The Knobe effect: what Brian Robinson and I (more or less) said in London

(Our collaborator, Paul Stey, was unable to attend this conference.  He’s a terrific up-and-coming psychologist at Notre Dame.)

I assume you’re all familiar with the seminal Knobe effect experiment, but let me briefly refresh your memory.  In this experiment, participants first read a description of a choice scenario: the protagonist is presented with a potential strategy that would result in a side effect.  Next, the protagonist makes an explicit disavowal about the side effect, and chooses to go ahead with the strategy.  The strategy results as advertised: both the primary and the side effect occur.  Finally, participants are asked to attribute an attitude to the protagonist.  The effect that Josh Knobe found was that when the side effect was bad, participants were more inclined to attribute intentionality than when the side effect was good.

 

This raises a number of questions.  Does the effect derive from the semantics of the attitude term, or the pragmatics of the attribution?  Does the effect provide evidence that morality infects supposedly non-moral folk psychological concepts?  Is the effect restricted to violations of moral norms?  Is it restricted to attributions of intentionality?

 

Over the last decade, researchers have greatly expanded the diversity of both the norm-violations that trigger the effect and the psychological states whose attribution exhibits the asymmetry.  The effect crops up not only after the violation of a moral norm, but also after the violation of prudential, aesthetic, legal, conventional, and even descriptive norms.  The attribution asymmetry is found not only for intentionality, but also for cognitive attitudes such as belief, knowledge, and memory, for conative attitudes such as desire, favor, and advocacy, and for the virtue of compassion and the vice of callousness.

 

We think that an ideal explanation of this wide-ranging effect both would unify the documented diversity and show how the attribution asymmetry is not irrational.  Our attempt to do this takes its cue from a study by Knobe that found that people were more likely to attribute intentionality to a protagonist who, in the pursuit of profit, violated rather than conformed to a Nazi racial identification law.  No matter what the protagonist did, he violated an important norm.  If he followed the law, he acted immorally; if acted morally, he violated the law.  There’s more call for genuine dilemmatic reasoning in this vignette than in any other we’ve seen because in all conditions there are strong reasons on both sides.   In Knobe’s experiment, however, only the legal norm was made salient.  We hypothesized that the effect is driven not by the violation of a norm as such, but by the violation of a salient norm.  From this it would follow that when contrary norms are alternately salienced in the same vignette, the attribution asymmetry will follow suit.  In other words, if only the racial identification law is salient, participants will be more inclined to say that the protagonist acts intentionally when violating the law (and conforming to morality).  But if only the moral norm of non-malificence is salient, participants will be more inclined to say that the protagonist acts intentionally when violating morality (and conforming to the law).

 

The narrative requirements of the racial identification law vignette made it hard to de-salience the law, so we designed a structurally similar vignette to explore our hypothesis.

In the story stem, Carl inherits some money and is thinking about what to do with it.  The next stage diverges from previous studies by having four conditions instead of two: an interlocutor saliences both norms, just the self-regarding norm, just the other-regarding norm, or neither norm.  When she saliences only one norm, we have a standard Knobe vignette, but when she saliences both or neither, we have something new.  The next stage is also an innovation.  Instead of just having the protagonist respond with his disavowal and decision, we ask participants for their normative and descriptive expectations; we ask what they think he will and should do.  Only after participants answer that question do we have the standard binary outcome.  Our final innovation is to ask not only for an attribution of an attitude but also for an attribution of mental processing, a point we’ll elaborate on later.

 

The results of our experiment are highly suggestive.  Consider first attributions of intentionally helping.

Unsurprisingly, when Carl invests the money, attributions of intentionally helping cluster around the midpoint.  But when Carl donates the money, a characteristic pattern emerges.  Consider first just the contrast between the self-regarding norm and the other-regarding norm conditions.  Participants were more inclined to say that Carl intentionally helped others when Diana salienced the self-regarding norm than the other-regarding norm.  In other words, they were more inclined to attribute intentionality when he violated a salient norm than when he conformed to a salient norm.  That’s just your standard Knobe effect.  When you look at the wider pattern, however, you can see that what’s happening is not that norm violation leads to a spike in attributions, but that conforming to the only salient norm leads to a drop-off in attributions.  It’s as if participants are saying, “He only donated because that’s what Diana told him to do.  He was just running on autopilot.”

 

A mirror of this pattern emerges when we look at attributions of intentionally making retirement more comfortable.

Unsurprisingly, when Carl donates the money, attributions cluster around the midpoint.  But then look at what happens when he invests.  If we focus only on the conditions where a single norm was salient, we find higher attributions of intentionality when Carl violates the salient norm than when he conforms to it.  And again, if we dilate our focus to look at the pattern across all four conditions, what emerges is not that norm violation leads to a spike in attributions, but that conforming to the only salient norm leads to a drop-off in attributions.  Again, it’s as if participants are saying, “He only invested because that’s what Diana told him to do.  He was just running on autopilot.”

 

The pattern that’s emerging seems to fit the dual-process theory of mental processing, most closely associated with Daniel Kahneman.

According to dual-process theory, people are capable of handling decisions in two quite different ways.  They can engage in slow, deliberative thought, weighing reasons, forming beliefs, desires, and intentions, and acting on the basis of this mental work.  Or they can engage in fast, automatic, heuristic processing.  Deliberation is cognitively costly, so people tend to engage in it only when they face a difficult problem.  A dilemma, where strong reasons oppose one another, is such a problem.  What we want to suggest is that the participants in our experiment have an intuitive grasp on dual-process theory (a claim that Kahneman would be unsurprised by), which explains the pattern of responses we’ve just displayed.  When the protagonist acts contrary to a salient norm, participants tend to think that he took the slow, deliberative path to behavior.  When he conforms to the only salient norm, they think he went on autopilot.

 

One way to test this hypothesis is to ask, as we did, for attributions of mental processing, not just attributions of mental attitudes.  As far as I know, the current study is one of only two to ask for attributions of something other than mental attitudes.  (The other is another study that we recently conducted, which asked for attributions of character traits.)  It would corroborate our interpretation if participants attributed higher levels of deliberation to Carl when he violates a salient norm and lower levels when he conforms to the sole salient norm.

 

As a matter of fact, that’s exactly what we find.  When Carl invests the money, attributions of considering whether to help cluster around the midpoint.  But when he donates the money, attributions of considering whether to help are high unless helping was the only salient norm.  This pattern of attributions is consistent with the dual-process model.

Again, this pattern is mirrored for attributions of considering his retirement.  When Carl donates the money, these attributions cluster around the midpoint.  But when he invests, attributions of considering whether to retire in comfort are high unless retiring was the only salient norm.  Again, this pattern is consistent with the dual-process model.

 

 

Two further points before we proceed with our interpretation of the data.  First, a remark on the neither condition: you might have noticed that this condition tended to have higher attributions.  We’re not sure what to make of this, but we suspect that it has to do with the fact that the neither condition is structurally dissimilar from the other three.  In the self, other, and both conditions, an interlocutor raises a norm to salience, and Carl makes a disavowal; only after that little conversation does he act.  In the neither condition, none of that happens: he just makes a decision.  Before we ran the experiment, we hypothesized that the neither condition would look like the both condition, since there was no asymmetry of salience in either condition.  That hypothesis was borne out in the cases where we found our characteristic pattern.  But the neither condition had higher levels of attribution in the “off” conditions.  If anyone has a conjecture about why that might be, we’re all ears.

 

Second, you might have noticed that, in general, attributions are higher when Carl donates the money than when he invests it.  In fact, the partial eta squared for this effect is .46; since anything above .14 is considered a large effect, this is huge.  Why do we get this effect?  Well, it probably has to do with the fact that about 95% of the participants thought that Carl both will and should invest the money.  As it turns out, mTurk participants are egoists.  This was unfortunate because it meant that we could test neither the effect of norm salience on expectations nor the effect of expectations on attributions.  That said, the uneven split in expectations probably does account for the overall higher attributions when Carl donates.  What we think is happening in those cases is that participants are surprised, which leads them to think Carl engaged in deliberative processing rather than automatic processing, even if the norm salience should have led them to think otherwise.

 

This leads us to our hypothesized connection between norm salience and expectations.  Although we weren’t able to manipulate expectations in this study, we think it’s plausible to interpret our findings as suggesting that attribution levels are higher whenever Carl does something unexpected.  There are two ways he can go against expectations.  First, he can donate the money, regardless of which norms are salienced.  Since almost everyone expects him to invest it, donating it is surprising, which in turn leads participants to think that Carl is engaged in deliberative processing, which in turn leads them to think that he formed an intention.  The other way Carl can act contrary to expectations is by violating a salient norm; to put the same point another way, if Carl does what Diane suggests, adhering to the sole salient norm, that’s only to be expected.  So, when Carl violates a salient norm, participants’ expectations are flouted, and they are therefore more inclined to attribute deliberation and intentionality.  And when he conforms to the sole salient norm, participants’ expectations are confirmed, and they are therefore less inclined to attribute deliberation and intentionality.

 

Furthermore, attributions of deliberation and attributions of intentionality were tightly correlated: .41 for retirement and .46 for helping.

 

Of course, in all of our cases, the same norms are operative.  It’ll help people to donate to Oxfam regardless of whether Diane makes this salient.  Investing the money will help Carl to retire in comfort regardless of what Diane points out.  Hence, what matters is not whether Carl violates or conforms to some norm.  What matters is what he’s expected to do, which in turn depends on which norms are raised to salience.

 

If this is right, it fulfills our first desideratum: unifying the Knobe effect.  What remains is to fulfill the second desideratum: showing that the attribution asymmetry is rational.

 

Now, if people actually do deliberate and form attitudes in the way we’ve hypothesized, it of course makes sense to attribute deliberation and attitudes in the same way.  The question is therefore not whether the attribution asymmetry is rational, but whether the attributed asymmetry is rational (or at least, not irrational).  We think it is.

 

Recall that deliberative processing is slow and cognitively costly.  Since we have limited cognitive powers, it makes sense to curtail our deliberative engagement to those cases where ignorance would be most deleterious.  We contend that it makes sense to pause and deliberate when the action you’re about to take would violate a salient norm.  Deliberation in turn leads to the formation of various attitudes, such as beliefs, desires, and intentions.

 

The point can be put another way: though true beliefs are typically worth having, some true beliefs are more worth having than others.  For instance, the true belief that n is the winning number in a lottery is more valuable than the true belief that m is not the winning number.  What makes the pattern of deliberation we’ve described rational is the fact that true beliefs to the effect that one is violating a norm are typically more valuable than true beliefs to the effect that one is conforming to a norm.  One may be sanctioned for violating a norm, so forming a true belief about whether one has violated a norm (hence potentiating such a sanction) is valuable, regardless of whether one endorses the norm.  The chairman in the help condition of Knobe’s original vignette, for example, does not need to say to himself, “Wait!  I need to stop and think carefully about whether helping the environment is something that I should be doing.”  In the harm condition, however, an inner monologue like this might well be appropriate.  The same seems to hold for the CEO who is considering violating or fulfilling a racial identification law in Nazi Germany and indeed for any of the other protagonists in the Knobe effect literature.

 

3 thoughts on “The Knobe effect: what Brian Robinson and I (more or less) said in London

  1. Very interesting work. One quick comment (I’m running off to work).

    “though true beliefs are typically worth having, some true beliefs are more worth having than others. For instance, the true belief that n is the winning number in a lottery is more valuable than the true belief that m is not the winning number. What makes the pattern of deliberation we’ve described rational is the fact that true beliefs to the effect that one is violating a norm are typically more valuable than true beliefs to the effect that one is conforming to a norm. One may be sanctioned for violating a norm, so forming a true belief about whether one has violated a norm (hence potentiating such a sanction) is valuable, regardless of whether one endorses the norm. ”

    I think this covers some cases, but not all. Harman’s original example involved heating the barrel of a rifle. A few years back when I thought I’d try to publish on this, I found that students would respond to cases where someone buys tuna knowing that it comes with mayo (where they don’t like mayo) as cases where they get a sandwich with mayo intentionally and cases where someone buys the car they want with a color they don’t like/don’t care about as cases where they get a car with that color intentionally. (This is all hazy recollection, but the point I was testing for is whether non-moral and neutral features of the choice situation would allow us to produce the side-effect effect.) If the effect can be produced in cases where there’s no norm to violate, no cost to pay, etc., then it’s not clear what work norm salience does.

    • I’m not sure we disagree, Clayton. In my article with James Beebe and Brian Robinson, we show that the Knobe effect occurs for violations of purely descriptive norms. (The protagonist goes walking with or without her dog; in her neighborhood everyone/noone else happens to be walking with a dog; people are more inclined to say that she intentionally does what noone else is doing than that she intentionally does what everyone else is doing.) Your cases seem to involve the violation of a prudential or self-regarding norm: don’t buy stuff with features you don’t like. It’s not a moral norm, but we already know that the Knobe effect crops up not only for violations of moral norms, but also for violations of prudential, legal, conventional, and descriptive norms.

      Our model is not that people notice that a norm is salient, decide whether it’s important or has big positive or negative payoffs, and then proceed to deliberate (if the norm is important) or use heuristics (otherwise). Rather, the idea is that people are automatically sensitive to norm violations, which tends to send them to the deliberative route when a norm — even an unimportant, merely descriptive norm — is (about to be) violated.

  2. Pingback: Stone Links: Identity Check - NYTimes.com

Leave a Reply