The Challenges Associated With Complexity, Exponential Growth, and Feedback Delay

September 2, 2021

James Reason, the author of Human Error, writes about different classes of error—skill-based, rule-based, and knowledge-based. For our purposes, we would focus primarily on knowledge-based errors.  The Uppsala DESSY studies (especially the challenges associated with complexity, exponential growth and feedback delay on performance) are especially pertinent.

The outcome of this pandemic was determined by the actions of a myriad of people from POTUS and his advisors, to Federal Department Secretaries (including HHS and CDC) and their advisors, to national experts in public health and emergency response, to legislatures, to governors, to mayors, to local public health and medical leaders, to school boards, to educators, to the managers of nursing homes and long term care facilities, to hospital administrators, to nurse leaders managing day to day staffing and operations, to medical supply staff anticipating the demand for PPE and other medical supplies, to pharmaceutical manufacturers, mask and respirator manufacturers, to the owners and operators of critical infrastructure/essential services (including police/fire/EMS, pharmacies, food supply/groceries stores/restaurants, transportation/buses/trains/planes/taxis/Uber), to religious leaders and the operation of places of worship, to the behaviors of all Americans (young and old). Certainly some individuals had a disproportionate impact on the outcome. But what is universal about all of them is they are human and predictably many people underestimated complexity (oversimplifying the pandemic), underestimated exponential growth (how often did we hear that COVID seemed to be nowhere and then it seemed to be everywhere), and all of us suffered from feedback delay (relying on data that told us where the pandemic was, not where the pandemic is). That is simply the context. We ignore it at our peril. It will be like investigating an accident and somehow not taking into account that it was dark and raining heavily and the driver was distracted by the ever present iPhone. We all take this as a given. It is like an ether that envelops us, but because of that, it is so easy to ignore. Like the old saying, if you want to know what water is, don’t ask a fish.

The other aspect of real life is its stochastic nature. We all know this after years of driving and our history of accidents and non-accidents associated with identical lapses or errors like running a light or looking down at a phone, etc.  We have seen how two nurses can commit the same error (mixing up patient medications) with very different outcomes. In one case, the patient receives a medication of little consequence, while a second patient receives a medication that causes serious injury or death. We treat the events (including the investigation of the events) very differently, yet the mistakes were identical (caused by our dependence on human vigilance and a system for prescribing and administering medications that is anything but foolproof). In one case, a patient dies, a family is devastated. There is a lawsuit and perhaps a settlement. A nurse and maybe a pharmacist and physician are fingered and their lives are changed. In the other case, not much happens—no harm, no foul. The problem of course is that those minor or non-events are so much more common, but invisible. The serious events are rare but glaringly obvious. As a result we get a very distorted view of the system (including people and how they tick). We see the exceptions, and not the rule. And we naturally blame individuals, and ignore the system.

So let's talk more about complexity, exponential growth, and feedback delay.


Dave Snowden developed a framework for decision-making and sense-making. Snowden’s model for decision-making includes four domains: (1) known; (2) knowable; (3) complex; and (4) chaos. Over the years, he has changed the names of the known and knowable domains to clear and complicated and identifies the central area as disorder.

Here is a quick overview.

The Known: The Known domain is where cause is clearly linked to effect.  This is the realm where linearity prevails—act in a certain way and the outcome is known and predictable.  It is the land of best practice.  As a consequence best practices are identified and standard operating procedures (SOP) are developed. In this sort of environment, we sense, categorize and respond. In this realm you can tell people what they ought to do—action is clearly linked to outcome. In this realm things are predictable. In many ways we operate as if we only exist in this Known realm. Our efforts related to issuing policies and operating procedures or operating plans work for some things, but not all. Snowden's point is that these sorts of practices work best when cause and effect are clearly linked (I do "A" and "B" reliably results). But when "B" doesn't clearly result from "A", these won't be the most effective strategies. Yet consider how often we utilize this approach. In the Known realm, leadership tends to a feudal model with budget replacing land. As Snowden notes, "…we impose order through laws and practices that have sufficient universal acceptance to create predictable environments."  All we need to do is provide information after we've had a chance to categorize it and we know what needs to be done.  (That sounds a lot like a bureaucracy.)

The Knowable: This is the domain of good practice.  As Snowden notes, "We do not yet know all the linkages, but they can be discovered.  This is the domain of experts, whose expertise enables us to manage by delegation without the need for categorization."  But he warns, "The very thing that enables expertise to develop, namely the codification of expert language in turn leads inevitably to entrainment of thinking. Exhortations to remain open to new ideas are unlikely to succeed.  Management of this space requires the cyclical disruption of perceived wisdom."  This is the realm in which cause and effect are linked but distant over time and place.  As a result causality isn't always easy to discern. Causality isn't known but it is knowable—if we have enough resources, capability and time. In this realm we sense, analyze and respond. Our process of analyzing is at its very core, reductionist.  It is the very basis for much of our scientific work—we reduce the problems (or linkages of cause and effect) to smaller more narrowly confined areas.  We control conditions (context) and reduce as many of the confounding variables as we can think of.  And if we accept that context may play an important role in determining response or effect in this domain, it becomes pretty obvious that one cannot simply provide an SOP or policy or even a best practice that could account for any context.  Yet adapting some sort of guidance to real life is actually something that human beings are very good at doing on their own—“they resolve conflicts, anticipate changes, cope with surprise, work around obstacles, close gaps between plans and real situations, detect and recover from miscommunications and misassessments.”  This is at the heart of their work practice.  In many ways we already know this about our plans and policies.  So in this realm, "we sense and respond based on our expert understanding of the situation, the leadership models are oligarchic requiring the consent of the elders of the community."

The Complex: This is the domain where patterns emerge. Snowden notes, "We need to identify the early signs of a pattern forming and disrupt those we find undesirable while stabilizing those we want. By increasing information flow, variety and connectiveness either singly or in combination we can break down existing patterns and create the conditions under which new patterns will emerge, although the nature of emergence is not predictable."  These patterns are the emergent properties of the interactions of the various agents. In this complex realm we must first probe, then sense and respond. Leadership in this domain can't be imposed—it emerges based on natural authority and respect. As Snowden notes, "but it is not democratic, it is matriarchal or patriarchal."

The Chaotic: This is the environment lacking any order. No patterns emerge.  As Snowden remarks, "Chaos represents the consequence of excessive structure or massive change."  This is the domain that requires crisis management.  In this realm the most important thing to do is act.  It is a bit like a fish that has been pulled from water and is lying on the ground.  The fish isn't sure with each flip or flop if it is getting nearer or farther from the water.  All it knows is that it must act.  There is no point wasting a lot of time analyzing or categorizing information—there are no patterns, there is no causality. In this domain we act, sense, and respond.  Snowden notes, "Leadership in this domain is about power; either power of tyranny, or that of charisma. Both models impose order."   Yet as Snowden notes, "(Chaos) is one of the most useful spaces, it provides a means by which entrainment of thinking, the inevitable consequence of expertise can be disrupted by breaking down the assumptions on which the expertise is based."  Different leadership styles are preferable and tend to emerge in the different domains.

One would argue that pandemics reside in the complex domain.  The behavior of complex systems is exquisitely sensitive to initial conditions (it is why we looked so closely at the timing of interventions during the 1918 pandemic). And the behavior of complex systems is very sensitive to tiny perturbations (small inputs = large outputs).  It is why these simple comparisons of states or countries were so meaningless.  Moreover, very simple rules can result in complex patterns (consider the 3-C’s in Japan—rules that could be easily operationalized by all Japanese citizens). Much of our scientific efforts are focused in the knowable domain.  In the complex domain you look for patterns. In chaos you just act.

During the past 20 months, how effectively have leaders recognized what domain they were in, and how effectively have they adjusted their leadership style and their response? That is where plan continuation bias (back to James Reason) comes in. It is just too taxing mentally for most people to change their strategies—especially when under significant stress.

Exponential Growth:

In everyday life we witness exponential growth—the growth of mold on a loaf of bread, the spread of a fire, interest on our credit card.  Yet we underestimate the consequences of exponential growth.  James Lawler put together a useful example—refilling Lake Michigan starting with a tablespoon of water and doubling the amount each day. By day 9, you add 1 gal. By day 23 a tanker truck of water is added.  By day 29, the lake has as much water as an Olympic sized swimming pool. On Day 58, the lake is about half full, but if you are standing on the original shoreline, you’d need a telescope to see the new shoreline, 17 miles away. On day 59 the lake is now completely full. And on day 60 you are dealing with a flood of the century. And it is the same way with outbreaks. It appears that disease was nowhere and suddenly it is everywhere.  There was exponential growth all along—it was just invisible to you until it wasn’t.  We underestimate exponential growth.

Data Lags and Feedback Delay:

Since June, Florida has publicly been reporting data on cases and deaths weekly. Each day, Florida reports the number of people that have died the day before (reporting deaths by the date the death actually occurred). A few times per week, Florida updates cumulative death totals (correcting prior date totals because of lags in the reporting of deaths). One might think that 20 months into this pandemic we would have a system for promptly and accurately capturing the numbers of daily deaths related to COVID. The vast majority of deaths would be occurring in hospitals so it wouldn’t seem difficult to capture all hospital deaths each day. Hospitals are fully aware of all deaths each day—it is simply a matter of determining whether the death was related to COVID or not. Recognize that deaths are a lagging indicator—time from disease onset to death is on the order of 2 weeks.

So look at Florida. Shown are several dates when Florida updated its cumulative death totals (Aug 13, 17, 19, 23, 30, and Sep 2). Below shows a table and a graph of the data.

On Aug 13, Florida reported 29 deaths. By Sep 2 (more than 3 weeks later), the death total for Aug 13 was estimated to be 247. That likely is still an underestimate of the true numbers of deaths that occurred on Aug 13. It will likely take another week or two until the full number of deaths in known. You can look at the graph and pick a date and realize how our picture of how severe this wave is (in terms of how many COVID deaths occurred on a given date) increased dramatically over time (assuming we could only look at the changing data up until that date). The outbreak itself wasn’t becoming more severe; our perception of severity was changing. Those patients had already died.  We were just unaware.  If we had perfect knowledge about all deaths (say we knew there were about 250 deaths in FL on Aug 13), would leaders have acted differently—with more urgency?

Could you imagine fighting a war and not knowing your losses for a 1-day battle until a month later? You see the same thing with case counts. Only it is pretty easy to distinguish alive patients from dead patients. It is another matter to distinguish infected from non-infected individuals. We are aware of the tip of the iceberg in disease outbreaks. That is called case ascertainment. Case ascertainment was shockingly low during the early part of the pandemic. The question are leaders and decision-makers aware of these data lags and do they adjust or do they just follow the numbers and assume the numbers reflect current reality?

Join our Team

We are actively hiring and would love to connect with talented individuals that want to be on the forefront of change. Contact Us.