As a Lean advocate, I used to be a little skeptical about the notion of “complexity thinking”. I thought it was just another new label on the same old bottle of wine. I am already familiar with (and a big fan of) the writings of W. Edwards Deming, Russel Ackoff, Peter Senge, and Derek and Laura Cabrera, who have all given me an appreciation for thinking about (complex) systems. Additionally, complexity thinkers are generally dismissive of Lean, writing it off as an industrial engineering discipline that is only good for mass manufacturing operations, and so does not pertain to complex situations. This, of course, is a gross misunderstanding of Lean.
Nonetheless, I started reading a little more about the Cynefin[1] decision-making framework that has been getting increasing airtime in the business world (particularly amongst Agilists) over the last decade or so. My initial impression was that it was a bit too abstract and academic, with its creator, David Snowden, using words like “ontology” and “epistemology” a lot (note: I love big words and abstract concepts—but I’m also suspicious of people who might use them as a way to put a veneer of intellectualism over shallow ideas). It’s not entirely new, I thought, but it does add some ways of looking at systems that could be useful.
And then, a few days ago, when I was reading about how some Americans were not in favor of using a stay-at-home lockdown as a temporary mechanism for “flattening the curve” of the spread of COVID-19, the framework suddenly seemed eminently helpful and practical.
As I write this, the number of confirmed COVID-19 cases continue to double every 2-3 days in the US, Canada, and in many other countries around the world. On March 27th, the number of cases in the US surpassed 100,000[2]. Only 13 states had strict lockdown measures in place, some had some moderate restrictions (such as large gatherings), and some states had no measures in place at all. Those trying to downplay the situation, and claiming that lockdowns were unnecessary, were saying the threat was “blown out of proportion”[3]. After all, they argued, the USA has 325 million people, so 100,000 is only .03% of the population. On the same day of March 27th, the total number of deaths reported by the US was 1701, so the mortality rate, using the most straightforward of measures[4], is around 1.7%. Even less of a big deal, right? Annualized, the 1701 deaths between February 29th (when the 1st COVID-19 death in the US was reported) and March 27th, amounts to 22,106, or only 1% of the total annual deaths in USA.
The anti-lockdown camp compared this paltry 1% to the number of annual deaths caused by car accidents (1.3%) or the “regular” flu/pneumonia (2.0%). Car crashes currently kill more people per day than does the coronavirus, “so should we stop driving too?”, they asked indignantly. They proffered graphs like Figure 1.1 below, using a “Pareto”-style distribution of various causes of death, ranked from highest to lowest. Why lock down the country, causing great harm to its economy, not to mention huge inconvenience to its citizens, over a measly 1% death rate? If we want to prevent deaths, we’d be far better off focusing on heart disease and cancer, no?
Figure 1.1 A Pareto diagram of Causes of Death in USA (source: Medical News Today)
The whole problem with this argument is that it assumes that the spread of the virus is in a stable state, where the number of confirmed cases, while it might vary day-to-day, is, on average, neither growing nor shrinking. If plotted on a graph, a stable situation is what statisticians like to call “in control”, meaning that it varies, but within upper and lower boundaries (delimiting the “common cause” variation)—see Figure 1.2. The make projections, this argument further assumes that this stable state will not change significantly—for the better or for the worse– in the future, so that we can take today’s numbers (around 61 deaths/day) and extrapolate them out for 12 months to come up with a total of 22,160 per year.
Figure 1.2 A graphical illustration of a stable situation, where the number of daily deaths varies from 51 to 71 and has an average of 61.
Unfortunately, the COVID-19 situation is far from stable. As you read this, I’m sure the number of confirmed cases, and the number of deaths, has increased exponentially since March 27th. Despite lockdown countermeasures in some states, the cases have continued to double every 3 days or fewer. The function is not linear, it’s quadratic:
While this may look like complicated math, it’s not: y (t+n) is simply the number of cases in n days from today(t). x (t) is the number of confirmed cases today. If we want to estimate 30 days out from March 27th, plug in the numbers and do the math:
Based on the fact that the US has confirmed 100,000 active cases on March 27th, our best guess, if various countermeasures (lockdowns, testing and quarantining the sick, hand washing and disinfecting routines, physical (social) distancing) have no effect, is that the US will have 102,500,000 confirmed cases 30 days later, on April 26th. Assuming the mortality rate remains at 1.7%, that means over 1.7 million deaths, or a daily average rate of 30,000. Yes, you read that right: 30,000 deaths a day. Now our Pareto chart looks like Figure 1.3. Suddenly, in just 30 days, our insignificant outlier at the tail end of Figure 1.1, representing 1% of all annual deaths, has jumped off the chart, to occupy 89% of all annual deaths. And the total number of actual deaths will have increase seven-fold in 2020 (19.5M) compared to 2019 (2.8M).
Figure 1.3 A Pareto diagram of Causes of Death in USA, 30 days later.
Before you panic too much, remember that this is an annualized projection 12 months into the future, and it is based on the assumption that the current measures already in place will have no effect. This is not likely to be the case. Additionally, in a couple of weeks, if there is still insufficient containment of the virus, I doubt the general public and the various levels of government will continue to do what they are currently doing and allow the pandemic to have that kind of widespread impact for 12 months running. They will enact stricter measures. But the point is this: complex, nonlinear, dynamic situations, such as viral outbreaks, when unmitigated, can get really, really bad, really, really quickly. And they will get exponentially worse every day (see Figure 1.4). Pareto charts are useless in such a situation.
Figure 1.4 The exponential trend of COVID-19 (source)
Using the Cynefin framework’s language, where situations can be classified as either Obvious, Complicated, Complex, or Chaotic, the COVID-19 virus is a good example of a situation in the Complex domain spilling over into the Chaotic domain. It’s complex because it has billions of interrelated elements (from human interactions to viral particles). There are many subsystems interacting, including political, economic, social, medical, biological, and so on, all contributing to the spread and/or containment of the virus and its wider effects on society. There is no singular “root cause” (as can often be the case in the Complicated domain) that can be simply fixed or corrected by an act of congress or a medical intervention– or at least not one that is currently understood and available to the public. It is a dynamic, unpredictable situation, meaning it is highly variable, changing day-to-day. It is, as I hope the Pareto charts above show, a fast-moving target. And that’s because it’s not at all “linear” or stable. Its growth is exponential, so that while today it is killing a relatively small percentage of the population, we know that, if left completely unchecked, it could quickly grow beyond the capacity of the medical system and become a huge killer in a very short time.
Or it might not. There are, at this time, so many unknowns: How many people actually have the virus? How many are asymptomatic and unwittingly spreading it? What are the factors, besides age and pre-existing conditions, that cause it to be severe in 20% of the cases but not in the other 80%? What will work and not work to contain the spread and the mortality rate? When will it peak? When will a vaccine or cure be found? Will we run out of ventilators, hospital beds, and medical staff? Because there are so many unknowns, politicians and the general public cannot act with any certainty. But, because it is a matter of life and death, we have to act.
Hence the complexity of the situation crosses the threshold into the Chaotic domain, which is often characterized by urgency and crisis. In this domain, people should act first, based on the limited amount of information they have, to stanch the proverbial bleeding and stabilize the patient. This is what seems to be happening right now, where some states (and most provinces in Canada) have imposed strict orders to stay home, to see if it contains the spread. Whether these decisions are good or bad can only be determined in hindsight, as cause and effect relationships emerge later on. Sooner or later, the COVID-19 situation will stabilize into some sort of discernible pattern, and it will move back to being more solidly situated in the Complex domain.
One of the main teachings of the Cynefin framework is that different thinking and decision-making approaches need to be used in each distinct domain to achieve optimal outcomes. Pareto charts, a tool frequently used by process improvement practitioners like me (the so-called Lean Six Sigma “belts”, etc.), are useful in the Complicated domain when a process is relatively stable over time (and there are known causes for the few times it might deviate outside of its boundaries). The range of variation may be far wider than your customers would like it to be, but at least it follows a fairly consistent pattern of disappointing your customers, and so can be improved. The Pareto chart is not the right tool when you have complex process characteristics that look like the “hockey stick” of Figure 1.4, because things are moving too quickly, there are too many variables at play, and the information is incomplete.
I think this is why complexity thinkers dislike Lean Six Sigma types. Our tribe tends to contain too many toolheads, who use the same tools for nearly every business problem they encounter, without much thought to the context, or domain, they are operating in. And I agree with the complexity thinkers! I disagree that we should throw the baby out with the bathwater though. Lean thinking, when practiced from a principled and systemic point of view, is very much suited to dealing with complexity. But I’ll save that discussion for a future date.
In the meantime, since we’ve got a viral outbreak to deal with, what should we do instead of a pointless Pareto analysis? First of all, since many aspects of the situation are chaotic, the approach should be fairly autocratic. Command-and-control leadership, which is largely ineffective in normal business operations, is called for in chaotic situations. Strong, swift, top-down measures should be enacted, even if it means stripping individuals of some of their civil rights: a total nationwide lockdown (temporarily) in the US and Canada, widespread and rapid testing everywhere (e.g. at airports, grocery stores, pharmacies nationwide), enforced quarantine of known cases (including digital tracking through cellphones, credit cards, and drone surveillance), rapid deployment of any and all supplies and human resources to the frontlines of the medical system, short-term economic relief measures to allow everyone to pay their bills and obtain minimal, basic necessities for survival.
Current evidence suggests that places that have fairly autocratic governments, like Hong Kong and Singapore, or those more liberal governments that have imposed more autocratic measures during this crisis, like South Korea and Japan, have done a much better job at containing the virus than places like Europe or North America.
For the aspects of the situation that remain complex, we should do as every good Lean thinker does: break down the problem and then experiment with countermeasures. Improvise, adapt and innovate towards increasing medical capacity, including protecting the frontline healthcare workers from health risks and burnout. Optimize public disinfection methods. Help redeploy factory machinery and labour to make ventilators and other needed supplies. Run research labs better to search for a vaccine. Improve the speed of vital supply chains. Experiment with removing the waste of bureaucratic procedures (without sacrificing human safety) to see how fast we can get new drugs approved, or shipments of needed supplies delivered. We don’t know ahead of time what will work and what won’t, but rapid, safe-to-fail experiments (a cornerstone of both the Lean and Cynefin approach) will allow us to find out in the shortest amount of time.
Once we find out what’s works and what does not, and then keeping doing the activities that work, the virus will come under containment and start crossing the threshold from the Complex to the Complicated domain. Then the experts can now effectively analyze (it will be safe to pull out the Pareto graphs again!) and respond: provide the best treatment to all those with severe symptoms, ensure as many recover—and as few die—as possible, continue to enforce the measures that are working at prevention and containment, determine when to say it’s safe to resume some normal activities. Meanwhile, the politicians and economists can manage the financial situation, to stabilize and resuscitate the economy. Virologists will look for root causes in the transmission of viruses from animals to humans and seek countermeasures to prevent such transmissions in the future.
But the dilemma the experts from the WHO and the CDC face right now is the uncertainty of complexity. They simply don’t know enough to predict the future with any confidence or know what to do with any certainty. If it emerges later that they were overly cautious, they will be blamed for being alarmist. If they were not cautious enough, they will be blamed for being incompetent. Maybe the anti-lockdown people will be vindicated. South Korea and other Asia countries have contained their situation without lockdown, but they acted in other ways swiftly and much earlier[5]. Personally, when it comes to potentially fatal health matters, and we are late in the game now, compared to Korea and Japan, I would prefer our governing officials to be overcautious and then have a better-than-expected outcome, rather than be cavalier and have a worse-than-expected outcome. It may come with some personal and economic sacrifice, but such is the nature of acting within the uncertain and ambiguous domain of Complexity.
[1] http://cognitive-edge.com/videos/cynefin-framework-introduction/
[2] https://www.worldometers.info/coronavirus/country/us/ (accessed March 28, 2020)
[3] https://www.npr.org/2020/03/17/816501871/poll-as-coronavirus-spreads-fewer-americans-see-pandemic-as-a-real-threat
[4] Calculating mortality rates is way more complicated than it might seem. Since COVID-19 gestates up to 14 days before manifesting any symptoms, some people are using the total number of confirmed cases as of 14 days prior as the denominator, which drives the percentage up. The number of confirmed cases, however, is highly dependent on the amount of testing being done, which currently is increasing from day-to-day. The numerator too, is hard to determine because often people who die with COVID-19 have other pre-existing conditions and it is near impossible to say how much one or the other contributed to the death. Layer onto that the variation between countries, since countries with older populations have more deaths from the virus. Access to healthcare is yet another country-based factor.
[5] https://www.npr.org/sections/goatsandsoda/2020/03/26/821688981/how-south-korea-reigned-in-the-outbreak-without-shutting-everything-down
keneakin.ca 