Reversing the Side-Effect Effect

Some of you are already familiar with the Side-Effect Effect by now. In a new article forthcoming in Philosophical Studies, Paul Stey, Mark Alfano, and I demonstrate that we can reverse the effect and therefore explain what the effect is as never before possible.

Tony Hayward
Tony Hayward – former CEO of BP (credit: Justin Thomas)

First discovered by Joshua Knobe (and hence sometimes called the Knobe Effect), the effect’s hallmark is an asymmetric attribution of some psychological state or other. In Knobe’s original study, the chairman of the board alternately harms or helps the environment as a side-effect of a new profit-increasing program. Knobe found an asymmetric pattern of attributions of intentionality in the two cases: the folk said he intentionally harmed the environment, but he didn’t intentionally help it. Since this seminal work, a literature on the effect has developed, expanding it in terms of the states asymmetrically attributed (intentionality, desires, beliefs, knowledge, etc.) and the type of norm involved (moral, legal, prudential, aesthetic, descriptive, etc.). As the literature on the effect has expanded, it has become increasingly difficult to precisely characterize the nature of the effect. While it initially looked like people were more willing to say a morally bad side effect was intentional, that interpretation doesn’t hold up since the asymmetry is present in studies involving on non-moral norms. So what is the Side-Effect Effect anyway? How do we describe it? We can already say that the effect refers to the asymmetric attribution rates for various psychological states, but based on what? When are the attribution rates higher or lower? My co-authors and I hypothesized that the Side-Effect Effect has everything to do with violations of salient norms, regardless of what type of norm it is. Imagine a protagonist who, by his or her action, will produce one of two side effects, A or B. An interlocutor makes a particular norm N salient to the protagonist, and side effect A violates norm N, but B does not. If the protagonist’s actions bring about side effect A (which violates N) then people are more willing to say the protagonists intended and desired to produce A and believed A would happen than cases in which side effect B resulted. This is the standard asymmetry we’ve come to expect from the Side-Effect Effect: higher attribution rates for A than B. Suppose, however, the interlocutor makes a different norm M salient to the protagonist, and side effect B violates M but B does not. We predicted that then the asymmetry will reverse. People will report higher attribution rates for B than for A, exactly opposite from before. If correct, we will have demonstrated a new level of control and understanding of the effect, since as Daniel Kahneman put it, “The proof that you truly understand a pattern of behavior is that you know how to reverse it” (2011, 133). We also predicted that if both norms N and M were salient to the protagonist at the same time, then there would be no asymmetry. To test these predictions, we ran two studies, of which I’ll only preview the second. (Mark Alfano has previously summarized our initial study.) We utilized a sample of 1422 participants (603 female; Mage = 29.34), who were recruited and compensated using Amazon.com’s Mechanical Turk. For this study, in order to lessen the likelihood that our results were being influenced by some peculiarity of the specific narrative context of our vignettes (i.e., to be sure that participants weren’t responding something about Nazis or the environment for example), we employed a large scale methodology with vignettes that were structurally similar, but in different narrative contexts, following the example of Cushman (2008) and Cushman & Young (2011). (We would especially like to thank Fiery Cushman for suggesting this methodology and helping us implement it.) Specifically, there were twelve different vignette contexts, each with three different norm salience conditions, and two different outcome conditions. In each narrative contexts, there were two conflicting norms (one legal and the other moral) that the interlocutor alternately made salient. The narrative contexts varied the gender of the protagonists. Some contexts were historical, such as the Nazi Racial Identification law, the Trail of Tears, and the Spanish Inquisition. (Nobody expects the Spanish Inquisition.) Others were fictional, such as Fahrenheit 451 or The Hunger Games. (See below for the full list of narrative contexts.) Each participant viewed only one of 72 different vignettes (12 x 3 x 2). The use of 12 different vignettes was intended to illustrate the generalizability of the observed effects. Thus, in analyzing results we are essentially treating vignettes (rather than participants) as the level of analysis. Therefore, the analyses reported below are for all twelve vignettes. We considered variable means from each vignette as the “responses” from that vignette and then examined effects across all vignettes. As an example, here is the Legal Norm Salient / Law Violated Condition for the Nazi Racial Identification narrative context:

In Nazi Germany, there was a law called the ‘racial identification law.’ The purpose of the law was to help identify people of certain races so that they could be rounded up and sent to concentration camps. It also made it impossible to fulfill the racial identification law and the moral duty to protect innocent people.

Shortly after this law was passed, the CEO of a small corporation decided to make certain organizational changes.

The Vice-President of the corporation said: “By making those changes, you’ll definitely be increasing our profits. But you’ll also be violating the requirements of the racial identification law.”

The CEO said: “Look, I don’t care one bit about that. All I care about is making as much profit as I can. Let’s make those organizational changes!”

As soon as the CEO gave this order, the corporation began making the organizational changes and the results were exactly as the Vice-President predicted.

Among other things, we then asked participants to rate their agreement with the following statements:

  1. The CEO intentionally violated the racial identification law/moral duty to protect the innocent. [Intentionally Violated]
  2. The CEO desires to violate the racial identification law/moral duty to protect the innocent. [Desired to Violate]

The results bore out our predictions. Consider first the Intentionally Violated variable.

Asymmetric response reverse and norms are alternately salienced.
Figure 1. Summary of means for the Intentionally Violated dependent variable. Error bars indicate ± 1 standard error.

As you can see, the asymmetry clearly reverses depending on which norm is salient. Within the Law Violated condition of the Outcome factor, when the legal norm is made salient in either the Legal Norm condition or the Both Norms condition and the law is violated, there is a much greater tendency to report that the law was intentionally violated, b = –1.46 (.12), p < .001. When the moral norm is made salient, and the law is violated, the intentionality attributions are around the midpoint, which is considerably lower than the other two conditions. Within the Morality Violated condition, using simple contrasts we see that when the moral norm is made salient (in either the Moral Norm condition or the Both Norms condition), and morality is violated, there is a greater tendency to report that morality was intentionally violated, b = –0.80 (.21), p < .001. This result mirrors the effects observed within the Law Violated condition of the Outcome factor. That is, whenever a norm is raised to salience and the protagonist goes on to violate that norm, participants are more likely to report that the protagonist intended to violate the norm, whether legal or moral. Next, consider the Desired to Violate variable, where participants rated their agreement that the protagonist.

Asymmetric response reverse as norms are alternately salienced.
Figure 2. Summary of means for the Desired to Violate dependent variable. Error bars indicate ± 1 standard error.

Within the Law Violated condition, we find a significant effect of the Norm Salience factor, F(2, 22) = 21.72, p < .001. Thus, as Figure 2 illustrates, when the law was violated and the legal norm was made salient (as in the Legal Norm and the Both Norms conditions), participants are more likely to report that the protagonist desired to violate the law, b = –0.71 (.14), p < .001. As Figure 2 also illustrates, we see the mirror image of this effect within the Morality Violated condition of the Outcome factor, where we find a significant effect of the Norm Salience factor, F(2, 22) = 5.96, p = .009. Furthermore, a simple contrast revealed that when morality is violated participants are more likely to report that the protagonist desired to violate morality in the two conditions where the moral norm was made salient, b = –0.47 (.18), p = .014. This mirror effect is reminiscent of the Intentionally Violated dependent variable discussed above. These results indicate that by alternately saliencing norms, we can alternate the direction of the asymmetry. And when both salient norms are violated, there is no asymmetry; the side-effect effect disappears. So, it would seem we now have a way to classify the effect. The side-effect effect refers to the attribution of various psychological states at a higher rate when a salient norm as been violated than when it has been fulfilled.


Appendix First Paragraph of Twelve Vignette Settings:

  1. Nazi Racial Identification: “In Nazi Germany, there was a law called the ‘racial identification law.’ The purpose of the law was to help identify people of certain races so that they could be rounded up and sent to concentration camps. It also made it impossible to fulfill the racial identification law and the moral duty to protect innocent people.”
  2. Japanese Internment: “In the United States during World War II, there was a law called the ‘Japanese Internment law.’ The purpose of the law was to round up all Japanese Americans, remove them from their homes, and send them to internment camps. It also made it impossible to fulfill the Japanese Internment law and the moral duty to protect innocent people.”
  3. Jim Crow: “In the last century in the United States, there were laws called the ‘Jim Crow laws.’ Part of the purpose of these laws was to segregate blacks from whites in the public school systems. It also made it impossible to fulfill the Jim Crow laws and the moral duty to treat everyone with dignity.”
  4. The Hunger Games: “In the nation of Panem, there was a law that called for ‘the Reaping.’ The purpose of the law was to have parents enter their children into a lottery so that a few of them would be selected and sacrificed for the good of the entire country. It also made it impossible to fulfill the Reaping law and the moral duty to protect children.”
  5. Soviet Propaganda: In the Soviet Union, there was a law called ‘the Anti-Soviet Agitation law.’ The purpose of the law was to require radio announcers to read only state approved news items as part of the state’s propaganda campaign. It also made it impossible to fulfill the Anti-Soviet Agitation law and the moral duty not to engage in propaganda.”
  6. Fahrenheit 451: “In a fictional country, there was a law called ‘the firemen’s law.’ The purpose of the law was to have firemen burn all books so that the conflicting views contained in them wouldn’t offend anyone. It also made it impossible to fulfill the firemen’s law and the moral duty not to engage in censorship.”
  7. Underground Railroad: “In the United States in the 1850s, there was a law called ‘the fugitive slave law.’ The purpose of the law was to make it illegal to help slaves escape from their owners. It also made it impossible to fulfill the fugitive slave law and moral duty not to treat people as slaves. Shortly after this law was passed, the owner of hotel decided to make certain organizational changes.”
  8. Trail of Tears: “In the United States in the 1830s, there was a law called ‘the Indian Removal Act.’ The purpose of the law was to require several Native American tribes to relocate west of the Mississippi River, sometimes by means of military force. It also made it impossible to fulfill the Indian Removal Act and the moral duty not to force people from their land.”
  9. Spanish Inquisition: “In the Spanish Empire in the fifteenth and sixteenth century, there was a law that called for the Inquisition. The purpose of the law was to force Jews and Muslims to convert to Christianity and used methods that included torture and murder. It also made it impossible to fulfill the requirements of the Inquisition and the moral duty not to torture and murder people.”
  10. Diocletianic Persecution: “In the late Roman Empire, there was a law called ‘the Edict of Diocletian.’ The purpose of the law was to persecute Christians, which included public execution. It also made it impossible to fulfill the Edict of Diocletian and the moral duty not to execute people for their religious beliefs.”
  11. Zimbabwe Land Resettlement: “In the nation of Zimbabwe, there was a law called ‘the Fast-Track Land Resettlement Program.’ The purpose of the law was to remove rich ethnic minorities from their farms without compensation and sometimes violently in order to give the land to poor ethnic minorities. It also made it impossible to fulfill the Fast-Track Land Resettlement Program and the moral duty not to steal from people.”
  12. English Married Women’s Property: “In nineteenth century England, there was a law called the ‘Coverture law.’ The purpose of the law was to deny married women the right to own any property, transferring anything she had owned to her husband. It also made it impossible to fulfill the Coverture law and the moral duty not to discriminate against anyone based on gender.”