The Need for a Systemic Understanding of Inequality

Human well-being as measured by social connectedness, income, wealth, health status, life expectancy, or life satisfaction varies substantially across people and places.  Some people live socially integrated lives while others live in isolation; some people earn millions per year in  salary while others live in poverty; and, some people can expect to live healthily into their eighties or nineties whereas others may be fortunate to reach seventy (Gee and Ford 2011).

Here, I provide a brief introduction to systemic effects by constructing simple models and explcating their logic to show the systemic thinking adds value to our collective understanding of inequities.   Racial disparities in the United States and other countires provide powerful examples of systemic inequality.  Evidence shows that when people view racial inequalities as due not to implicit biases but to systemic effects, i.e. structural racism, that they become more likely to advocate reform (Rucker and Richeson 2021).  

Compounded Disadvantages and Systemic Effects

While we should expect some variation to arise due to differences in natural capacities,  effort, luck, and circumstances, the extent of variation and its spatial concentration within particular communities reveals the presenceof compounded disadvantages that limit the prospects for many  (Perkins and Sampson 2015).   Compounded disadvantages  create systemic effects.  Each disadvantage reinforces others: an unsafe environment contributes to poor health, which in turn contributes to worse performance in school; high school dropouts are more likely to be unemployed and to lead less healthy lifestyles.  In addition, people suffering from poor health and poor economics prospects are less likely to place trust in government and social services, contributing to a system of forces that make life difficult. 

The compounded disadvantages are interdependent not additive, a distinction that must be front of mind if we hope to reduce inequalities.   In a complex system, single isolatable interventions will not lead to improvements.   Evidence from fifty years of randomized controlled trials testing policies to reduce crime reveals that few prove replicable and those that do have effects of small magnitude at best (Stevenson 2023).

To explain how inequality emerges within a system and how that produces robust, systemic inequality, this site relies on a collection of simple models to demonstrate how inequality that emerges from systemic effects differs from inequality that is the sum of separable effects and describes some of the component processes that cause inequality to emerge and how through mutual reinforcement they often render single policy solutions ineffective in the long run.

The Use of Models

The models are  intended to elucidate the constituent processes that produce disparities, demonstrate the mechanisms through which those processes reinforce one another, and highlight the non intuitive, that is to say nonlinear, ways these processes aggregate to produce robust inequality.  

The purpose of these models is to help us clarify our thinking.  They should be read and enjoyed as such. They were not designed to explain data, predict outcomes, or design policy interventions.   They are not calibrated in any way to data.   A deep understanding of inequality requires a many model approach informed by data and history (Page 2018).   The animations built from the models are intended not as stand alone explanations but a collection of insights that enhance our general understanding.

The models are agent-based computational models.  Such models explicitly represent each individual in a population, as opposed to say a mean value for a population.  The animations that you see are computer simulations that iterate through a series of steps.  In each step, an agent may or may not change its status in response to other agents of the environment.  The configuration of agents changes over time as the members of the population respond to their environment.   The models track the well-being of individuals over time.  The model represents well-being in two dimensions:   socio-economic well-being and health status.  Each is measured on a scale from zero to one hundred.  

Systemic Compounded Disadvantages

Systemic compounded disadvantages differ from additive (separable) disadvantages because the interactions among negative forces imply that the inequality that emerges will be robust, that is, difficult to undo.   As a result, policies aimed at single dimensions or that lack sufficient resources, commitment, or engagement may prove inadequate.  In addition, policies that appear to work in the short term may fail in the long term as the systemic forces eventually dominate.

The socio-economic status and health of a community of two hundred people can be represented on a scatterplot, two of which are shown below.  Each red dot represents an individual.  The larger blue dot represents the community means (averages) on the two dimensions.  The two means provide a crude measure of whether the community is thriving or struggling.   The dispersion in the dots captures the differing circumstances of individuals.   Greater dispersion implies more inequality.

Scatterplots allow us to see both means and dispersion.  A thriving community will be represented by individuals and a mean located in the upper right as in the diagram on the left.   In contrast, a struggling community will be represented by individuals and a mean in the lower left as shown in the figure on the right.  Notice that even in the thriving community, some individuals have quite low well-being, while in the struggling community, some individuals have above average health status and socio-economic  well-being.

 Examples of a Thriving Community and a  Struggling Community

To generate the diagram on the left, the population is normally distributed with a mean value of seventy-five on each dimension with a standard deviation of fifteen. In that diagram on the right,  the mean value is set to twenty-five with a standard deviation of fifteen. 

Why Emergent Systemic Inequality Differs From Additively Separable Disadvantages

Educational and public institutions, policies, business opportunities, public health offerings, environmental quality, policing and many other features of a community contribute to whether a community thrives or struggles.   In a separable effects model,  an individual's health and socio-economic well-being can be written as an innate or initial value plus or minus the sum of the various contributions of the environment, the community, school quality,  public spending, and institutional support. 

The animation below shows a community that encounters negative additive effects on both health and socio-economic well-being.  The green path represents the trajectory of a single, randomly chosen individual.   Note how  green dot and the entire population move to the lower right.   They struggle.  

Two Views of Inequality: Separable and Systemic

Separable Effects

To generate this animation, in each iteration the health status and the socio-economic well-being of each individual decreases by a fixed amount but all effects are separable additive.

Systemic Effects

In a complex systems model of inequality, the effects on a community interact with one another.  Health influences school performance.  Poor school performance contributes to increases in crime.  Crime, in turn, increases stress, lowering school performance, producing a vicious cycle.    The animation below shows the degradation of a community's health and socio-economic status in model that includes a variety of interdependent effects shown in the diagram on the right. 

To generate this animation, at each iteration the health status and the socio-economic well-being of each individual decreases by a fixed amount in a complex systems model with systemic effects.

In watching this animation, notice that the path of the blue dot follows a similar path as in the previous animation with separable effects.  In both animations, average health and average socio-economic well-being decline.   Yet, what occurs within the population differs markedly across the two models.    These animations demonstrate why the mean may not be that meaningful when evaluating a  complex system.

Deconstructing Inequality Using Complex Systems Methodology

The complex systems model of inequality exhibits four phenomena not present in the separable effects model. 

Heterogeneous Effects: Some individuals improve on both dimensions even though the majority do not.  

Nonlinear Dynamics:  The individual represented by the green dot takes a curved path.  Following other individuals reveals that some even alternative between improvements and declines in socio-economic well-being.

Clustering: On the socio-economic dimension, the individuals cluster into three groups: one high performing, one moderate performing, and one low performing.

Diagonalization (Correlation): The distribution tends toward the upper sloping diagonal implying a correlation between heath status and socio-economic well-being. 

These various phenomena can be explained (in part) by considering the micro-level processes that underpin the model.    Even though  the aggregate effect of these processes will not be sum of the individual effects as in the separable (additive) effects model , we can still glean insights into the workings of complex systems by isolating individual forces and structures and then building up to a more complex model (Miller and Page 2008).

The Building Blocks of Emergent Inequality 

The inequality model shown above is an example of a complex adaptive system.  Such systems of heterogeneous actors who interact with one another (often through networks) whose behaviors depend on the actions of others.  The collective actions produce aggregate or emergent phenomena that also influence individuals, such as when poor performing students drop out of school and contribute to levels of violence which produce stress for those individuals still in school.  Complex social systems, which consist of people and institutions, as opposed to say neurons, termites, or oceanic flora and fauna, often include common component processes, or what might be called building blocks.

The inequality model includes the following building blocks:

Interdependencies & Feedback Loops

The first building block, an interdependent effect, occurs when the value of one attribute affects the value of another attribute.  The animation on the left shows the effect on the population when individuals' health levels influence their socio-economic well-being.  The model assumes that the socio-economic well-being of an individual with health status above fifty increases while the socio-economic well-being of an individual with health scores below fifty decreases.   The amount of the increase (or decrease) equals a constant times the individual's health status minus fifty.   A person with a health status of ninety will see a greater increases in socio-economic well-being than a person with a health status of sixty.

The individual below marked in green, who has a health level above sixty, sees a marked increase in socio-economic well-being.   As a result of the health effect on socio-economic status, the distribution becomes more diagonalized.  That diagonalization  produces a positive correlation between health status and socio-economic well-being. 

The animation on the right shows the effect of socio-economic well-being influencing health.   Those individuals with socio-economic well-being above fifty see an increase in their health level, and those with a socio-economic status below fifty see a decrease.   This interdependency also causes a diagonalization of the scatterplot 

To generate animation on the left,  at each iteration the socio-economic status of an individual changes by a constant times their health status minus fifty.   In the animation on the right, in each iteration the health status of an individual changes by a constant times their  socio-economic status minus fifty.   

Feedback Loops

Including both a  health effect on socio-economic well-being and a socio-economic well-being effect on health creates a  feedback loop as shown in the animation below.   This animation shows the paths of two individuals.  The individual in the upper right has an initial health status and socio-economic well-being above fifty.   Her good health improves her socio-economic well-being and high socio-economic well-being improves her health, a classic positive feedback loop, which amplifies each individual effect.   The amplifications that arise from positive feedback loops distinguishes the complex system models from a model with separable effects.

The other individual tagged as green follows a curved route.   Initially, because his health status exceeds fifty, socio-economic well-being increases.  At the same time, however, his low level of socio-economic well-being causes a decline in his health status.   When his health status falls below fifty, he then finds himself caught in a negative feedback loop: poor health lowers socio-economic well-being and low socio-economic well-being lowers health status.   This curved path would not be possible in a model in which the effects of health and socio-economic well-being were separable. 

To generate this animation,  at each iteration the socio-economic status of an individual changes by a constant times her health status minus fifty and her health status changes by a constant times her  socio-economic status minus fifty.   

Self Amplifying Loop

The next systemic building block, a self amplifying loop,  consists of a feedback loop for socio-economic well-being, in which an individual's value on an attribute changes as a function of that attribute's value.   Socio-economic status could create a self amplifying loop through multiple causal paths.   One route would be through executive function.   Evidence suggests that socio-economic status influences cognitive ability, as measured by executive function tests (Last et al 2018).  All else equal,  higher executive function should corresponds to higher socio-economic status.  Thus, higher socio-economic status would produce a feedback loop.  

Other paths exist as well.  People with high socio-economic status likely possess skills, networks, levels of trust in the fairness of educational and legal systems, as well as opportunities  for acquiring information and access.  All of these would produce self-amplification of socio-economic status.   The result will be a positive feedback (as shown by the green individual in the lower right). Individuals who are above average in socio-economic status will see their status increase.  Individuals with below average socio-economic well-being will move in the opposite direction (as shown by the green individual in the upper left).     The amplification loop can be thought of as a an agency effect causes an increase in the dispersion of socio-economic well-being.

Note that health status has no impact on the agency effect directly, but it have an effect once we build the entire model.  The effects will not be separable.

To generate this animation,  at each iteration both the socio-economic well-being of an individual changes by a constant times their health status minus fifty and the health status of an individual changes by a constant times their socio-economic well-being minus fifty.   

Clustering: Peer Effects

The next building block, a clustering effect, occurs when individuals match the behavior of other individuals like them or adapt their behavior to be more like that of other individuals.  In the model, this takes the form of a peer effect: individual adjust their socio-economic status to be closer to that of their peers.  Good students form friendships and adopt similar study behaviors.  Bad students to the same.   A peer effect creates a clustering of levels of socio-economic well-being, but on its own creates no clustering of health status. 

To generate this animation,  at each iteration an individual  randomly meets another individual and if their socio-economic statuses are similar, the first adjusts its socio-economic status to be nearer to that of the other individual.

The animation below shows what happens if we include both an agency effect (the positive feedback loop on socio-economic well-being) and a peer effect. 

Level Effects

A level effect causes an upward or downward shift of a variable.  The model assumes an environmental effect that produces a consistent decrease in health status over time as shown in the animation below.   This shift could result from the presence of lead in the water supply or high crime levels that create stress, which in turn has an effect on overall health.   Living in a neighborhood with high levels of crime or violence induces stress.  Stress  increases an individual's allostatic load, the physiological burden that a person confronts on a daily basis and leads to poor health outcomes (Guidi 2021).  

The System: Whole More Than the Sum of the Parts

When we put all of these pieces together, we have a model model that includes interdependence between health status and socio-economic well-being,  a positive feedback agency effect on socio-economic status, a peer effect that produces clustering and a negative environmental (level) effect.   The model also includes a tail effect as in which  increasing the proportion of lowest performing members of a community amplifies the environmental effect through an increase in stress. 

The mutual reinforcements include the following:

Agency effects create more individuals of low socio-economic well-being.  They then increase the tail effect and increase the negative environmental effect  lowering everyone's health status.

The interdependence of health and socio-economic well-being amplifies agency effects, which then cause the increased tail effect and environmental effect just mentioned. 

In addition to lowering health status, environmental effects lower socio-economic well-being through the interdependence of health and socio-economic well-being.  The negative feedbacks then become amplified through  agency effects.

Finally, peer effects  create clusters of similar socio-economic well-being.  The interdependence of health and socio-economic well-being and agency effects then prevent those people with  low socio-economic well-being clusters from thriving.  

The aggregate result as seen in the diagram below is a system in which many individuals wind up with low health status and low socio-economic well-being.  And, as shown next, these many mutually reinforcing processes create a system that proves robust to single dimensional or small magnitude policy interventions.

Policy Efficacy at Reducing Separable and Emergent Systemic Inequality

The complex systems inequality model, as just noted, produces robust, systemic inequality.  Attempts to reduce inequality require larger, more sustained efforts.  To see why that is the case, suppose, for example, we have a community with both low average health status and socio-economic well-being.   Though, transforming this community into a thriving community would require resources and will, it could be accomplished by policy  interventions that improve health status and socio-economic well-being.  Those policies could be implemented in sequence.  Schools could be improved, then infrastructure, then the local economy and so on.    Such an approach would work because of the assumption of separability - each positive and negative effect on health and socio-economic well-being operates independently of the others. The animation on the left show the effect of a health policy that increases health in each period by a small increment.  The animation on the right shows the effect of also adding a policy that increases socio-economic status each period.   In a separable system, both policies work as intended.

Policies to Improve Health Alone and to Improve Health and Socio-Economic Well-Being Assuming Separable Effects

To generate the animation on the left, at each iteration the health status of each individual increases.   In the animation on the right both the health status and the socio-economic well-being of each individual  increases by a fixed amount.

The Robustness of Systemic Inequality

In contrast, suppose that the identical policies were implemented within the complex systems inequality model that includes the feedback loops, interdependencies, level effects, and  peer effects.  The animation on the left shows the policy of improving health status each period by a moderate amount.  As can be seen in the animation, the intended positive effect is swamped by the negative feedback loop between socio-economic status and health status.  Note that is not true for everyone though.  A handful of individuals manage to get a sufficient health boost to create a positive feedback loop between health status and socio-economic well-being.  They thrive.

 The animation on the right shows the effect of supplementing the health intervention with a policy to boost socio-economic well-being each period.  The combined policies reduce the power of the negative feedback loop and even more individuals thrive by becoming part of positive feedback loop.   The two policies, therefore, work better than just one, but they remain largely ineffective.   Systemic inequality often proves robust to policies because of the interdependent forces. 

Policies to Improve Health Alone and to Improve Health and Socio-Economic Well-Being Assuming Systemic Effects

To generate the animation on the left, at each iteration the health status of each individual increases.   In the animation on the right both the health status and the socio-economic well-being of each individual  increases by a fixed amount.

Basins of Attraction and Escape Velocity

The low performing community finds itself stuck in a basin of attraction.   Imagine a ball sitting at the bottom of a bowl.  The policies intended to improve health status and socio-economic well-being push the ball up the side of the bowl, but only temporarily.  The gravity within the system, in the form of the level effects and negative feedback loops, pull the ball back down to the bottom of the basin.

Solutions to overcome systemic inequality require achieving escape velocity.  If we strengthen the policies by giving people even more of a boost on each dimension, more members of the population attain sufficient levels of health status and socio-economic well-being.  They escape the negative feedback loop and enter the positive feedback loop as shown in the animation below.   Notice that even in this case of large and sustained multi-pronged policies many individuals, including the individual marked in green, do not see an improvement in either their health or socio-economic status. 

The graph on the right captures the mean effect on the average of health status and socio-economic well-being as a function of the magnitude of the policy interventions.  In the separable model, the mean effect increases linearly with the magnitude of the policy intervention.  In the complex systems model, mean effects are non existent for interventions of small magnitude.  And, even as the mean effect grows, much of the system-wide change may be due to a subset of the population achieving escape velocity as opposed to the entire community improving.

Escape Velocity

The animation reveals a nonlinearity in the path to a thriving community.   With sufficiently large policies, over time more and more people enter the positive feedback loop and fewer get stuck in the negative feedback loop.   The dynamic produece a thriving socieity occurs at a different rate for different people. The combined effect is that some begin to thrive and others cease to struggle.  Over time, this second group also enters the positive feedback loop, and they too thrive.  

Concluding Thoughts 

This site should be read as an attempt to provide a gentle introduction to systemic effects, how they emerge, why they often prove robust to policy interventions, and why policy interventions may have early success but fail to escape the basin of attraction of the current equilibrium.   The site also points to the potential value of using models to identify interactions between the multiple effects that produce systemic inequality to guide policy interventions.

References

Gee, G., and C Ford. 2011. ``Structural Racism and Health Inequities: Old Issues, New Directions.'' Du Bois Review: Social Science Research on Race, 8(1), 115-132. 

Guidi J, Lucente M, Sonino N, Fava G, A. 2021. ``Allostatic Load and Its Impact on Health: A Systematic Review.'' Psychotherapy Psychosometrics  90:11-27. doi: 10.1159/000510696

Last BS, Lawson GM, Breiner K, Steinberg L, Farah MJ. 2018. ``Childhood socioeconomic status and executive function in childhood and beyond." PLoS ONE 13(8):e0202964. https://doi.org/10.1371/journal.pone.0202964

Miller J,  and  SE Page. 2008. Complex Adaptive Social Systems, Princeton University Press

Page, S. . 2018. The Model Thinker, Basic Books. 

Perkins, KL, and RJ Sampson. 2015. "Compounded Deprivation in the Transition to Adulthood: The Intersection of Racial and Economic Inequality among Chicagoans, 1995-2013." RSF: The Russell Sage Foundation Journal of the Social Sciences. Special volume on "Severe Deprivation in America,"

Rucker, JM, and JA Richeson,  2021. ``Toward an understanding of structural racism: Implications for criminal justice. ''Science,  374: (6565), 286-290. 

Stevenson, M. 2023 "Cause, Effect, and the Structure of the Social World." Available at SSRN: https://ssrn.com/abstract=4445710 

The models were constructed in response to a request from two medical doctors, Erika Newman of the University of Michigan and Victor Garcia of Cincinnati Children's Hospital, engaged in making practical efforts to apply systems thinking to improve community health as a part of Project Hope.  I thank Victor Garcia, John Miller, Jenna Bednar, and Ken Frank for helpful comments.