Exciting new findings!

John Bengson and I have been conducting some new research that those interested in the potential influence of evaluative considerations on intentionality attributions -- the 'Knobe effect' -- may find interesting. Specifically, our research has focused on the connection between judgments of responsibility (blame/praise) and intentionality attributions. In previous research, we found strong evidence for the view that judgments of responsibility (but not goodness/badness) are strongly related to intentionality attributions: that is, to the extent that people blame/praise, they judge that the action was done intentionally; and to the extent that they do not blame/praise, they fail to judge that the action was done intentionally. In this new research, we're testing whether or not a judgment of blame/praise can actually lead to (result in) an intentionality attribution, as our account of the Knobe effect predicts. Preliminary data using various manipulations strongly suggests that, as a matter of fact, judgments of blame/praise are leading to (resulting in) attributions of intentionality. Roughly, what we are finding is that manipulating whether participants think that the actor is blameworthy or not blameworthy leads participants to give different judgments about whether the actor acted intentionally – even when everything else about the vignette remains the same.

These sorts of manipulations are difficult because they require controlling for the actual judgment(s) that participants have already formed upon reading the vignette. Our strategy for doing this has been to give participants vignettes in which another subject is forming judgments about the original vignettes. So, for example, we have been giving participants vignettes in which they read about someone else (say, Joe) who has been given the HARM vignette and has been asked to judge both whether the chairman is blameworthy and whether he harmed the environment intentionally. Participants are then told in the vignette either that Joe judges that the chairman is blameworthy or that he judges that he is not blameworthy. Participants are then asked whether Joe will judge that the chairman acted intentionally or did not act intentionally.

The preliminary findings are quite provocative. When told that the character (Joe) judged that the chairman was blameworthy, the majority of participants judge that Joe also judged that the chairman acted intentionally. Yet, when told that Joe judged that the chairman was not blameworthy, the majority of participants judge that Joe also judged that the chairman did not act intentionally. This was so for 22 of 31 participants (71%), which is of course significantly higher than chance (2 additional participants changed their judgments, though not in the hypothesized direction). This means that, for the majority of participants, input regarding blameworthiness (or lack thereof) influences output regarding intentionality (or lack thereof), as it were. Of course, manipulations like the ones we're employing cannot provide conclusive evidence: nevertheless, they provide strong empirical support for the existence of a certain sort of relation. So, although we're still neck-deep in data collection, we believe that these preliminary findings provide reason to think that judgments of blameworthiness can and do influence intentionality attributions. We’ll keep you posted!

Comments

[Richard]

This raises some interesting methodological questions. E.g. aren't you really just measuring people's folk-psychological judgments (making inferences about what Joe thinks), rather than what they think about blame and intentionality? I guess you're assuming that participants' folk-psychological judgments derive from judgments about what they would think (about intentionality) if they were in Joe's shoes (re: ascribing blameworthiness). This is all very indirect, though; do you think it's reliable?

Also, I'm not sure I understand how any of this is evidence that "judgments of blame/praise are leading to (resulting in) attributions of intentionality." Couldn't participants simply hold that intentionality ascriptions are a precondition for ascriptions of praise/blame, and so tollens from Joe's lack of the latter to his lack of the former?

[Chris]

I share Richard's first concern. One of the things that has always impressed me about Knobe's experiments is that they are very simple, thus making the data easy to interpret, even if the ultimate theoretical interpretation is more difficult (for a variety of reasons). When you add the second observer, so that you now have an observer observing an observer, it gets pretty convoluted, and the potential interpretations multiply.

The problem lies specifically in trying to extend Knobe's method. It has several problems if you're trying to look at blame as opposed to what he designed it to look at. For example, there's a great deal of evidence that people will tend to assign blame when negative outcomes were avoidable. In Knobe's scenarios (at least the harm scenarios), it's quite clear from the way they're worded that the outcome was avoidable.

Instead, I think it would be better to design new scenarios that utilize things we know about blame assignment. For example, if an immoral act is a necessary cause of a negative outcome, but not a sufficient cause, then blame will tend to be assigned to the sufficient cause than the immoral necessary one. So in order to look at the relationship between blame and intention, you could vary whether the immoral act was a necessary cause of the negative outcome, or a sufficient cause of it. Your hypothesis would then predict that when the immoral act is only a necessary cause, and thus not assigned blame, people would not believe that the actor intentionally caused the negative outcome.

For examples of scenarios like this in the blame literature, check out:

N'gbala, A., & Branscombe, N. R. (1995). Mental simulation and causal attribution: When simulating an event does not affect fault assignment. Journal of Experimental Social Psychology, 31, 139-162.

Of course, this isn't the only way to do it. There's a large literature on blame, and thus a variety of ways of manipulating whether people assign blame.

[Simon Rippon]

Hi Jen,
I share Richard's second concern. Why say that the participant thinks Joe attributes intentionality _as a result_ of his pre-existing blameworthiness judgment? It seems more prima facie plausible that Joe attributes blameworthiness only when he thinks the act was intentional, and that if Joe doesn't attribute blameworthiness, one explanation of that is that he doesn't think the act was intentional.

[Brandon N. Towl]

To add to the fray-- I think both worries go hand-in-hand. True, the methodology is such that either attributions of praise/blame could influence attributions of intentional/not-intentional. And, if the subjects are just trying to understand why *Joe* made the ascriptions he did, then subjects can just "tollens their way back" as Richard suggested. But such ascriptions simply show how subjects reason about Joe, not how they themselves would describe the situation. In effect, the experiment is such that is "washes out" the Knobe effect completely. I can see why you wanted to do this... but it doesn't provide counter-evidence to Knobe's original findings.

Just out of curiosity-- why not also ask participants whether or not they thought that Joe *should* find the people in the vignettes blameworthy, etc. It would be interested to see if subjects thought that Joe had gotten the intentionality attribution wrong in some cases (for example, finding a "bad" action unintentional) but still got the prasise/balme attribtion correct.

[tnadelhoffer]

Jen,

Since I have defended the view that judgments about blame often bias judgments concerning intentionality, I would welcome more results that suggest this is the case. But I am unconvinced that the kind of studies you mention will help.

First, Richard, Chris, Simon, and Brandon all raise legitimate concerns that would first need to be addressed.

Second, I think you would have to make sure that there are not other non-moral explanations of the results you get. Consider, for instance, Machery's attempt to show that you get similar results in morally neutral but evaluative cases. As far as I can tell, Machery has successfully shifted the burden back to those of us who have assumed that moral considerations do all the work in the Knobe-style cases. Before we try to explain how and why the Knobe-effect occurs, we must respond to Machery's attempt to show that it does not occur at all!

Third, your approach only makes sense if we assume that when we engage in folk psychological explanations of other people's judgments and decisions we do not simply place ourselves in their shoes and ask ourselves what we would think or judge if we were in similar circumstances. After all, if the participants in your studies merely place themselves in the agent's shoes to see what they would judge, then why bother having their judgments mediated in the first place? Why not simply ask them directly whether they think the agent brought about the side effect intentionally? If, on the other hand, they are not placing themselves in the agent's shoes, the worry is that the Gricean issues raised by Adams and Steadman raise their ugly head.

Fourth, several pilot studies have shown that people's intuitions differ when they are presented with second-person rather than third-person vignettes. Your studies further complicate the matter by making them both second and third person.

These difficulties notwithstanding, I look forward to hearing more about the studies and results. As I said at the beginning, the view you are defending is a view I tentatively share. So, good luck!

[John and Jen]

First of all, thanks for all the great feedback! We (this is both John and Jen speaking) want to try to address each post in turn. But, first, we want to emphasize that our goal is simply to test whether

(C) a variation in input regarding blameworthiness sometimes results in variation in output regarding intentionality.

We want to test (C) because our account of the Knobe effect entails that

(T) judgments of blame/praise sometimes influence judgments of intentionality.

The specific thesis (T) entails the more general claim that the answer to the question "Is x blameworthy/praiseworthy for A-ing?" sometimes infliuences the answer to the question "Did x A intentionally?" This, in turn, means that our account is committed to the even more general claim (C): a variation in some sort of input regarding blameworthiness sometimes results in variation in some sort of output regarding intentionality. Given that (C) is a very general claim, the truth of (C) does not entail the truth of (T) or any other part of our account. But it does (indirectly, since abductively) support our account, for our account, unlike many other accounts, seems to be in a good position to explain why (C) would be so.

All of this aside, our main suggestion in the post was that the preliminary data we've collected so far supports (C). Why? Because when we vary the input regarding blameworthiness (e.g., Joe judged that the chairman was (not) blameworthy) we get significant corresponding variation in the output regarding intentionality (e.g., Joe judged that the chairman did (not) act intentionally). Hopefully these preliminary remarks illuminate the point of our research--it might not have been clear in the original post. Now onto your more specific comments:

Richard, first, you've asked exactly the right question: are our manipulations reliable? Although they are not conclusive, have limitations, and get at the relation between judgments of blameworthiness and judgments of intentionality indirectly, we believe that such manipulations can serve as reliable (and direct) tests of claims like (C). (See below for more on the directness/indirectness point.) Second, you ask: "Couldn't participants simply hold that intentionality ascriptions are a precondition for ascriptions of praise/blame, and so tollens from Joe's lack of the latter to his lack of the former?" Perhaps; if so, then (C) is confirmed: input regarding blameworthiness is leading (by way of an application of modus tollens) to certain output regarding intentionality. (Interestingly, we described this sort of modus tollens-esque inference in our paper. We're happy to hear that it's not outlandish to think that this inference seems pretty important.)

We want to be sure to distinguish (i) that our manipulations directly test the general claim (C) from (ii) that our manipulations indirectly test (T). Given how difficult it is to control for participant's original judgments (which seem extremely difficult to experimentally manipulate), it seems like an indirect test of (T) is necessary. But we're open to suggestions about how to test (T) directly, if you have any.

Chris, we see nothing to disagree with in your observation that "there's a great deal of evidence that people will tend to assign blame when negative outcomes were avoidable." That seems perfectly consistent with (C) (and (T)).

Simon, we hope we addressed your concern above.

Brandon, we agree that our manipulations are, in one sense, orthogonal to the Knobe effect. We're not trying to replicate the Knobe effect or provide counter-evidence to anybody else's findings, including Knobe's own. We're just testing (C). We take it that the standard way to test claims about the influence of one variable on another is to manipulate one in order to see what happens to the other. Hence our methodology. (Incidentally, we had considered using "should" in the way you suggest, but our worry was that this would not adequately control for participants' original judgments.)

Thomas, first, we hope to have addressed the concerns you share with Richard, Chris, Simon, and Brandon. Second, we completely agree with Machery that the Knobe effect is not specifically a moral phenomenon. So, we, too, are skeptical of moral explanations of the Knobe effect. Our account appeals to blame/praise, which need not be moral. (This is one place where our accounts seem to diverge: since your account (i) appeals to Alicke's work on moral blame and (ii) entails that the Knobe effect is the result of an affective bias, your view appears to differ in at least two significant respects from our own). Of course, blame/praise are always evaluative. But, as you indirectly point out, Machery's findings and arguments are consistent with the Knobe effect being an evaluative phenomenon. Third, it seems plausible that participants are putting themselves in Joe's shoes by making counterfactual judgments along the following lines: if I were to judge, as Joe did, that the chairman was (not) blameworthy, then I would judge, as Joe presumably did, that the chairman did (not) act intentionally. It would be extremely difficult to get participants to entertain such counterfactuals--or, more generally, to manipulate their original judgments--in the absence of another character (Joe). Fourth, we're interested to hear more about why you think that the reliability of manipulations like ours is undermined by the fact that they require participants to make second-person judgments. Since these sorts of manipulations are not uncommon in experimental psych (e.g., in theory of mind research), we find your suggestion intriguing and would be interested to hear it fleshed out in detail.

Once again, thanks for all the very helpful comments -- we appreciate all of your input!

[tnadelhoffer]

John and Jen,

A couple brief (and likely somewhat confused and confusing) early-morning remarks:

First: I would like to hear more about the notion of blameworthiness you have in mind that is purely evaluative and not moral. I, for one, was never comfortable with the way you drew this distinction in your earlier work. For me, it does not make sense to say that someone is blameworthy for x, if x is entirely devoid of moral import. It would, however, make sense to say that a person is responsible but not blameworthy for x-ing. On my view, blame is thoroughly moral.

Second: Keep in mind that in the original Knobe HARM vignettes, the harm was supposed to be perceived as immoral in addition to being blameworthy. Yet, you use modified versions of these vignettes to get at a different type of responsibility. I, for one, do not think you are entitled to such a move. Indeed, I predict that if you ran two studies--one of which involves a serious moral violation, the other of which merely involves an evaluative mistake--the Knobe effect will be greater in the moral case than in the evaluative case. This possibility reminds us that even if you can get the Knobe effect in non-moral cases, it may nevertheless be true that in moral cases--especially bad ones--the effect is further magnified by the affective mechanisms that get implicated when participants encounter the moral feature of the vignette.

Third: You suggest:
"Third, it seems plausible that participants are putting themselves in Joe's shoes by making counterfactual judgments along the following lines: if I were to judge, as Joe did, that the chairman was (not) blameworthy, then I would judge, as Joe presumably did, that the chairman did (not) act intentionally. It would be extremely difficult to get participants to entertain such counterfactuals--or, more generally, to manipulate their original judgments--in the absence of another character (Joe)."

Why, then, have them think counterfactually at all? If you only use Joe as a means of getting at what they would think if they were Joe, why not just ask them what they think rather than using Joe as the middle man? Put differently, if you are interested in how the participants make judgments about intentional action, why not just investigate that directly rather than examining how they judge other people to make such judgments. After all, either they assume the other people make judgments precisely as they do--in which case, using the other people is pointless--or they think other people make judgments differently--in which case, you can no longer take their answers as indicative of how they would respond.

Finally: The worry about second-person vs. third-person is something I plan on further investigating this summer. For now, I just want to give a simple example. If I ask an individual whether they would be tempted to behave immorally if they ceased to believe in free will and moral responsibility, I will get a different answer than if I ask them whether they think other people will behave immorally. I am confident this is partly tied up with the fundamental attribution error--but I am sure other issues are involved as well. The worry is that if you make me the star of the vignette, you might get different responses than you would have gotten had you asked me to imagine that someone else has to make the decision. I am not saying that this is definitely a problem. I am merely suggesting that it is an issue that we cannot simply sweep under the rug. If you ask me whether I should hit the switch in order to save five workmen by killing one, I may very well reason differently than if you asked me whether John should do the same. In your studies, you seem to include both the second person and the third person--further blurring the line between the two. After all, if I understand your methdology correctly, you get participants to make judgments not about the intentionality of the agent in the vignette, but rather about the judgments about intentionality made by one of the characters in the vignette. I take it that was partly what was worrying everyone in the comment thread. After all, you seem to assume that the participants were judging that Joe judged that the agent x-ed intentionally because Joe judged that the agent was blameworthy. But how do you know whether they were using modus ponens or tollens with respect to Joe's thinking? Now, it seems, we have come full circle!

Either way, I look forward to seeing the vignettes themselves.

[Chris]

Regarding (T), I'm pretty sure this is an empirically confirmed statement. That is, there's at least a 50-year history in psychology of theories of blame (and the empirical testing of those theories) in which intentionality and blame are shown to be related. An act is blameworthy only if it was intentional (or if there is some other sort of personal control involved, intention being a member of a larger category of types of personal control). Therefore, if an act is blameworthy, it was intentional, and if an act is the only available cause of an event, and that act is not blameworthy, then it was likely not intentional (or under personal control).

The problem is that it doesn't directly address the Knobe effect. For example, do you still get the Knobe effect when an act leads to a negative outcome, but is not the sufficient cause of that outcome, and therefore tends not to be assigned blame for the outcome? If so, then the Knobe effect isn't explainable in terms of blame (or at least, not in terms of blame alone), and his contention that the moral valence of the act affects its perceived intentionality still stands.

On Machery's experiments (I assume the reference here is to the commemorative cup and extra dollar cases), I honestly don't think they have any relevance to the hypothesized relationship between intentionality and moral valence. One of the great things about Knobe's scenarios is that they are perfectly alignable; the only difference is in the valence of the outcome. That's not true of Machery's scenarios. For one scenario, participants are asked whether Joe intended to cause the outcome of a scenario (getting the big cup). The outcome is morally neutral, Knobe's hypothesis predicts that participants would not assign intentionality in this case.

In the other scenario, participants are asked whether Joe's action itself -- paying another dollar -- was intentional. The analogous question, in Knobe's scenarios, would be, "Did the CEO intentionally adopt the new policy?" For participants to say that Joe didn't intentionally pay the extra dollar (or that the CEO didn't intentionally adopt the new policy), they'd have to believe either that he was forced, against his will, to do so, or that he did so without knowing it. Since the scenario makes it clear that neither of these is the case, it would make absolutely no sense for participants to say he didn't pay the extra dollar intentionally. It's not surprising, then, that Machery finds the assymetry. He's built the assymetry into the scenarios themselves! Machery is comparing apples and lawn mowers.

I know Machery mentions this objection (or at least a similar one) in his paper, but he doesn't actually address it. Instead, he argues that further research is needed to address it. I don't think further research.

Machery's second set of scenarios (the dog and worker scenarios) actually confirms Knobe's hypothesis, I think. Causing the death of a worker, even if appropriate (to save the life of five) has a much stronger negative moral valence than causing the death of a nonhuman animal. If moral valence plays a role in the assignment of intnetionality, then Knobe's hypothesis would predict that people would be more likely to assign intentionality in the worker case than in the dog case. Instead of presenting evidence against Knobe's theory, I think these two cases actually provide a more nuanced picture of the role of moral valence in judgments of intentionality. In fact, since judgments of intentionality were pretty much at chance in the worker condition (did Machery test to see whether the percentage, in that condition, was different from chance? I don't remember), it could be argued that his data show that conflicting moral valences (it's good to save the five, bad to kill 1, but the good of saving the five ultimately outweighs the bad of saving the five, though not to the extent that it outweighs the bad of killing a dog) lead to ambivalence with regard to intentionality.

[John and Jen]

Thomas, thanks for the follow-up. Let us try to respond to each of your comments in turn.

First, the literature on blame is rather large, and it appears to quite clearly recognize that blame is not specifically moral. Blame is oftentimes moral blame -- but not always. If it helps, think of praise for a moment. Surely someone can be laudable in a non-moral way. “Praise Bob -- he deserves it” one says after Bob does well in the big game. Similarly, someone can be criticizable or blameworthy in a non-moral way. “Bob’s to blame” one says after we lose the big game (or, after it is realized that we took a left turn rather than a right, on Bob’s command, and so on). You might have a sort of revisionary view according to which even these instances of praise and blame really are moral (and not just evaluative), appearances notwithstanding. But we’re just following what we take to be a very straightforward observation made in the literature on responsibility, so we think our usage is quite standard.

Second, we don’t see how our previous research used “modified versions of [Knobe’s HARM/HELP] vignettes to get at a different type of responsibility”. If we understand its implications correctly, that’s a pretty heavy charge. But since the only modifications we made were (1) to change ‘chairman’ to ‘chairperson’ (to avoid gender bias) and (2) to ask participants more questions about the case after the fact, we don’t see why it’s deserved.

Third, we’re starting to wonder if we’ve (mistakenly) represented ourselves as making a different, perhaps stronger, suggestion than we’re actually making. Just in case, we want to pause to clarify our main suggestion. Our main suggestion is that the preliminary findings support (C). Full stop.

We’re not presently testing “how the participants make judgments about intentional action”, as you suggest. That’s not our concern. Our concern is (C). We’re testing (C) using manipulations. We’re using manipulations because it is basically impossible to test (C) by simply asking participants to make judgments about blame and intentional action. Why? Because one cannot just read direction of influence off of that sort of data. Even if the judgments are significantly correlated (as we’ve found in previous research that they are), that is little or no indication of which one is leading to the other. The way to figure out direction of influence is to manipulate one variable (blame) to see if that results in a significant corresponding change in the other variable (intentionality). If so, then there is good reason to think that the one (blame) is influencing the other (intentionality). We cannot get at this simply by asking participants to make judgments about blame and intentionality. To test claims of direction of influence (such as the claims made by Knobe about badnes, you about moral blame, and Machery about cost, etc.), it is necessary to vary the alleged influencer to see whether that results in a significant corresponding variation in the alleged influencee. Clearly, this cannot be done simply by asking participants for their judgments about blame (or badness or cost) and intentionality, for that sort of study involves no variation of the alleged influencer--it involves no manipulation. Of course, that sort of study is great to use if one is trying to determine whether judgments of x (blame, badness, or cost) are significantly related to judgments of y (intentionality). In our previous research we found strong evidence that judgments of blame are significantly related to judgments of intentionality (and that once this relation was controlled for, badness was actually significantly *negatively* related to intentionality). Now we’re trying to test the directionality of this relation. And for that we need to engage in manipulations, since more of the same won’t do.

Importantly, we’re not taking participants judgments about Joe to be indicative of how participants themselves would respond to the HARM vignette. But we are taking their judgments to provide insight into how they understand the relaiton between blame and intentionality. For, presumably they’re relying on their own understanding of the relation between blame and intentionality to make the judgments about Joe that they’re making--that is, to infer that Joe would judge as they say he would. But, again, we’re not trying to replicate the Knobe effect; that we’ve done in previous research. In this new research, we’re just testing (C). (C) is not a claim about participants’ responses to cases like HARM, but the sort of claim one needs to use manipuluations to test -- that’s why we think our manipulations are appropriate. Joe’s purpose is to control for participants’ own judgments. We cannot simply tell participants: “ok, now judge that the act is blameworthy; given this, was it done intentionally? ok, now judge that the act is not blameworthy; given this, was it done intentionally?” Participants own judgments about blameworthiness are not so easily controlled for. As a result, this sort of naive manipulation would not be very effective, since we wouldn’t be sure that we were really varying the one variable. In order to control for participants’ original judgments, we felt that we needed to do something more. Hopefully this makes it clear why we ended up with the manipulations we did.

Finally, about second vs. third-person judgments: we agree that researchers need to be wary of the sorts of possibilities you gesture at. But we’re still not sure why you think that our research is undermined by these possibilities. We’re just testing (C). The research you mention regarding free will strikes us as quite different than the sort of research we’re doing. Our manipulations are modelled upon the sorts of studies used in theory of mind research, in which second and third person judgments are not uncommon. Just as theory of mind researchers ask participants to make second and third person judgments about how another person thinks about a third person’s mental states in order to determine how they themselves think about mental states (presumably they’re drawing upon their own understanding of mental states to make these judgments), we’re asking participants to make second and third person judgments about how another person thinks about blame and intentionality in order to determine how they themselves think about blame and intentionality (presumably they’re drawing on their own understanding of the relation between blame and intentionality to make these judgments). In any event, our methodology seems to pretty clearly test (C): that is, does varying some sort of input regarding blame result in a significant corresponding variation in some sort of output regarding intentionality? We still think that our preliminary data suggests that the answer is yes.

You mentioned that you’d like to see the vignettes. Here they are:

1. HARM - BLAMEWORTHY

Joe is given the following case:

The VP 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, but it will also harm the environment." The chairman of the board answered, "I don't care at all about harming the environment. I just want to make as much profit as I can. Let's start the new program." They started the new program. Sure enough, the environment was harmed.

After reading the case, Joe was asked to consider whether or not the chairman was blameworthy for harming the environment and also whether or not the chairman had harmed the environment intentionally.

Joe judged that the chairman WAS BLAMEWORTHY FOR HARMING THE ENVIRONMENT. Did Joe judge that the chairman harmed the environment intentionally?

[   ] PROBABLY, YES
[   ] PROBABLY, NO

2. HARM- NOT BLAMEWORTHY

Bob is given the following case:

The VP 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, but it will also harm the environment." The chairman of the board answered, "I don't care at all about harming the environment. I just want to make as much profit as I can. Let's start the new program." They started the new program. Sure enough, the environment was harmed.

After reading the case, Bob was asked to consider whether or not the chairman was blameworthy for harming the environment and also whether or not the chairman had harmed the environment intentionally.

Bob judged that the chairman WAS NOT BLAMEWORTHY FOR HARMING THE ENVIRONMENT. Did Bob judge that the chairman harmed the environment intentionally?

[   ] PROBABLY, YES
[   ] PROBABLY, NO

[John]

Chris, thanks for the comments. Just two quick points.

First, we don't think the case for a non-moral Knobe effect rests on Machery's very provocative studies. For instance, there are the producer cases in Turner (2004) and the decrease/increase sales cases in Knobe and Mendlow (2004), Phelan and Sarkissian (forthcoming), and our own paper. But since this issue is orthogonal to our main suggestion in the post, we'd prefer to put it aside for now.

Second, a brief response to one of your suggestions. You write:

"Regarding (T), I'm pretty sure this is an empirically confirmed statement. That is, there's at least a 50-year history in psychology of theories of blame (and the empirical testing of those theories) in which intentionality and blame are shown to be related. "

This sounds great -- for one, it would save us a lot of work! But, unfortunately, showing that two variables are significantly related does not by itself show that one of those variables sometimes influences the other variable. The direction of influence could be uni-directional, or there could be a common influencer. To test claims about the direction of influence, one needs manipulations. As far as we know, no one has ever tested the direction of influence between blame and intentionality; no manipulations have ever been done. If you happen to know of any such manipulations, we'd be interested to hear more.

[tnadelhoffer]

John and Jen,

First, I was not trying to make any "heavy charges" at all. I was simply confusing your earlier stuff with the paper by Phelan and Sarkissian--where the distinction between blame and responsibility is important. Hey, I admitted in advance that I what I was going to say might be confused and confusing! It was early and I was not yet properly caffeinated. My goal was simply to generate discussion and give you feedback. I was certainly not trying to sling undeserved charges...

Second, obviously I am a fan of (C), so I am happy with any additional evidence that speaks in favor of it. In that respect, I think we are birds of a feather.

Third, I, too, would like to know how to test whether blame attribution comes before ascriptions of intentional action. And I also realize that to date, the studies have only shown that the two are related, not that one comes before the other. As such, I applaud your effort to examine the issue. I am just still unsure how it is supposed to work. Keep in mind, you ask:

"Bob judged that the chairman WAS NOT BLAMEWORTHY FOR HARMING THE ENVIRONMENT. Did Bob judge that the chairman harmed the environment intentionally?"

Assume they answer yes. How is that supposed to rule out the participants having judged that Bob reasoned thusly: "Given that the CEO knowingly harmed the environment and harming the environment is bad, he intentionally harmed the environment. And because he intentionally harmed the environment, he is blameworthy."
It's as if you've given them an argument for the conclusion that Bob is blameworthy with all of the premises supressed. They've simply filled in the missing premises. How do you address this interpretation of the results?

[tnadelhoffer]

I also wanted to make a suggestion for testing (C) more directly: You could use fMRI studies to see which regions get implicated first by scanning people while having them read Knobe-style action theory vignettes. Assume that the regions associated with blame and punishment get implicated before the regions associated with folk psychological judgments. Wouldn't that provide evidence for (C)? Just a thought really...

[Chris]

A couple things:

First, you're definitely right, showing that they are related is not enough. There is plenty of evidence that they are. Some form of personal control (which would include intentionality) is necessary for blame assignments. There are plenty of experiments demonstrating that, even in young children. I'm not sure of I know of any studies that specifically demonstrate that blame implies intentionality, or that the lack of blame implies a lack of intentionality. But extrapolating from the data we do have (from previous studies), I think it's pretty safe to assume that with one possible cause, intentionality will not be inferred if blame is not attributed to an act leading to a negative outcome.

That's why I think you need to present cases in which the same action will be blamed or not blamed, depending on its role in a causal chain.

As for Thomas' fMRI study suggestion, I don't think it's a good way to go. For one, a well-known problem with current imaging technologies is that they're either good at time course analysis or good at localization, but none are good at both. So trying to figure out the order of activation in particular areas would be difficult. Second, trying to localize things as abstract as blame and intentionality would be, well, beyond our capabilities at present, I think.

[Jen]

Perhaps it would help to mention the blame/not blame vignettes are only two of eight vignetes in total that we are giving to participants. With the HARM vignette, they are also being told that the reader either thinks the chairman acted intentionally or not intentionally and asked whether the reader will find the chairman blameworthy or not. In addition, they are being given the HELP vignette in a way that mirrors the set up for the HARM vignette: that is, they are either told that the reader finds the chairman praiseworthy or not or they are being told that the reader thinks the chairman acted intentionally or not. And then we ask what they think the reader will judge on the other issue (I hope this makes sense).

When we vary the intentionality information that participants are given (Jane thinks the chariman acted intentionally while Fred does not), 84% of the participants likewise vary their blame answers: Jane finds the chairman blameworthy, but Fred does not. Interestingly, only 71% of the participants vary their intentionality answers when given varying judgments of blame. So, it isn't clear that participants reasoning is the same in both the cases in which they have intentionality information and those in which they have blame information (otherwise one would expect the percentages to be approximately the same).

Nor do I see any reason to think that what they are doing is what Thomas suggests above: "Given that the CEO knowingly harmed the environment and harming the environment is bad, he intentionally harmed the environment. And because he intentionally harmed the environment, he is blameworthy." Why? Because, if this was the case, then the change in blame information would have no effect on their judgments about intentionality -- if the chairman did it intentionally, he did it intentionally. End of story. Thus, whether or not Joe blames him, he should still judge that he acted intentionally. In other words, particpants would just be confused by Bob's judging the chairman to be not blameworthy -- and certainly, it should have NO influence on whether or not they think Bob would judge that the chairman acted intentionally. After all, intentionality is being determined by "knowingly" bringing about a "bad" outcome. The presence or absence of blame should be irrelevant.

Yet what we are finding is that the presence or absence of blame is NOT irrelevant.

As we put it in our paper, the folk-psychological "equation" being used might look something like this:

good/bad act + intentionality = responsibility (blame/praise)

You can "solve" this equation for one variable by having information on the other two -- whichever they are.

I should mention, interestingly enough, that we are NOT finding the same effect for praise/not praise (or, for that matter, for intentional/not intentional) for the HELP vignette. Only 58% of participants have varied their judgments of praiseworthy or not when we vary the intentionality information and only 42% have varied their judgments of intentionality when we vary the praiseworthy or not information.

Of course, we are operating on a low n so far -- I'm continuing to collect data so the % may change.

[Edouard Machery]

A few comments related to my work.

Chris writes:

"An act is blameworthy only if it was intentional (or if there is some other sort of personal control involved, intention being a member of a larger category of types of personal control). Therefore, if an act is blameworthy, it was intentional, and if an act is the only available cause of an event, and that act is not blameworthy, then it was likely not intentional (or under personal control)."

This is obviously a fallacious argument. Being intentional is certainly a necessary condition for being blameworthy (at least among Occidental cultures). But the last sentence does not follow from this: Me typing is the cause of the appearance of the letters on this screen and me typing is not blameworthy, but certainly my typing is intentional. Or am I missing something...

Chris then writes: "the only difference is in the valence of the outcome. That's not true of Machery's scenarios"

Indeed, the very point is to replicate the asymmetry with a pair of cases that do not vary with respect to the valence of the outcome. So, I don't quite see how this counts as an objection.

Christ goes on: "In the other scenario, participants are asked whether Joe's action itself -- paying another dollar -- was intentional. The analogous question, in Knobe's scenarios, would be, "Did the CEO intentionally adopt the new policy?""

Of course not. In both the harm case and in the extra-dollar case, the question bears on an action of the agent--harming the environment and paying an extra-dollar. Thus, the analogous question in K's scenario is "did the CEO intentionally harm the environment?" I really don't see what justifies Chris' claim...

So much the worse for Chris' conclusion: "Machery is comparing apples and lawn mowers.".

Chris then discusses the second pair of cases. He expresses skepticism: "Machery's second set of scenarios (the dog and worker scenarios) actually confirms Knobe's hypothesis, I think. Causing the death of a worker, even if appropriate (to save the life of five) has a much stronger negative moral valence than causing the death of a nonhuman animal."

But Chris is again mistaken. K's view (at least his official view) is not that the negative moral valence explains the asymmetry. Rather, the official view is that the asymmetry results from the fact that harming the environment is *morally wrong* while helping the environment is *morally right*. Because people judge that causing the death of the single work was morally appropriate, they should judge that it is non-intentional if K's view is correct.

And as noted by John, the asymmetry is found in numerous other non-moral cases. This entails the following constraint on a correct explanation of the asymmetry between judgments about the intentional status of actions: It should be applicable both to moral cases and to non-moral cases. The trade-off hypothesis (proposed in my forthcoming paper) might not be the final word on the issue, but at least, it fulfills this constraint.

I'll try to comment on John and Jen's findings later.

[tnadelhoffer]

Jen,

OK, now it makes more sense. But can we see all of the cases? It would sure make it a helluva lot easier to visualize.

For now, I have another question. You say,

"Interestingly, only 71% of the participants vary their intentionality answers when given varying judgments of blame."

That is interesting indeed. For if (C) were true, wouldn't we expect that changing the blame would change the intentionality rating? Am I misunderstanding something here?

On a related note, why not think the folk are operating with the following simple heuristic:

P is blameworthy for bringing about x iff P brings about x intentionally and x is bad.

If participants had something like this in mind, it would explain why they answered the way they did in the two cases you mentioned. In HARM-BLAMEWORTHY, you tell the participants that Bob judged that the CEO is blameworthy, which just affirms one side of the bi-conditional thereby entailing the other side (i.e., if blame then intentional). In HARM-NOT BLAMEWORTHY, participants are told that Bob judged that the CEO is not blameworthy, which just denies one half of the bi-conditional, thereby denying the other half (i.e., if not blameworthy, then not intentional). I take it that was the conditional worry expressed early on. Am I just being dense? How do you know which judgments allegedly came first?

Perhaps what we should say is that so long as the relationship between the two could be mutual entailment as far as the folk are concerned, it may not make sense to talk about one coming first as a rule. In some contexts, blame attribution comes first. In other contexts, ascriptions of intentional action come first. Of course, all of this is consistent with (C). Which is fine with me--my own view (or at least the view I have defended in print) depends on the truth of (C).

Finally, you mentioned earlier that you had fixed the vignettes to control for gender bias (chairperson vs. chairman), but the ones you posted still use the gender-biased term. I suppose it was a typo :)

[Jen]

Thomas, you wrote: "if (C) were true, wouldn't we expect that changing the blame would change the intentionality rating?" But this is precisely what we are finding -- 71% (actually, it is 74% now with new data that I entered last night) are changing their intentionality attributions based on changes in blame information. The only reason I put "only" is that this percentage is lower (though still highly significant) than the percentage that are chaning their blame judgment based on changed intentionality information.

I think your bi-conditional is ok: the crucial point being that information from either side can influence judgments formed about the other side. This is all our account needs to show.

RE: the vignettes, there are too many to post on the blog, I think, but I'd be happy to post them on my website so that they are available there if that would help.

[Chris]

Jen and John,

I'd be interested in seeing the other scenarios that you've used too. I think I've been suggesting for a couple years in comments here and on my own blog that research on counterfactual reasoning and blame might explain Knobe's results, though every time he asked me how we might test that hypothesis, I drew a blank (I think I even said to him, at one point, that I didn't know how to separate intentionality and blame). So whatever reservations I've expressed, I'm definitely interested in your studies, 'cause you've definitely succeeded where I've failed for a long time.

I still think Thomas' criticism, which is similar to the one I was trying (and failing) to express, is valid, though. Specifically, I don't think this is the case:

"Because, if this was the case, then the change in blame information would have no effect on their judgments about intentionality -- if the chairman did it intentionally, he did it intentionally. End of story."

The necessary conditions for the assignment of blame are threefold (which are, together, sufficient to lead to the assignment of blame): negative outcome, personal control (e.g., intentionality), and sufficient cause. It's quite likely that if participants see a negative outcome and a sufficient cause (as in the scenarios you've described so far), but are told that the person is not to blame, their only possible explanation for the lack of blame is that the act was not under personal control (in this case, not intentional). Or at least, they have to infer that the person reading the scenario doesn't believe the action was under personal control Otherwise, that person would have assigned blame. Assuming, that is, that participants are acting as though the individual in the scenario is using the same reasoning process they do. If that's the case, then it certainly qualifies Knobe's hypothesis. It says that people don't infer intention for morally bad outcomes if they have independent evidence that the outcome was not under personal control. But it doesn't argue against Knobe's hypothesis more generally. That is, it doesn't say that people are using moral information to infer intention, it just says that it's not all they're using. I'm pretty sure that Knobe has explicitly agreed with that.

To provide evidence against Knobe's hypothesis more generally, you'd have to provide conditions in which the lack of blame does not directly imply a lack of intention. That is, you'd have to provide conditions in which there are explanations for the lack of blame other than a lack of personal control (e.g., intervening causes). If, in those cases, the Knobe effect still disappears when blame is not assigned, then you'd have strong evidence against his hypothesis.

[John]

Chris,

Sorry it's taken me so long to get back to you. Typepad hasn't been letting me post a response. I had four thoughts:

First, the following recipe should make the contents of the other vignettes clear enough. Consider the pair of vignettes posted above. First, just replace 'blame' with 'praise' to get another pair. Then, to get the rest of the vignettes, just switch blame/praise and intentionality in the first set of 4 (so that now the vignettes are giving participants the intentionality judgment and then asking what the blame/praise judgment would then be, rather than the other way around).

Jen is going to post the vignettes on her website, just in case you're still having trouble. Just go to her website (click on her name in the 'contributors' column) and make your way to her philosophy page. There will be a link from there to an experimental philosophy page, which will have a link to the vignettes.

Second, would you mind sharing the source of your necessary conditions for blame? Also, are they supposed to hold generally, or just typically (or perhaps normally)? I'm a bit worried that all three conditions are open to counterexamples.

Third, I'm not sure whether we're miscommunicating or what, but we seem to be on different pages. The suggestion that Jen and I have made is simply that the data supports (C), repeated below:

(C) a variation in input regarding blameworthiness sometimes results in variation in output regarding intentionality.

As we've tried to emphasize, we're not attempting to provide evidence against Joshua's view right now; nor are we are arguing that the Knobe effect is non-moral. Those issues are orthogonal to this new research.

Fourth, you say that you (and Thomas) are criticizing our suggestion. You say that your criticism is that:

"It's quite likely that if participants see a negative outcome and a sufficient cause (as in the scenarios you've described so far), but are told that the person is not to blame, their only possible explanation for the lack of blame is that the act was not under personal control (in this case, not intentional)."

I’m sorry if I’m not getting something that’s obvious, but I just don't see how this is a criticism of our suggestion. Your idea seems to be that participants are using the input regarding blame, together with some other information they have about the case, in a (strong, straightforward abductive) inference to the intentionality conclusion. But this is entirely consistent with our suggestion that the data supports (C): the blame input is being used to determine the intentionality output.

Regardless of how participants are reasoning, the fact remains (and this is just pure data speaking--no interpretation), 71% (well, now it's even higher: 74%) produce a different output regarding intentionality when they are given one input regarding blameworthiness than when they are given a different input regarding blameworthiness. Given this, I think we are warranted in concluding that a variation in input regarding blameworthiness results in a significant corresponding variation in output regarding intentionality. In other words, the data supports (C).

It's interesting to speculate about why input regarding blame influences ouput regarding intentionality. Jen and I believe that the answer lies in the diagram she mentioned in one of her posts above. But, for all this new research tells us, participants may be using your set of necessary conditions or Thomas's heuristic or some other set of necessary conditions or some other heuristic or no conditions or heuristic at all. Perhaps they're inferring via some extensionally or intensionally correct or incorrect biconditional, reasoning by modus ponens or modus tollens or reductio ad absurdum, making a Harman-esque inference to the best total explanation, intuiting the third realm, or what have you. This new research is silent on this issue. And, as far as I can tell, this issue is silent on this new research. All this new research tells us is what we take to be the rather exciting fact that variation in blame input results in a significant corresponding variation in intentionality output. In short, it provides evidence that the blame input influences the intentionality output. That’s pretty cool, no?

I hope this clarifies the position Jen and I are defending in this thread. If you think your criticism still sticks, would you mind walking me through your worry, explaining why exactly you think that (C) is not supported by the data? Thanks.