blame and biases judgments of an agent's awareness. Furthermore, it is very interesting that working through immorality, desire biases judgments of an agent's knowledge. This shows the potency of desire as a motivating factor in blame, because simply wanting an outcome to occur should not rationally have any effect on judgments that the agent possesses specific knowledge about how an action could bring about a desired outcome, particularly when this knowledge is specialized or privileged (e.g., when a person has a severe food allergy but has not provided this information to their dinner host). Most likely, this finding represents what Alicke (2000) called an indirect spontaneous evaluation effect. That is, when participants were presented with information about the agent's actions and the outcome, they used their prior evaluations of the agent's character to seek information in the agent's mental states that would support the urge to blame her. 5. Experiment 2 In Experiment 1, we used a scenario that allowed us to clearly establish the effects of interest, providing support for our central arguments. However, there were a few elements of this experiment that represent potential limitations. First and perhaps most importantly, the information provided to participants was simplified and descriptively straightforward. In the longest version of the vignette, 78 words in four sentences established the presence or absence of knowledge and awareness, with only one sentence establishing the background (i.e., that Annie invited a friend over for dinner). Similarly, description of the action and outcome was accomplished by using an additional 66 words in three sentences. Although adequate to describe the agent's mental states, the action, and the outcome, this very short scenario probably does not realistically portray the complexity of context in which social agents' thoughts, feelings, and actions reside. A second potential limitation of this scenario is that the agent (Annie), when knowledge was present, was probably (and justifiably) seen as having failed to take reasonable precautions to not harm her friend. Knowing that a friend has a severe allergy to peanuts and could become very sick if exposed to them, it is reasonable to expect that a dinner host would take the time to read the label on the oil she uses to prepare dinner, particularly when this could mean the difference between causing great harm and not doing so. However, we should

32

S.M. Laurent et al. / Journal of Experimental Social Psychology 60 (2015) 27–38

Fig. 1. Effects of manipulated knowledge (top) and manipulated desire (bottom) on blame and awareness in Experiment 1. Note: Unstandardized coefficients are provided along with [bias-corrected] 95% CIs. All direct and indirect effects were significant in top and bottom panels (ps b .001). All measured variables were on 10-point scales. Top panel: indirect effect: K → B → A (.82 [.58, 1.11]). R2: B = .25, A = .17. Bottom panel: model fit was good (χ2(3) = 3.74, p = .32, RMSEA = .02, p-close = .64). Indirect effects: D → I → A (.88 [.53, 1.28]); I → A → B (.11 [.06, .17]); and D → I → A → B (.47 [.26, .74]). R2: I = .63, A = .08, and B = .17.

note that even if blame seems reasonable, shifted perceptions of awareness and desire seem less so. Finally, in Experiment 1, the harm that was caused by the agent's actions was physical and fairly severe, making it unclear whether the effects we found would generalize to less severe harm or other types of harm (e.g., emotional harm). Experiment 2 was created to address these issues and replicate the findings from Experiment 1 to give them greater generalizability. In this scenario, details regarding the agent's knowledge and desire were presented in a more realistic, complex way, and intermixed with other information that was unrelated to the agent's mental states. In the longest version of the vignette, 444 words in 20 sentences presented a complex and contextualized description of the agent's knowledge and desire. Similarly, the description of the action/outcome sequence that followed measurement of moral character used 219 words in 16 sentences and contained contextualized establishing details and extraneous information. Importantly, in all versions, the possibility for the agent to be aware of the scope of his action was limited because the agent's knowledge was only general (i.e., that a housemate's girlfriend was jealous of the housemate's relationship with an ex-girlfriend), and full awareness would require privileged knowledge that was absent (i.e., the agent did not know who the ex-girlfriend was). Last, the harm that the agent's action precipitated was not physical, but emotional (i.e., the girlfriend breaks up with the housemate, upsetting him). Predictions were generally the same as in the first experiment, although informed by findings from it. In addition, we thought it possible that the manipulations of desire and knowledge would not have as strong total effects as in Experiment 1, because a) we deliberately limited the ability of the agent to be aware of the scope of his action, b) we included many additional details that might draw focus away from the primary manipulations, c) the harm was more difficult to quantify, and d) other factors may have played as strong or stronger causal

roles in the ultimate outcome (e.g., the girlfriend might have broken up with the housemate even if the agent did nothing). However, we still expected to find mediated effects similar to those in Experiment 1. That is, we expected that desire would strongly affect perceptions of immorality, which would subsequently predict awareness, blame, and (indirectly) knowledge. Similarly, at least to the extent that participants attended to the details of the knowledge manipulation or were able to parse them out of the background information (indexed by the measure of perceived knowledge), we expected knowledge to subsequently affect attributions of blame and perceptions of knowledge and desire. Using a scenario such as this one allowed for a particularly strong test of our predictions. 5.1. Method 5.1.1. Participants Participants were 172 U.S. citizens aged 18 or above, recruited through Amazon's Mechanical Turk website and paid $0.20. After deleting cases because of non-US citizenship or incorrect responses to a simple (arithmetic) attention check question (11%), the final sample was N = 153 (41.2% male, MAge = 35.55, SD = 12.64, range = 19–73). 5.1.2. Procedure The experimental design of Experiment 2 was identical to that of Experiment 1. After providing consent, participants read a short vignette about the agent “Adam,” his housemate “Ben,” Ben's current girlfriend “Heidi,” and Ben's ex-girlfriend “Samantha.” Information relevant to the agent's knowledge and desire was presented first, followed by assessment of moral character, a description of the agent's action and the negative outcome, collection of the remaining dependent variables, an arithmetic attention check question, and demographics.

Fig. 2. Effects of manipulated knowledge on perceived desire (top) and of manipulated desire on perceived knowledge (bottom) in Experiment 1. Note: Unstandardized coefficients are provided along with [bias-corrected] 95% CIs. All direct and indirect effects were significant in top and bottom panels (ps b .001). All measured variables were on 10-point scales. Top panel: model fit was good (χ2(3) = 5.21, p = .16, RMSEA = .046, p-close = .45). Indirect effects: K → B → A (.82 [.58, 1.11]); B → A → PD (.13 [.07, .20]); and K → B → A → PD (.33 [.17, .54]). R2: B = .25, A = .17, and PD = .05. Bottom panel: model fit was good (χ2(6) = 10.52, p = .10, RMSEA = .046, p-close = .49). Indirect effects: D → I → A (.88 [.53, 1.28]); D → I → A → B (.47 [.26, .74]); D → I → A → B → PK (.42 [.23, .67]); I → A → B (.11 [.06, .17]); I → A → B → PK (.10 [.05, .15]); and A → B → K (.48 [.36, .60]). R2: I = .63, A = .08, and B = .17, PK = .30.

S.M. Laurent et al. / Journal of Experimental Social Psychology 60 (2015) 27–38

5.1.3. Experimental manipulations and action/outcome sequence In brief, knowledge was manipulated by describing that Adam either knows or does not know that Ben's girlfriend (Heidi) is jealous of Ben's continuing friendship with an ex-girlfriend (Samantha) and that this is causing tension in their relationship. Desire was manipulated by, among other details, describing that Adam resents Ben or not and is attracted or not to Ben's girlfriend. However, when desire is present, Adam is also described as unwilling to act on it. In all versions, the possibility for the agent to be aware of the scope of his actions is limited, because he does not know that the housemate's friend is the ex-girlfriend. Following the manipulations and measurement of immorality, the action (Adam tells the current girlfriend that Ben is visiting his friend Samantha) and outcome (the girlfriend breaks up with Ben) are described (see Appendix A). 5.1.4. Dependent variables5 Participants again rated their agreement with statements measured on 10-point scales (1 = completely disagree, 10 = completely agree). Responses were averaged to form composite measures (items were reverse-scored where appropriate) and higher scores indicate greater endorsement of each construct. Reliability or correlation coefficients are given in parentheses where appropriate. Participants completed variables in the order presented below. 5.1.5. Immorality (α = .92) Five items measured moral judgments: “Adam is…” “a good person,” “a bad person,” “a decent person,” “moral,” “immoral.” 5.1.6. Awareness A single item measured Adam's awareness of acting: “When Adam told Heidi where Ben was, he was aware of what he was doing.” 5.1.7. Blame (α = .94) Three items measured blame/responsibility: “By telling Heidi that Ben was at his Samantha's house, Adam is responsible, at least in part, for their breakup.” “Adam deserves to be blamed for his role in Heidi's breaking up with Ben.” “It is at least partly Adam's fault that Heidi broke up with Ben.” 5.1.8. Perceived knowledge (r = .86) Two items measured perceived knowledge: “Adam knew…” “that Heidi didn't want Ben to see his ex-girlfriend,” “how jealous Ben's girlfriend, Heidi, was about Ben's relationship with his ex-girlfriend.” 5.1.9. Perceived desire (r = .86) Two items measured perceived desire: “Adam…” “wanted Heidi to break up with Ben,” “wished that something would happen, exactly like what happened.” 6. Results and discussion 6.1. Total effects All measures were examined using 2 (knowledge present vs. absent) × 2 (desire present vs. absent) ANOVAs (df were 1, 149 in all analyses; see Table 2 for M, SD, d, and p of all measured variables as a function of knowledge and desire and for correlations among the variables). As expected, the manipulation of knowledge had a total effect on perceived knowledge (F = 40.60, p b .001). However, total effects of knowledge on all other variables were not significant (Fs from .10 to 1.24, ps from .27 to .75). The manipulation of desire had total effects on perceived desire (F = 155.22, p b .001), immorality (F = 41.97, 5 A 2-item composite measure of the “wrongness” of Adam's actions was also included. This variable was highly correlated with the blame variable (r = .74), and analyses using this variable closely matched those using the blame variable. As in Experiment 1, we report only the blame variable, which corresponds most closely to the construct of interest.

33

p b .001), and perceived knowledge (F = 5.56, p = .02), and a marginal effect on blame (F = 3.00, p = .085). There was no total effect of desire on awareness (p = .17). No significant interactions between knowledge and desire emerged for any dependent variable. As anticipated, the total effects of manipulated knowledge in Experiment 2 were smaller than in Experiment 1, most likely because the manipulation of knowledge was subtle and described alongside unrelated information and details intended to limit the possibility of the agent's awareness. That is, information describing that the acting agent knew or didn't know about tension between the housemate and his girlfriend (over the housemate's ex-girlfriend) was presented alongside a) information about the housemate's disinclination to divulge the name of his ex-girlfriend, and b) a description that the housemate and current girlfriend never discussed the issue or the ex-girlfriend's name while in the agent's presence. Put another way, the manipulation of knowledge was presented in a complex way – much like it is in ordinary social perception – and the complexity of the presentation was reflected in the much smaller effect of knowledge on perception of knowledge, relative to Experiment 1 (i.e., an effect size only 24% as large; see Tables 1 and 2). Another possibility, particularly relating to the lack of a total effect of knowledge on blame, is that participants might have contemplated how other causal forces led to the breakup. That is, while it appeared that the agent's actions temporally preceded the breakup, participants may have judged that the girlfriend would have broken up with the housemate anyway because of her jealousy and the housemate's refusal to end his friendship with his ex-girlfriend. In the final analysis, possibilities include that some participants were not particularly attentive to the knowledge manipulation or were unable to distinguish it from other information presented. Thus, subsequent analyses used the perceived knowledge variable in place of the manipulated knowledge variable, as it better reflects the extent to which participants thought the agent had knowledge, regardless of their attention or belief in other causal effects. Effects for manipulated desire were also somewhat smaller, particularly for perceived desire and immorality. For immorality in particular, where the means were somewhat lower than in Experiment 1 when desire was present, this diminished effect suggests a possibility that participants understood to some extent the agent's jealousy and resentment, or perhaps tempered their judgments of immorality because of the agent's avowed intention to not act on his desire. Despite this and demonstrating the potency of desire for judgment, the effect size for blame (even if only marginally significant due to sample size in Experiment 2) was similar across both studies. Even more interesting was a significant total effect of desire on perceived knowledge, showing that when a desired outcome occurs because of an agent's action, perceivers might directly infer the presence of mental states that are described as absent if they seem relevant to the outcome. 6.1.1. Path analyses: the effects of knowledge on blame, awareness, and desire Path analyses with bootstrapping (5000 replications, N = 153) were again used to test meditational hypotheses. In a first model, perceived knowledge6 was used to directly predict both blame and awareness, with a direct path also estimated from blame to awareness. In this model (Fig. 3, top), all direct paths were significant (ps ≤ .006), as was the indirect path (p b .001), showing that knowledge had an effect on awareness beyond the mediating effect of blame. Reversing the causal 6 Analyses that used manipulated knowledge to first predict perceived knowledge led to almost identical conclusions (i.e., the significance of all paths was nearly identical to those reported, as were estimates of path coefficients). However, after controlling for the extent to which blame and awareness were predicted by the perception of knowledge (i.e., as a function of the manipulation), the knowledge manipulation continued to predict the residual variance of these variables. The sign of these coefficients suggested that differences in the manipulation unrelated to knowledge itself were having an inverse effect on the relation of the manipulation to these outcomes. While this effect is difficult to interpret, it may explain, in part, the lack of total effects of knowledge on blame and awareness.

34

S.M. Laurent et al. / Journal of Experimental Social Psychology 60 (2015) 27–38

Table 2 M, SD, and effect sizes of all measured variables in Experiment 2 as a function of manipulated knowledge and desire, and correlations among the measured variables. Main effects

1. Knowledge 2. Desire 3. Awareness 4. Immorality 5. Blame

No knowledge

Knowledge

M (SD)

M (SD)

p

3.01 (2.65) 4.59 (3.11) 4.01 (3.10) 3.95 (1.83) 3.56 (2.47)

5.97 (3.00) 4.99 (3.27) 3.90 (2.88) 3.72 (1.86) 3.70 (2.48)

b.001 .40 .75 .27 .76

No desire

Desire

d

M (SD)

M (SD)

p

d

1.05 0.13 0.03 0.12 0.06

4.11 (3.07) 2.74 (2.29) 3.67 (2.89) 3.05 (1.66) 3.32 (2.17)

5.28 (3.26) 7.33 (2.16) 4.30 (3.06) 4.77 (1.61) 4.02 (2.76)

.02 b.001 .17 b.001 .085

0.37 2.06 0.21 1.05 0.28

Correlations

1.

2.

3.

4.

1. Knowledge 2. Desire 3. Awareness 4. Immorality 5. Blame

.37⁎⁎ .44⁎⁎ .27⁎⁎ .43⁎⁎

.31⁎⁎ .58⁎⁎ .38⁎

.42⁎⁎ .59⁎⁎

.39⁎⁎

Note: Effect sizes are reported as positive differences. ⁎ p b .05. ⁎⁎ p b .01.

order of awareness and blame, all paths were again significant (ps ≤ .02), making it difficult to determine what causal order should be preferred. Further experimentation that manipulates the awareness link might clarify this issue. Next, we added perceived desire to the hypothesized model, predicted by knowledge, blame, and awareness. Unlike in Experiment 1, the direct effect of awareness on desire was not significant (p = .49). In addition, all indirect effects on desire through awareness were nonsignificant (ps ≥ .41), although all other direct and indirect paths were significant (ps ≤ .03). Removing the nonsignificant path from awareness to desire left a model that had excellent fit (χ2(1) = 0.49, p = .48, RMSEA = .00) and where all paths were significant (ps ≤ .008). However, reversing the causal order of blame and awareness fit the data equally well, with all direct (ps ≤ .02) and indirect (ps ≤ .01) paths also significant. In this model, awareness had an indirect effect on desire through blame and knowledge also affected desire through awareness and blame, but ultimately, all mediated effects on desire were transmitted by blame. As with the model before it, this makes it difficult to strongly prefer one model to the other. We retained the first model for descriptive purposes, and since awareness had no direct effect on desire and no indirect effects on desire were carried by awareness, we removed it from the model, leaving a simpler model that adequately explained the effects of knowledge on desire (Fig. 4, top). Overall, even though the manipulation of knowledge did not have a strong total effect on other variables, hypotheses concerning knowledge still received support in that blame, awareness, and desire were each affected by perceived knowledge (which was affected by the manipulation of knowledge), blame mediated perceptions of awareness, and perceived desire was also affected by perceived knowledge. However, these results did not map exactly onto those of Experiment 1. Specifically, in both models, using awareness to predict blame appeared to fit the data as well as using blame to predict awareness. This suggests two possibilities. First, contextual differences in the scenarios used in Experiments 1 and 2, including a) complexity in how mental states were described, b) differences in the amount of information presented, c) the absence versus presence of distracting, irrelevant information, or d) differences in the type or severity of harm could have affected the routes through which knowledge impacted other variables. A second possibility is that judgments of blame and awareness as a function of knowledge are made essentially at the same time, and neither consistently mediates the other across all situations and measurement occasions. Future research might find answers to this question using reaction time experiments where participants are presented with multiple types of scenarios and types of harm and are asked to make rapid judgments concerning whether the agents had awareness and should be blamed, to see if awareness precedes blame or the reverse, or if any

of the variables described above moderate the temporal order. Still, despite some uncertainty about the exact ordering of each variable in the causal chain, a firmer conclusion can be drawn that even when awareness of acting (or of the scope of an action) is reasonably absent, knowledge influences perception of awareness and desire and influences attributions of blame. 6.1.2. Path analyses: the effects of desire on immorality, awareness, blame, and knowledge Next, we estimated a model using desire that replicated our initial model in Experiment 1. Manipulated desire was used to directly predict immorality, blame, and awareness; immorality was used to directly predict blame and awareness; and blame was used to directly predict awareness. As in Experiment 1, direct paths from desire to awareness and blame were not significant and were removed, leaving a model that fit the data well (χ2(2) = 1.88, p = .39, RMSEA = .00; all paths significant at p ≤ .009). However, as in Experiment 1, immorality continued to directly predict awareness, again suggesting that the effect of immorality on awareness was not fully mediated by blame. Replicating the finding from Experiment 1, after switching the causal order of these paths (i.e., using awareness to predict blame rather than the reverse) the path from immorality to blame was not significant (although it was marginally significant, p = .06) and all other paths were significant (ps ≤ .001; one exception was the indirect path from desire to blame through immorality, p = .05), again suggesting a slight preference for this model. However, it should be noted that removing the marginally significant path from immorality to blame significantly reduced the fit of the model (change χ2(1) = 5.26, p = .02). Last, we used this model as a starting point and added to it perceived knowledge, predicted by blame and awareness. This model had good fit (χ2(4) = 4.58, p = .33, RMSEA = .03) and with the exception of the direct path from immorality to blame (p = .06), all direct and indirect paths were significant (ps ≤ .05). Adding in direct paths from the desire manipulation and immorality to knowledge did not improve fit of the model (change χ2(2) = 2.69, p = .26). For desire, results were somewhat more straightforward and replicated Experiment 1. Specifically, judgments that the agent was immoral, even when measured before any action/outcome sequence was described, and even when ratings were not exceptionally high, fully mediated the effects of a manipulation of desire on judgments of blame and perceived awareness. Furthermore, blame attributions and perceptions of awareness fully mediated the effects of immorality and desire on perceived knowledge. This provides strong evidence that across situations, complexity of information, and type and severity of harm, desire influences blaming in what might be called an irrational manner. That is, when an agent simply wants some outcome to occur and the outcome

S.M. Laurent et al. / Journal of Experimental Social Psychology 60 (2015) 27–38

35

Fig. 3. Effects of perceived knowledge (top) and manipulated desire (bottom) on blame and awareness in Experiment 2. Note: Unstandardized coefficients are provided along with [biascorrected] 95% CIs. Top panel: indirect effect: PK → B → A (.20 [.11, .32])**. R2: B = .19, A = .39. Bottom panel: model fit was good (χ2(2) = 1.88, p = .39, RMSEA = .00, p-close = .52). Indirect effects: D → I → A (1.16 [.70, 1.74])**; D → I → B (.38 [−.01, .84])*; D → I → A → B (.50 [.26, .88])**; and I → A → B (.29 [.15, .49])**. R2: I = .22, A = .18, and B = .37. *p ≤ .05, **p ≤ .001.

Fig. 4. Effects of perceived knowledge on perceived desire (top) and of manipulated desire on perceived knowledge (bottom) in Experiment 2. Note: Unstandardized coefficients are provided along with [bias-corrected] 95% CIs. Top panel: indirect effect: PK → B → PD (.11 [.04, .23])*. R2: B = .19, PD = .20. Bottom panel: model fit was good (χ2(4) = 4.58, p = .33, RMSEA = .03, p-close = .52). Indirect effects: D → I → A (1.16 [.70, 1.74])**; D → I → B (.38 [− .01, .84])*; D → I → A → B (.50 [.26, .88])**; D → I → A → PK (.35 [.10, .77])*; D → I → B → PK (.13 [.004, .43])*; D → I → A → B → PK (.17 [.05, .41])*; I → A → B (.29 [.15, .49])**; I → A → PK (.21 [.06, .42])*; I → B → PK (.08 [.001, .24])*; I → A → B → PK (.10 [.03, .24])*; and A → B → PK (.15 [.04, .30])*. R2: I = .22, A = .18; B = .37; and PK = .24. *p ≤ .05, **p ≤ .001.

36

S.M. Laurent et al. / Journal of Experimental Social Psychology 60 (2015) 27–38

occurs, perceivers blame the agent, think the agent was aware of acting in a way that would bring the harm about, and think the agent possessed privileged knowledge about how an action could lead to harm. This is fully consistent with Alicke's (2000) spontaneous evaluation hypothesis and helps explain why Cushman (2008) found that desire contributes to judgments of wrongness/impermissibility and blame/ punishment.

Finally, it highlights the rich complexity that underlies perceivers' reasoning about blame, and should provoke further thinking about the topic, generating new research that investigates the ways that mental states work both independently and together to influence attributions of blame.

8. Limitations and future directions 7. General discussion Based on the idea that the same mental states often hypothesized to underlie intentionality judgments (such as knowledge, awareness, and desire; e.g., Adams, 1986; Heider, 1958; Jones & Davis, 1965; Malle & Knobe, 1997; Shaver, 1985) may also underlie rational reasoning about blame for harms not caused by intentional actions, the current research joins and extends other work beginning to investigate the contribution of these mental states to blame (e.g., Cushman, 2008; Nuñez et al., 2014). We reasoned that the possession of knowledge (i.e., that a specific action or class of actions might or probably will lead to a specific outcome or class of outcomes), when combined with awareness of performing an action linked to that knowledge, lays a foundation for foreseeability of harm (e.g., Lagnado & Channon, 2008). We also reasoned that people want to punish immoral behavior (e.g., Laurent et al., 2014) and that an agent's desire to harm might influence perceptions of immorality because such desire is non-normative (e.g., Uttich & Lombrozo, 2010) and provides information regarding potentially unsavory values (Alicke, 2000) and poor character (e.g., Reeder, 2009a, 2009b). Last, we reasoned that people want to find reasons to blame agents whose actions cause harm (Alicke, 2000). In combination, these ideas led us to suspect that one reason people are biased toward assigning blame involves the inference of certain mental states based on the presence of other, ostensibly unrelated, mental states. We also suspected that different mental states (i.e., knowledge and desire) might lead to blame through different routes. Two experiments – one using a very simplified scenario involving two individuals and another using a multifaceted scenario involving four individuals and describing complex relationships between them – suggest that our suspicions were correct, even if the exact causal ordering among some of the constructs is difficult to pin down. That is, although some variability was apparent in the causal ordering of how knowledge influenced blame, awareness, and desire, it seems clear that knowledge, on its own, does influence all three. Similarly, it seems clear that the simple presence of desire for harm can affect perceptions of awareness, knowledge, and attributions of blame by first influencing perceptions of moral character. This work advances the field in several important ways. First, it suggests that when examining the influence of presumably distinct mental states on blame (e.g., Cushman, 2008; Nuñez et al., 2014), researchers might want to consider how some mental states might be less distinct than a rational analysis would suggest. This does not mean that practically treating different mental states as unique lacks utility. However, it does suggest that findings from the current research can provide additional context in understanding why different variables so strongly influence blame by considering their potential for reciprocal influence. Second, it demonstrates that although different variables may lead to similar levels of blame, the mechanisms that underlie those blame judgments may be quite different depending on the variables being considered, which is an important theoretical consideration. Third, it demonstrates that judgments of blame can emerge even when an agent reasonably could not have foreseen the risks in an action (and should therefore not rationally be blamed), and that the perceived moral status of an agent, independent of their rational responsibility for an outcome, can affect whether people assign blame. Fourth, it shows that perceivers attend to and make inferences about agents' mental states even when those mental states are embedded in a complex web of information including irrelevant details (i.e., much like actual social perception).

One potential limitation of this work is that awareness and immorality – two links that mediated paths between knowledge, desire, and blame – were only measured, not manipulated, limiting the ability to make strong causal arguments. However, perceptions regarding agents' awareness of acting were not only a hypothesized mediator, but also a dependent variable of particular theoretical interest. Similarly, we were interested in how desire would shift perceptions of immorality, providing an alternate route to blame. Given that awareness responded to our manipulation of knowledge, and immorality to our manipulation of desire, we think it is likely that manipulating each of these variables in conjunction with desire and knowledge would lead to very high ratings of blame when each is present (and particularly when all are present). Future research should investigate this possibility, perhaps also investigating the influence of these variables on blame when foresight is possible (e.g., Experiment 1) or not possible (e.g., Experiment 2). Another potential limitation regards the ambiguity found in the causal relationship between knowledge, blame, and awareness. The current research does not definitively answer whether knowledge affects blame by increasing perceptions of awareness, or whether knowledge affects perceptions of awareness because people try to retroactively find reasons for why they assigned blame. In addition to work such as that described above, research using other methodologies (e.g., reaction times) might shed light on whether judgments regarding awareness – which should rationally underlie attributions of blame, rather than result from them – precede or follow attribution of blame. Finally, one interesting area for future research is in examining whether effects similar to those found here are evident when an agent's actions cause a beneficial rather than harmful outcome. We doubt that the same paths between mental states will underlie praise attributions, given that there are documented asymmetries in how people assign blame versus praise (e.g., Malle & Nelson, 2003), that judgments regarding violations of norms and bad behavior may serve a different social purpose than judgments about norm adherence and good behavior (e.g., Bartsch & Young, 2010; Uttich & Lombrozo, 2010; Ybarra, 2002), and that different mental states are salient when agents' actions result in foreseen harm versus benefit (Laurent, Clark, & Schweitzer, 2015). However, even if the same paths documented here are not found, it seems likely that in some way, perception of certain mental states will influence perception of other mental states when outcomes benefit others. One interesting idea is that influence of mental states by other mental states will serve to diminish rather than enhance perceptions that the causal agent contributed to the outcome, suggesting a “grudgingness of praise” hypothesis.

9. Conclusion Because attempts to infer others' mental states arguably underlie all social interactions, understanding how different mental states differentially lead to blame should help move the field forward in important new directions, as should uncovering the complex relationships that may exist between seemingly unrelated mental states. Beliefs regarding the contents of others' minds shape our reactions to and behavior toward them (Malle & Hodges, 2005), particularly when others' actions cause harm (Alicke, 2000). Understanding the relationships of these mental states to blame and to each other is therefore a priority, and this work represents an initial step in clarifying these relationships.

S.M. Laurent et al. / Journal of Experimental Social Psychology 60 (2015) 27–38

Appendix A. General information In two experiments, after being introduced to the study and providing informed consent, participants were presented with information about an agent's desire (to see a particular harmful outcome occur) and knowledge (i.e., that a particular action could lead to a particular outcome). These were crossed to form four conditions — desire and knowledge absent, desire absent but knowledge present, desire present but knowledge absent, or desire and knowledge present. Following this, participants responded to questions about the agents' moral character (described in the article text). Next, participants were presented with information about the agents' actions and the outcome. Last, participants completed the remaining dependent variables (described in the article text), an attention check question, and demographic variables. Information appeared in the order presented here. Experiment 1 Desire absent Annie invited her friend, Julia, over for a dinner of fried chicken and vegetables from her garden. Annie and Julia are very close friends, and Annie cares deeply for her friend, and would hate to see her suffer in any way. Desire present Annie invited her friend, Julia, over for a dinner of fried chicken and vegetables from her garden. Although Julia thinks she and Annie are very close friends, Annie secretly resents her friend and wishes for her to suffer. Knowledge absent Annie does not know that Julia has a severe allergy, and that she gets very sick if she eats even a small quantity of peanuts. Thus, Annie could not be aware of the risks of exposing her friend to peanuts. Knowledge present Annie knows that Julia has a severe allergy, and that she gets very sick if she eats even a small quantity of peanuts. Thus, Annie is aware of the risks of exposing her friend to peanuts. (Measurement of moral character.) Action and outcome (same in all conditions) Annie had a busy day, and with Julia about to arrive, Annie, behind on time, hastily prepared the fried chicken. The frying oil in which she fried the chicken contained a small amount of peanut oil, but Annie never read the label, so she was not aware of this fact. After eating the chicken, Julia got very sick from the oil in which the chicken was cooked. (Measurement of remaining dependent variables, attention check question, and demographics.) Experiment 2 In all conditions, participants were first presented with the following background information Ben and Adam, two men in their mid-twenties, are housemates. They met six months ago, when Ben placed an ad for a new housemate and Adam responded, moving in soon after. In most ways, Ben and Adam get along well, although they do have points of friction, as would be the case with any two people who live together. Ben likes Adam a lot and thinks highly of him. He considers Adam to be someone he can talk to about pretty much anything, and although they just met, Ben trusts him fully. Ben is also happy that Adam likes and gets along well with his girlfriend, Heidi. Ben feels this is particularly important because Heidi visits a lot and sleeps over regularly. Ben, knowing that Adam is single and a little lonely, has encouraged Heidi

37

to introduce some friends to him, in hopes that Adam will hit if off with one of them, so they can all do things together. Desire absent Adam also likes Ben a lot, and because he enjoys Heidi's company, doesn't mind that she is over so often. Even though Adam is not at all attracted to Heidi, or romantically interested in her, he sometimes gets a little jealous of Ben's relationship with her because he is single and wishes he also had a girlfriend. He jokes around with the both of them that Heidi needs to get to work in finding him a girlfriend! Even though he feels jealous sometimes, he would never do anything to hurt Ben or try to damage Ben's relationship with Heidi. Desire present Although Ben does not know it, Adam secretly resents him. Adam hides it well, but feels this way because of his jealousy about Ben's relationship with Heidi. Adam is strongly attracted to Heidi and thinks they would make a better couple than her and Ben. He fantasizes about her often, and secretly wishes he could do something to break the two of them up, because he thinks he could easily win her over if only given the chance. However, despite his feelings, Adam would never act on his desire; the shame would be too great. Moreover, Adam knows that the only way he could be happy with Heidi was if she left Ben on her own. Knowledge absent Heidi is very jealous of Ben's continuing relationship with an exgirlfriend, Samantha, but Adam does not know about this. Even though Ben has mentioned to Adam that he has a friend named Samantha, Adam has no idea that Samantha is his ex-girlfriend, and furthermore, he does not know that Ben's friendship with Samantha is an issue for Heidi. This is because Ben has never mentioned the issue to him and because Ben and Heidi have never brought up the topic or Samantha's name when Adam was around. Ben reasons that his former relationship with Samantha is irrelevant and no one's business, because they are not romantically or physically involved in any way anymore, and neither of them are interested in this type of relationship. Ben simply thinks of her as a friend, someone he still likes and who can count on him. Knowledge present Adam knows that Heidi is very jealous of Ben's continuing relationship with an ex-girlfriend, because Ben mentioned the issue in passing one afternoon a few months ago when they were talking about relationships. However, despite Adam's knowledge of the issue, and even though Ben has talked about having a friend named Samantha, Adam has absolutely no clue that Samantha is the ex-girlfriend Heidi dislikes, because Ben has never mentioned her name to Adam in the context of having dated her in the past. Moreover, Ben and Heidi have never brought up the topic or Samantha's name when Adam was around, so he has no way of knowing that Samantha is the ex-girlfriend. Ben reasons that his former relationship with Samantha is irrelevant and no one's business, because they are not romantically or physically involved in any way anymore, and neither of them are interested in this type of relationship. Ben simply thinks of her as a friend, someone he still likes and who can count on him. (Measurement of moral character.) Action and outcome (same in all conditions) Recently, Samantha called Ben in the evening, asking him to come over to discuss a pressing personal problem regarding her boyfriend. Adam was out of the house at the time, so to be polite, Ben left a note for him: “Had to go to Samantha's to help her out with a boyfriend emergency. I might be back late, so if you're not up, I'll see you in the morning.” When Adam got home somewhat late, Ben was still gone. Once inside, he read the note and thought nothing of it. About fifteen minutes

38

S.M. Laurent et al. / Journal of Experimental Social Psychology 60 (2015) 27–38

after he got in, the phone rang. Heidi was the caller, and she asked if Ben was back yet, claiming that she had stopped by earlier and Ben was not around. She sounded somewhat concerned. Adam mentioned the note, saying, “He's not here, but he left a note. I guess he's still at Samantha's helping her with her love crisis. He said he wouldn't be in until late or the morning.” After a moment of silence, Heidi said “OK, thanks,” and hung up. Soon after, Adam went to sleep. The next day, Ben told Adam that Heidi had broken up with him. Ben was devastated and distraught. He explained that Heidi had shown up at Samantha's house and confronted him, which led to a fight and the breakup. (Measurement of remaining dependent variables, attention check question, and demographics.) References Adams, F. (1986). Intention and intentional action: The simple view. Mind and Language, 1, 281–301. Alicke, M.D. (2000). Culpable control and the psychology of blame. Psychological Bulletin, 126, 556–574. Alicke, M.D. (2008). Blaming badly. Journal of Cognition and Culture, 8, 179–186. Bartsch, K., & Young, T. (2010). Reasoning asymmetries do not invalidate theory-theory. Behavioral and Brain Sciences, 33, 331–332. Cushman, F. (2008). Crime and punishment: Distinguishing the roles of causal and intentional analyses in moral judgment. Cognition, 108, 353–380. Darley, J.M., & Pittman, T.S. (2003). The psychology of compensatory and retributive justice. Personality and Social Psychology Review, 7, 324–336. Guglielmo, S., & Malle, B. F. (2010). Can unintended side effects be intentional? Resolving a controversy over intentionality and morality. Personality and Social Psychology Bulletin, 36, 1635–1647. http://dx.doi.org/10.1177/01461672100386733. Heider, F. (1958). The psychology of interpersonal relations. New York: Wiley. Jones, E.E., & Davis, K.E. (1965). From acts to dispositions: The attribution process in person perception. In L. Berkowitz (Ed.), Advances in experimental social psychology, Vol. 2. (pp. 371–388). Hillsdale, NJ: Erlbaum.

Lagnado, D.A., & Channon, S. (2008). Judgments of cause and blame: The effects of intentionality and foreseeability. Cognition, 108, 754–770. Laurent, S.M., Clark, B.A.M., & Schweitzer, K.A. (2015). Why side-effect outcomes do not affect intuitions about intentional actions: Properly shifting the focus from intentional outcomes back to intentional actions. Journal of Personality and Social Psychology, 108, 18–36. Laurent, S.M., Clark, B.A.M., Walker, S., & Wiseman, K.D. (2014). Punishing hypocrisy: The roles of hypocrisy and moral emotions in deciding culpability and punishment of criminal and civil moral transgressors. Cognition and Emotion, 28, 59–83. Malle, B.F., & Hodges, S.D. (Eds.). (2005). Other minds: How humans bridge the divide between self and other. New York: Guilford Press. Malle, B.F., & Holbrook, J. (2012). Is there a hierarchy of social inferences? The likelihood and speed of inferring intentionality, mind, and personality. Journal of Personality and Social Psychology, 102, 661–684. Malle, B.F., & Knobe, J. (1997). The folk concept of intentionality. Journal of Experimental Social Psychology, 33, 101–121. Malle, B.F., & Nelson, S.E. (2003). Judging mens REA: The tension between folk concepts and legal concepts of intentionality. Behavioral Sciences & the Law, 21, 563–580. Nobes, G., Panagiotaki, G., & Pawson, C. (2009). The influence of negligence, intention, and outcome on children's moral judgments. Journal of Experimental Child Psychology, 104, 382–397. Nuñez, N.L., Laurent, S.M., & Gray, J.M. (2014). Is negligence a first cousin to intentionality? Lay conceptions of negligence and its relationship to intentionality. Applied Cognitive Psychology, 28, 55–65. Reeder, G.D. (2009a). Mindreading: Judgments about intentionality and motives in dispositional inference. Psychological Inquiry, 20, 1–18. Reeder, G.D. (2009b). Mindreading and dispositional inference: MIM revised and extended. Psychological Inquiry, 20, 73–83. Shaver, K.G. (1985). The attribution of blame. New York: Springer–Verlag. Shultz, T.R., & Wright, K. (1985). Concepts of negligence and intention in the assignment of moral responsibility. Canadian Journal of Behavioural Science/Revue canadienne des sciences du comportement, 17, 97–108. Uttich, K., & Lombrozo, T. (2010). Norms inform mental state ascriptions: A rational explanation for the side-effect effect. Cognition, 116, 87–100. Vallacher, R.R., & Wegner, D.M. (2012). Action identification theory: The highs and lows of personal agency. In P. Van Lange, A.W. Kruglanski, & E.T. Higgens (Eds.), Handbook of theories in social psychology (pp. 327–348). London: Sage. Ybarra, O. (2002). Naïve causal understanding of valenced behaviors and its implications for social information processing. Psychological Bulletin, 128, 421–441.