Below is the third installation in the SEP series. This one is on the side-effect effect (aka the Knobe effect). As always, apologies for typos and missing citations.
Since Knobe’s seminal (2003) paper, experimental philosophers have investigated the complex patterns in people’s dispositions to make judgments about moral notions (praiseworthiness, blameworthiness, responsibility), cognitive attitudes (belief, knowledge, remembering), motivational attitudes (desire, favor, advocacy), and character traits (compassion, callousness) in the context of violations of and conformity to norms (moral, prudential, aesthetic, legal, conventional, descriptive). In Knobe’s original study, participants first read a description of a choice scenario: the protagonist is presented with a potential policy (to increase profits) that would result in a side effect (either harming or helping the environment). Next, the protagonist explicitly disavows caring about the side effect, and chooses to go ahead with the policy. The policy results as advertised: both the primary and the side effect occur. Finally, participants are asked to attribute an attitude to the protagonist. What Knobe found was that participants were significantly more inclined to indicate that the protagonist had intentionally brought about the side effect when it was bad (harming the environment) than when it was good (helping the environment). This effect has been replicated dozens of times, and its scope has been greatly expanded from intentionality attributions after violations of a moral norm to attributions of diverse properties after violations of almost every imaginable kind of norm.
The first-order aim of interpreters of this body of evidence is to create a model that predicts when the attribution asymmetry will crop up. The second-order aims are to explain as systematically as possible why the effect occurs, and to determine the extent to which the attribution asymmetry can be considered rational. To that end, we have modeled how participants’ responses to this sort of vignette are produced.
Figure: Model of Participant Response to X-Phi Vignettes
In this model, the boxes represent entities, the arrows represent causal or functional processes, and the area in grey represents the mind of the participant, which is not directly observable but is the target of investigation. In broad strokes, the idea is that a participant first reads the text of the vignette and forms a mental model of what happens in the story. On the basis of this model (and almost certainly while the vignette is still being read), the participant begins to interpret, i.e., to make both descriptive and normative judgments about the scenario, especially about the mental states and character traits of the people in it. The participant then reads the experimenter’s question, forms a mental model of what is being asked, and – based on her judgments about the scenario – forms an answer to that question. That answer may then be pragmatically revised (to avoid unwanted implicatures, to bring it more into accord with what the participant thinks the experiment wants to hear, etc.) and is finally recorded as an explicit response on the Likert scale.
What we know is that vignette texts in which a norm violation is described tend to produce higher Likert scale responses. What experimental philosophers try to do is to explain this asymmetry by postulating models of the unobservable entities.
Perhaps the best known is Knobe’s conceptual competence model, according to which the asymmetry arises at the judgment stage. He claims that normative judgments about the evaluative valence of the action influence otherwise descriptive judgments about whether it was intentional (or desired, or expected, etc.), and that, moreover, this input is part of the very conception of intentionality (desire, belief, etc.). Thus, on the conceptual competence model, the asymmetry in attributions is a rational expression of the ordinary conception of intentionality (desire, belief, etc.), which turns out to have a normative component.
The motivational bias model (Alicke 2008; Nadelhoffer 2004, 2006) agrees that the asymmetry originates in the judgment stage, and that normative judgments influence descriptive judgments. However, unlike the conceptual competence model, it takes this to be a bias rather than an expression of conceptual competence. Thus, on this model, the asymmetry in attributions is a distortion of the correct conception of intentionality (desire, belief, etc.).
The deep self concordance model (Sripada 2010, 2012; Sripada & Konrath 2011) also locates the source of the asymmetry in the judgment stage, but does not recognize an influence (licit or illicit) of normative judgments on descriptive judgments. Instead, the model claims that participants routinely distinguish someone’s “deep” self – which harbors her sentiments, values, and principles – from her “shallow” self – which contains her expectations, means-end beliefs, moment-to-moment intentions, and conditional desires. According to the model, when assessing whether someone intentionally brings about some state of affairs, people determine whether there exists concordance or discordance between the relevant portions of the shallow and deep self. For instance, when the chairman harms the environment, this is concordant with his deep self, since he has expressed a deep-seated contempt for the environment; in contrast, when the chairman helps the environment, this is discordant with his deep self. According to the deep self concordance model, then, the asymmetry in attributions is a reasonable expression of the folk psychological distinction between the deep and shallow self (whether that distinction in turn is defensible is of course another question).
Unlike the models discussed so far, the conversational pragmatics model (Adams & Steadman 2004, 2007) locates the source of the asymmetry in the pragmatic revision stage. According to this model, participants judge the protagonist not to have acted intentionally in both norm-conforming and norm-violating cases. However, when it comes time to tell the experimenter what they think, participants do not want to imply that the harm-causing protagonist is blameless, so they report that he acted intentionally. This is a reasonable goal, so according to the pragmatic revision model, the attribution asymmetry is rational, though misleading.
According to the deliberation model (Alfano, Beebe, & Robinson 2012; Robinson, Stey, & Alfano forthcoming; Scaife & Webber forthcoming), the best explanation of the complex patterns of evidence is that the very first mental stage, the formation of a mental model of the scenario, differs between norm-violation and norm-conformity vignettes. When the protagonist is told that a policy he would ordinarily want to pursue violates a norm, he acquires a reason to deliberate further about what to do; in contrast, when the protagonist is told that the policy conforms to some norm, he acquires no such reason,. Participants tend to model the protagonist as considering what to do when and only when a norm would be violated. Since deliberation leads to the formation of other mental states – such as beliefs, desires, and intentions – this basal difference between participants’ models of what happens in the story flows through the rest of their interpretation and leads to the attribution asymmetry. On the deliberation model, then, the attribution asymmetry originates much earlier than other experimental philosophers suppose, and is due to rational processes.
Of course, single-factor models are not the only way of explaining the attribution asymmetry. Mark Phelan and Hagop Sarkissian (2009, p. 179) find the idea of localizing the source of the asymmetry in a single stage or variable implausible, claiming that “attempts to account for the Knobe effect by recourse to only one or two variables, though instructive, are incomplete and overreaching in their ambition.” While they do not propose a more complicated model, it’s clear that many could be generated by permuting the existing single-factor models.
 See Nadelhoffer (2004, 2006); Knobe & Mendlow (2004); Knobe (2004a, 2004b, 2007); Pettit & Knobe (2009); Tannenbaum, Ditto, & Pizarro (2007); Beebe & Buckwalter (2010), Beebe & Jensen (forthcoming); Beebe (forthcoming); Alfano, Beebe, & Robinson (2012); Robinson, Stey, & Alfano (forthcoming).
 The idea that seemingly predictive and explanatory concepts might also have a normative component is not entirely original with Knobe; Bernard Williams (1985, p. 129) pointed out that virtues and vices have such a dual nature.
 See also Scaife & Webber (forthcoming), who helpfully point out that “the same words spoken by a character in the story do not necessarily have the same meaning, or give the reader the same impression, when the surrounding story has changed.”