US education spending runs around $1T a year, roughly 7% of GDP, underscoring the high importance we place on it. In a field otherwise characterized by widespread disagreement virtually everyone agrees on one thing: this massive investment produces unacceptable results. Delivery of education competes for funding with education reform and improvement initiatives. Delivery suffers, the improvement efforts produce disappointing results, spawning new improvement programs and more pressure on delivery.
Dissatisfaction with education outcomes and reform efforts are nothing new. US Education has been at Level Orange for 50 years and more. After lots of arguments, lots of theories, experiments, programs, campaigns, and evaluations we’re more or less where we started across a broad range of measures.
Why is education so resistant to improvement? It’s not because we’re evil or stupid. The problem is complexity.
Education is a complex domain. Outcomes depend on the actions of a huge number of people, each of whom has their own intentions, motivations, and desires. These change frequently and unpredictably, influenced by the behaviors of each other and events that have little or no direct relationship to education concerns. How will people react to a given change? The uncertainty of answering this question is intrinsic and unavoidable.
Rather than a line of dominoes, where one event causes exactly one event and itself results from one event, many events might contribute to one event, which in turn might influence many other events, connected in a network of self-adapting feedback loops. As a result cause and effect are intrinsically difficult to understand in complex systems.
These patterns of interactions might eventually settle down into what appears as a stable system that produces expected outputs for given inputs, but frequently the system is “metastable,” where a small change in inputs can make large changes in overall behavior, producing different outputs or perhaps becoming highly unstable, resulting in hysteresis (wild fluctuations), runaway, or collapse. Small changes can create large scale shifts in behavior, and since there are fewer better states than worse ones, these large shifts are often for the worse. They also tend to be one-way, lacking an ‘undo’ button, making it impossible to return to the previous state.
As a result, when we try to improve education outcomes we’re shooting in the dark, no matter how confidently we advocate our approach. Particularly when the initiative is large scale and driven top-down.
We’ve learned a few valuable tricks to help us deal with large, complicated systems. One of the most important is carving the system up into a number of smaller subsystems. For example, when we design a car we break the work down into power train, electrical, suspension, and so on. Bringing them all together into an automobile design requires making a series of choices that generate constraints and requirements for each subsystem, but we have confidence that bringing them together will provide a complete transportation solution.
Similarly we can factor a complex system into a number of subsystems. But it’s far harder to define those subdomains as cleanly for complex systems as we can for simple and even complicated ones. We lack the ability to describe the complete system in a way that assures subsystems don’t overlap in some cases and leave gaps in others.
The education system varies widely, with no broadly adopted design specifying modular subsystems and a standard recipe for connecting the pieces. Lacking clear boundaries and defined interaction patterns generates confusion, redundancies, breakdowns in handoffs between subsystems, and guarantees good-faith efforts unintentionally working at cross purposes.
The varieties of approaches means that expertise in one subdomain doesn’t necessarily transfer well to similar-but-subtly different subdomains. Experiments are difficult to replicate and learnings hard to share.
In the face of these challenges despair seems justifiable, but all is not lost. In many domains of human endeavor progress depends on sophisticated design insights, even in the face of complexity. An abstract model of the system enables reasoning about it, and running simulations to gain insight into the potential impacts of proposed changes. It allows identifying dependencies across subdomains, potentially addressing non-obvious constraints that would negatively impact an otherwise beneficial change. It emphasizes the divergent perspectives on whether a prospective outcome would be beneficial or harmful. It enables translating a given solution to the varying circumstances across education subdomains (geographic, economic, target student population, etc.).
Attempting to apply traditional design methods to large complex domains fail in a few typical patterns. Their emphasis on understanding processes is incompatible with domains characterized by lightly constrained participants. The descriptive focus on how things are done bogs down under the volume and the variety of actions taken. A detailed description of a subdomain takes inordinate effort to create, additional effort to adapt for each slightly different instance, and provides little leverage independent of designs for adjacent subdomains. Only when the complete system is exhaustively described can the design be used, but the required up-front investment in time and effort before achieving any useful result virtually guarantees the design will never be completed.
A useful design approach must provide value quickly, allow a balance between breadth-first and depth-first analysis, and provide incremental value for incremental investment.
To accomplish this we’ve developed a design method that focuses on what and why, not how, which provides insight without bogging down in details. It places particular emphasis on information, asking this question: “What information is essential for resolving a particular situation?” It codifies interactions between subdomains by identifying the authoritative source for each information element, defining the required exchanges between subdomains. It provides for technology-independent descriptions that map to more detailed designs by applying choices and constraints, providing guidance to applying technology without mandating it.
Applying this design method to the education domain makes improving education more tractable. We’ll explore that in more detail in subsequent posts.