YAAS social norms

Yet Another Amateur Summary:
Bicchieri's Norms in the Wild: How to Diagnose, Measure, and Change Social Norms

When agents act or plan to act, they do so on the basis of both beliefs and preferences. Alternatively, we can phrase these as reasons and passions. Or we can say that a reward function and a model of the environment are required for a policy.

Foundation

Before we can analyze collective behavior, we have to make a few foundational distinctions.

Preferences

Individual or social

Preferences can be either individual or social preferences. Social preferences are those that “take into account the behavior, beliefs, and outcomes of other people” while individual preferences do not (Bicchieri 2016). (Example to come.) This distinction is important because changing individual preferences can plausibly happen in isolation, one person after another. Changing social preferences on the other hand is more likely to require coordinated group action.

Conditional or unconditional

Preferences can also be conditional or unconditional¹. Conditional preferences are those that vary with some feature of the environment while unconditional preferences do not. This is an important distinction because a social engineer can change the way conditional preferences manifest by changing the environment while unconditional preferences can only be altered in a direct confrontation.

NitW also draws a distinction between prudential and moral preferences. I think these can be viewed as a special case of conditional and unconditional preferences. It seems we can gloss these to preferences which seek to satisfy instrumental or intrinsic values. Preferring not to cheat out of a fear of getting caught is a prudential preference; preferring not to cheat because it’s an odious breach of faith is a moral preference. This distinction is important because moral preferences are typically more stable—prudential preferences may change with changing circumstances while moral preferences will not.

Together

If we put these two axes together, we end up with a classificatory grid like this:

Beliefs

Normative and non-normative

We can also distinguish between normative and non-normative beliefs (ought vs is). This distinction is important when discussing norms because demonstrations and other empirical evidence can be used to change non-normative beliefs but are harder to apply to normative beliefs.

Social and non-social beliefs

Some beliefs are social and others are not. Social beliefs are “expectations we have about other people’s behaviors and beliefs” (Bicchieri 2016). This distinction is important for much the same reason as the distinction between social and non-social preferences is important—it determines whether a social engineer has a divisible problem or one that must be tackled at the scale of groups.

Together

If we put these two axes together, we end up with a classificatory grid like this:

To be explicit, empirical expectations are “beliefs about how other people are going to act or react in certain situations” while normative expectations are “beliefs about other people’s personal normative beliefs (i.e., they are second-order beliefs)” (Bicchieri 2016). That is, “All do it” is an empirical expectation while “All approve of it” is a normative expectation.

Collective behavior

With those distinctions made, we can start to suss out the varieties of collective behavior. NitW suggests something like the following overview:

We’ll start from the bottom. There are three different types of collective behavior identified here:

Custom

“a pattern of behavior such that individuals (unconditionally) prefer to conform to it because it meets their needs” (Bicchieri 2016)

Descriptive norm

“a pattern of behavior such that individuals prefer to conform to it on condition that they believe that most people in their reference network conform to it (empirical expectation)” (Bicchieri 2016)

Social norm

“a rule of behavior such that individuals prefer to conform to it on condition that they believe that (a) most people in their reference network conform to it (empirical expectation), and (b) that most people in their reference network believe they ought to conform to it (normative expectation)” (Bicchieri 2016)

If the distinctions aren’t quite clear yet, hopefully walking through the diagnostic process will clarify. We start at the top of the diagram with an observed collective behavior.

Socially conditional

Our first fork is about whether the behavior is conditional on social expectations–that is, do social expectations have causal influence. In other words, if our preferences are in the bottom right of the preferences grid above—they’re socially conditional—we take the right of the fork. If they’re not socially conditional, we take the left side of the fork. If we’ve taken the left side of the fork, we don’t care to make any further distinction at present and just call all such collective behaviors customs. In these cases, we see many agents making similar choices because they are all acting out similar preferences in a similar environment; no communication is required. An example of a custom like this is (almost) everyone wearing a coat in cold weather—the fact that I expect others to also wear a coat is irrelevant to my decision.

Type of social expectation

When we took the left side of the first fork, we ended in customs. What happens if we take the right side of that initial fork? In that case, we’re affirming that social expectations have causal influence over the collective behavior.

The question that remains and constitutes the second fork is “What type of social expectations?”. If the only expectations that are a basis for the action are empirical, we call the collective behavior a descriptive norm. If both empirical and normative expectations are relevant, we call it a social norm. In terms of the beliefs grid above, we’ve already restricted ourselves to the right column by taking the right side of the first fork (“It’s socially conditional”) and now we’re determining whether we’re in the top right or bottom right.

Examples might help further clarify the distinction: - Everyone walking on the right side of a walkway is a descriptive norm. If you expect everyone else to act that way, you achieve your aims best by conforming. Violations are more likely met with eye-rolls than disapprobation. - Waiting your turn in line is (in many cultures) a social norm. Not only do others behave that way, but they believe that everyone ought to behave that way. Violations are likely to meet with sanctions.

Measuring norms

Phew. Done with all the conceptual analysis. Now we can start to try to apply this analysis in the world. If a social engineer wants to change a norm, their first step is to confirm that there is actually a norm in effect. This is a multistep process that will mostly follow the diagram above.

Collective pattern of behavior

The first step is simply to confirm that there is indeed a collective pattern of behavior. Just because a social engineer thinks something is common practice, doesn’t mean it actually is—perceptions don’t always match reality. The practices part of a KAP survey is one common tool that can be used for this. This step is generally straightforward social science and doesn’t have any norm-related special sauce so we’ll move on.

Social expectations

The next step is to establish whether there are normative social expectations behind the collective pattern of behavior. To this end, a social engineer must measure the normative beliefs of the individuals in the relevant community.

Questionnaires are a fairly straightforward way to do this. Crucially, the survey should ask about each of personal normative beliefs (“Do you personally think female genital cutting is morally obligatory, permissible or forbidden?”), empirical social expectations (“How prevalent do you think female genital cutting is in your community?”), and normative social expectations (“What fraction of your community do you think believes that female genital cutting is morally obligatory? Permissible? Forbidden?”). We can actually go even further as outlined in the following table:

Pluralistic ignorance

A social engineer needs to ask about personal normative beliefs and normative social expectations because these don’t always align. When each person (or most people) in a community falsely believes that others support a norm and their private disapproval is exceptional, we have pluralistic ignorance. The table on female genital cutting, for example, suggests a serious divergence between private beliefs and public professions due to pluralistic ignorance in countries like Djibouti and Somalia.

The distinction between norms which are the result of pluralistic ignorance and those which are the result of genuine community endorsement is, of course, an important one. We’ll talk more about the different strategies these scenarios demand shortly.

Methodological tips and tricks

While the basic idea of questionnaires is straightforward, there are several subtleties that apply in this domain.

Disinterest

People may simply try to complete the questionnaire as quickly as possible and, in doing so, give inaccurate responses. This problem can sometimes² be ameliorated with monetary incentives—“If your answer about the prevalence of this moral belief in your community turns out to be correct, we’ll pay you a bonus.”. Unfortunately, monetary incentives like this can induce distortions of their own.

Social desirability bias

Social desirability bias is likely to be particularly important when surveying people about moral beliefs. To combat this, a social engineer can use randomized responses.

This is by no means a complete list but does hint at the care which must be taken when designing questionnaires in the investigation of social norms.

Conditional social expectations

Our diagram actually compresses things and elides an important distinction. “if they have social expectations” requires both that there are social expectations and that those expectations are causally relevant. In the previous measurement step, the social engineer confirmed that social expectations are present. But, because actions are multiply determined, this information alone isn’t enough to determine whether the social expectations are causally relevant. A collective pattern of behavior might actually be sustained by legal mandates or identical private moral beliefs or ignorance of alternatives (or etc.) with the social expectations incidental.

When I hear causal, I think counterfactual and when I hear counterfactual, I think experiment. However, outside the lab, it’s difficult to manipulate social expectations. The best option for assessing the counterfactual may be describing and asking about hypothetical scenarios.

Hypotheticals that contradict known fact are often difficult for people to answer. Thus, it’s usually better to use vignettes describing possible futures or fictional characters. Instead of “Would you still practice female genital cutting if your community opposed it?”, prefer “Would you still practice female genital cutting if you moved to a new community that opposed it?” or “Nkemdilim lives in a community similar to yours, but it does not approve of female genital cutting or sanction those that refrain from it. Should she still participate in female genital cutting?”.

While hypotheticals like this are imperfect—people probably have difficulty imaginatively transposing themselves and “forgetting” their current perspective entirely—they do offer our best clues. If people suggest changed behavior in response to hypotheticals which alter social expectations, that’s good evidence the collective behavior is indeed conditional on social expectations—that the expectations are causally relevant. Hypotheticals also allow us to suss out under which conditions the norm applies and which expectations are most important for the maintenance of the norm.

Changing norms

Let’s recap for a moment. We started out with a conceptual analysis of social norms and related concepts. Then we looked into how a social engineer could apply these concepts in the real world via measurement. If the hypothesis was correct, measurements confirm that the social engineer is dealing with a social norm—a collective behavior that’s conditional on empirical and normative social expectations.

Now, we get to the part that justifies (at least my) interest in all this: the possibility of changing norms. (Bicchieri 2016) outlines four components of norm change:

1. “[P]eople must face a collective action problem”
2. “[T]hey must have shared reasons to change”
3. “[T]heir social expectations must collectively change”
4. “[T]heir actions have to be coordinated”

We’ll tackle these in order.

Problem

A harmful (from the social engineer’s perspective) norm may persist for one of many reasons (Bicchieri 2016):

• “People may not see the current norm as problematic, and even if presented with alternatives, they may defend their ways as superior.”
• “A second reason why a norm may not seem problematic is the lack of knowledge of possible alternatives”
• Pluralistic ignorance

Shared reasons

Because of each of the above problem scenarios has distinct implications, we’ll consider them separately when walking through the shared reasons part of norm change.

Expectation change

Even if the above persuasion succeeds, that’s not enough. Individuals privately supporting an alternative norm doesn’t necessarily lead to changed behavior. The reason is that social norms, unlike customs and descriptive norms, are rarely self-enforcing.

By self-enforcing, we mean that individuals will follow some pattern of collective behavior even in the absence of any social response to their behavior. People don’t need social approval before putting on a coat in cold weather. Walking on the wrong side of the sidewalk would be self-defeating because you’d get to your destination slower due to the need to dodge everyone walking against you. On the other hand, an unscrupulous person might cut in line every time they thought they could get away with it. The near absence of line cutting (in certain cultures) is maintained only through continuous social sanctions for norm violators.

Because social norms are socially enforced, unilateral defiance is rarely advantageous. It’s only when social expectations collectively change that changed behavior becomes possible. All of this means that attempts to advocate for alternative collective behaviors must be broadly effective before any change will be visible—there’s a threshold. But even convincing a majority of people that some alternative is preferable isn’t enough.

The social engineer must convince each person in a community of three things:

1. An alternative norm is superior (accomplishing this is covered in the shared reasons section)
2. Other people in the community support an alternative norm
3. Other people in the community are also convinced of the above

If the social engineer succeeds only at the first two, each individual will still be afraid to defy the existing norm. For example, even if Fiorenzo believes that others in his community privately dislike some norm, if he thinks they believe the norm still has public support, he’ll think they’ll still enforce the norm and sanction him for violating it³.

The necessity of common knowledge⁴ means that public fora are quite useful. When people discuss norms, their concerns, and their changing attitudes with other members of their community, it becomes possible for people to “infer that others’ beliefs are changing alongside their own” (Bicchieri 2016).

Collective action

We’re almost there. In the first three steps, the social engineer identified why a problematic norm persists, convinced people privately that an alternative is superior, and then made people’s preferences for change common knowledge. The final step is to realize these common preferences and enact a new norm.

Of course, this isn’t easy; it is a problem of collective action. The core issue is that individual incentives can diverge from group incentives. For example, cutting in line is advantageous for any individual if not punished. But if every individual tries to cut, the outcome is worse for everyone than adhering to the no cutting norm.

This echoes back to the discussion on social norms not being self-enforcing. Just as a social engineer wanted to eliminate social sanction for defying a bad norm, they may want to introduce social sanction for defying a good norm. This is one way to solve collective action problems.

Sanctions aren’t the only way to solve collective action problems. There are other opportunities too, but the literature on collective action is deep and we won’t cover any more of it here. I hope to address it more directly in a future post.

Tools

That last section is the last of the core content in Norms in the Wild. This last section will just cover some tools it outlines as potentially useful for a social engineer.

Economic

One obvious option for changing norms (in particular, providing shared reasons for change) is altering economic incentives. However, this approach has several limitations:

• “The activity may be reframed as a monetary transaction, thus obscuring its moral significance.” (Bicchieri 2016)
• “If the price is too low, it may signal that the activity is of low import, discouraging engagement in it.” (Bicchieri 2016)
• “Finally, a monetary incentive may implicitly suggest that an inducement is needed, otherwise people would skirt the activity.” (Bicchieri 2016)

Legal

Alternatively, one could simply try to mandate adherence to a new norm with the force of law. This should provide strong shared reasons to change. However, “if a new legal norm imposes harsh penalties against an accepted social norm, police will be less likely to enforce the legal norm, prosecutors will be less likely to charge, and juries to convict, with the effect of ultimately reinforcing the social norm that was intended to be changed.” (Bicchieri 2016).

The efficacy of legal mandates for changing norms is strongly correlated with trust in formal institutions.

Trendsetters

Trendsetters may be useful for both providing reasons to change and altering shared expectations. As previously mentioned, the positive deviance approach suggests that lived demonstrations by one trendsetter may be worth a thousand exhortations from a social engineer. Additionally, if a trendsetter is not sanctioned or only sanctioned lightly, this may change observers expectations about the cost of their own defiance.

Trendsetters are likely to be:

Risk-insensitive

They may not much care about the possibility of sanction

Misperceive the risks of deviance

They may underestimate the likelihood or magnitude of sanction.

Norm-insentive

They may not have much esteem for the norm and what it represents

Media

Finally, edutainment is, reportedly, shockingly effective at changing norms.

Edutainment may be effective for some combination of the following reasons:

• It can change scripts and schemata by having representative and relatable characters experience events that diverge from existing schema.
• Characters can function as trendsetters.
• It can make alternative norms cohere with the existing local moral structure.