Social Norms and Relational Mobility

Social scientific theories and/or real-world contexts that your model is based on:

Group norms theory suggests that the attitudes that individuals hold are largely the result of socialization with other group members (Sherif & Sherif, 1953). Individuals see can norms as an indication of what behaviors or attitudes are normal and acceptable. Social norms influence one’s attitudes towards social groups, and this relationship has been studied extensively (Li, Liang, Wu, Lin, & Wen, 2009; Norman, Sorrentino, Windell, & Manchanda, 2008; Stuber, Galea, & Link, 2008). There is a strong positive correlation between such social norms and the outward expression of one’s attitudes and prejudices (in the populations that have been examined so far (Crandall, Eshleman, & O’Brien, 2002). The Justification-Suppression Model (JSM; Crandall and Eshleman, 2003) suggests that one of the reasons individuals may do this is to avoid social consequences for their attitudes, if such attitudes are counter-normative.

Research indicates that the exposure to an anti-prejudice norm does indeed result in less explicit prejudice, particularly when individuals feel a high normative pressure (Blanchard, Crandall, Brigham, & Vaughn, 1994; Monteith, Deneen, & Tooman, 1996; Periera, 2002). Similarly, norms for outgroup exclusion increase outgroup prejudice among children, while inclusion norms decrease outgroup prejudice (Nesdale, 2011). Attitudes are more susceptible to influence from immediate others when the general norms for a particular target social group are less clear (Zitek & Hibel, 2006). On the whole, this research indicates that individuals will shift their expression of prejudices to align with social norms.

It is important to consider this effect within the lens of cultural psychology. Shweder (1990) suggests a mutual constitution of culture and psyche. Individuals live in worlds that are populated by products of their own understanding of the world. The way they understand the world shapes their psyche, and their psyche shapes the way they understand the world and process cultural and environment cues. This mutual constitution shapes the way we view and form attitudes.

Knowing that social norms influence the expression of prejudice, and knowing that target social groups are subject to differing levels of perceived social acceptability of prejudice (with more disagreement for some groups than others), we can consider a crucial question: what variables may moderate the relationship between social norms and the expression of prejudice?

Relational Mobility

Relational mobility has been used to explain many different relational and social behaviors, from gift-giving (Komiya, Ohtsubo, Nakanishi, & Oishi, 2019) to similarity attraction (Schug, Yuki, Horikawa, Takemura, 2009). It is defined as defined as “the amount of opportunities people have to select new relationship partners in a given society or social context” (Yuki, et al., 2007, p.3). High relational mobility suggests an element of perceived discretion and choice in one’s interpersonal relationships; the individual is more abstracted from their network, and entering or exiting relationships is voluntary (Oishi, Schug, Yuki, Axt, 2015). Low relational mobility suggests that the individual’s network is perceived to be stable and inflexible; the individual is embedded in their network and cannot easily enter or exit relationships (Oishi et al., 2015). Relational mobility can vary between and within countries (Thomson et al., 2018).

Cultural context informs one’s degree of perceived relational mobility, which also influences social behaviors (Thomson et al., 2018). Research has demonstrated that relational mobility increases the likelihood that individuals will take social risks (Li, Hamamura, & Adams, 2016). In this work, researchers determined that individuals who were primed with high relational mobility primes scored higher on the interpersonal section of the Domain-Specific Risk-Taking Scale. They indicated that they were more likely to take interpersonal risks, such as disagreeing with friends or admitting they have different tastes to others. This can be considered norm violation; as such, those who are higher in relational mobility were more likely to respond favorably to items intended to assess the likelihood of future norm violation. This may also be considered interpersonally inharmonious, as per Kim and Markus (1999). However, participants were not given the opportunity to participate in any non-normative actions themselves, and no behavioral measures were included (Li, Hamamura, & Adams, 2016).

Further work has demonstrated that in societies with low relational mobility, individuals are motivated to follow norms to avoid rejection, as opposed to earning a positive reputation (Iwatani, Muramoto, Kasahara, 2016). The desire to avoid consequences is consistent with the JSM, a model of prejudice expression that emphasizes social norms as a crucial reason for why individuals alter the expression of their prejudice (Crandall & Eshleman, 2003).

To examine the relationship between relational mobility and prejudice, I will construct an agent-based model.

Agent properties:

Within this model, agents will represent people. Their color will range from a light peach shade to black; the higher the prejudice, the darker the agent will be. Agents can have two different shapes: when they are happy they will be squares, and when they are unhappy they will be X’s. Happiness will depend on whether they are surrounded by agents they deem similar to themselves in prejudice. The total number of agents will depend on the density the observer chooses.

The agent will have several characteristics that change over time. In addition to color and shape, they will have a location that can change (and will change based on a probability related to their relational mobility) and they will have a level of prejudice that can change. There is one characteristic that cannot change, and that is their level of relational mobility.

Initial location, initial explicit prejudice, and relational mobility are all randomly assigned.

Interaction rules:

In each tick, agents will assess their own level of explicit prejudice, and the level of explicit prejudice of each of their 8 neighbors.

If they have enough neighbors that are similar to their level of prejudice (the percentage of similar neighbors required and the definition of what constitutes similarity are parameters and will be described in the next section), nothing will happen. They will remain stationary, and their level of explicit prejudice will not change. These agents are happy agents

If they do not have a high enough percentage of similar neighbors (once again, the required percentage is a parameter and will be defined in the next section), they will be unhappy. Unhappy agents will move with a probability equal to that of their relational mobility. That is, p of moving when unhappy is equivalent to their randomly assigned level of relational mobility.

Agents that move when they are unhappy will undergo no other changes during that tick, besides moving location. Agents that are unhappy but do not move will adjust their prejudice to be more similar to that of their neighbors, representative of a normative effect. They will assess the average prejudice level of their neighbors and their own prejudice, and adjust their prejudice to be the average of those two numbers.

After this, the tick concludes. If there are still unhappy agents, the model runs again. If all agents are happy, the model stops.

These interaction rules are based on theories such as relational mobility and group norms of prejudice (see Figure 1).

NetLogo initial setup:

Upon setup, agents populate the field with a density equal to that of the density parameter set by the observer (this can range from 0% to 100%). They update their color to be consistent with that of their randomly assigned explicit prejudice level, and they update their shape depending on if they are happy with the composition of their neighbors or unhappy with that composition.

Their satisfaction with their neighbors is dependent on two parameters: tightness/looseness and percent of similar neighbors wanted.

Tightness/looseness defines what neighbors will be seen as similar by setting the range of prejudices around an agents prejudice that are defined as similar. For example, if tightness/looseness is 5, each agent will see neighbor agents as similar if they have prejudice between 5 points below their prejudice and 5 points above their prejudice.. If tightness/looseness is 20, each agent will see neighbor agents as similar if they have prejudice between 30 points below their prejudice and 30 points above their prejudice. Thus, tightness/looseness sets the range what neighbors will be considered similar.

Percent of similar neighbors wanted defines how many neighbors must be similar to the agent to be happy. If it is set to 25%, at least 25% of neighbors must fall within the previously defined range of prejudices for the agent to be happy. If it is set to 50%, at least 50% of neighbors must fall within the previously defined range of prejudices for the agent to be happy.

The default density is 65%, the default percent of similar neighbors wanted is 75%, and the default value of tightness/looseness is 10.

Simulation procedure:

I am planning to sweep three parameters: density (65% and 85%), tightness/looseness (10 and 20), and percent of similar neighbors wanted (25%, 50%, and 75%).

I am interested in three outcome variables: how long it takes the model to reach equilibrium, the mean of prejudice for all turtles, and the standard deviation of prejudice for all turtles. In this way, I can track how prejudice changes, how dissent increases or disappears, and how long it takes to establish satisfactory relationships with others.

I plot several variables as the model runs. I create several line graphs: the mean of prejudice for all of the turtles, and the mean change in prejudice for all turtles (meaning, how much change, on average, have turtles undergone between their original level of prejudice and their current level of prejudice), the standard deviation of prejudice for the turtles, and number of unhappy turtles, and the overall percent of similar neighbors. I also create two histograms, showing the distribution of prejudice and the distribution of the amount of change in prejudice turtles have undergone.

Simulation results:

Standard deviation:

Standard deviation is, to me, a marker of how much disagreement there still is on prejudice within the ‘society’ I have created within the model. A low SD of prejudice indicated everyone agrees quite closely; a higher SD of prejudice means that there is more disagreement, with some individuals having a lot more prejudice and some having a lot less prejudice than the average agent.

The effects of the three parameters on the standard deviation of prejudice are as follows: looser societies have a significantly higher SD, percent of similar neighbors wanted significantly affects SD wherein lower percentages of similar neighbors wanted have a significantly higher SD than higher percentages of similar neighbors wanted, and density has no effect on the standard deviation of prejudice (See Figure 2)

Mean prejudice:

In this model, mean prejudice is the average amount of explicit prejudice held by agents. When mean explicit prejudice is higher, the average agent has more explicit prejudice.

Neither density, tightness/looseness, or percent of similar agents wanted significantly affected the mean prejudice of the agents. With each model run, mean prejudice settled at around 50. I believe this is because the normative effect brought everyone closer to the initial mean prejudice. (See Figure 3).

Run time:

Run time is the amount of ticks it takes for the model to reach equilibrium. Equilibrium occurs when every agent is happy, meaning they have a sufficient percent of similar neighbors to satisfy them.

Tightness significantly affects run time. The tighter the society is, the longer it takes to reach equilibrium. The percent of similar neighbors wanted also has a significant affect on run time; it takes significantly longer to achieve equilibrium when 75% similar neighbors are wanted than when only 25% similar neighbors are wanted. There are no significant differences between 50% and either 25% or 75%. Density has no significant affect on run time. (See Figure 4).

Combinations:

It seems that within the model, the tighter and higher percent of similarity wanted makes it take longer it takes to reach equilibrium, and decreases the standard deviation of prejudice.

Patterns:

When running the model, I noticed an interesting pattern emerge. Particularly when the percent of similarity wanted was high, nearly all agents would adjust their prejudice to be close to the mean prejudice. However, regardless of density and tightness/looseness, when the percent of similarity was high, there would often be pockets in the model with just one agent who had no neighbors. That agent was typically highly mobile, and had a level of prejudice that was either much lower or much higher than the average. These isolated agents had no neighbors, so they had no effect on other agents, but their still retained non-normative views (see Figure 5).

Implications and future directions

This work was largely exploratory, and I did not have many pre-established hypotheses. I had initially expected that high levels of relational mobility would decrease the power of the normative effects—that those with high relational mobility would adjust their prejudice less to conform with their neighbors to achieve happiness. This is, to an extent, a pattern I saw: sone isolated agents would move until they had no neighbors, retaining their original prejudice level because they were too different from most other agents, and these agents had high relational mobility. However, there were plenty of other agents who had a high level of relational mobility but which did not and up in isolated pockets, and instead conformed to their neighbors.

These results suggest that, in a society with varying levels of relational mobility, most individuals will conform to a middling level of prejudice—even those who start out with extremely high or low prejudice. A few mobile agents will remain outliers and retain either an extremely high or extremely low level of prejudice. The average prejudice levels do not change, and certain parameters force closer adherence to norms and the opinions of others.

My future directions involve several new avenues of exploration. First, I’d like to try systematically varying relational mobility. Within this model, it was randomly assigned to all agents. Thus, I created a model where I was able to see how a society with different levels of mobility might experience changes in prejudice and disagreement. However, relational mobility was not manipulated, so I was not able to see how different levels of relational mobility would impact mean prejudice or the standard deviation of that prejudice. In the future, I’d like to do this. I don’t want all agents to have the same level of relational mobility, but I would like to find other ways to systematically vary it. Would this make the normative effect stronger? Would prejudice increase?

I’d also like to change tightness/looseness. Right now, it’s a parameter of the entire space, and is the same for all agents. However, it’s unlikely in a real-life setting that this would be the case—difference communities are known to be tighter or looser. Keeping this in mind, I could change tightness-looseness to be a patch property, so that it changes for different communities/neighborhoods.

Finally, I am extremely interested in the few agents who remained prejudice outliers—finding a secluded pocket with no neighbors, and maintaining an extremely high or extremely low level of prejudice. In my model, they had no effect on others. However, in a modern society we might expect them to still influence others who are no direct neighbors through social media. Doing this might increase (or decrease) the overall prejudice of other agents in the space, necessitating more movement or normative adjustments. I’d like to explore what this would look like.