Policy-Making Through a Declustered Lens

Commentary

Canada’s public policies are built on flawed averages. Housing is a prime example. Since the 1960s, the national average ratio of house prices to after-tax income has more than tripled, from around 2.2 in 1965 to over 7.5 by 2024. At its peak in 2022, that ratio surpassed 9.0.

But these figures are heavily skewed. When adjusted to account for spatial concentration, we find that Toronto and Vancouver alone inflate the national average by several points. For example, housing in Thunder Bay still averages under $350,000, while in Toronto it exceeds $1.1 million. Yet both cities weigh equally in national policy metrics. The national average masks regional affordability and creates a false sense of universal crisis. When we layer in the rising share of income consumed by taxes, up from around 25 percent in the 1960s to over 43 percent by 2024, the picture becomes even more distorted. The public perception of housing unaffordability is not wrong, but the data used to define and respond to it is uncorrected for geographic bias. That’s a critical warning sign.

When governments build policies around these averages, such as setting mortgage eligibility rules, allocating funding, or triggering affordability measures, they create a ripple effect. Rural and small-town Canadians get hit with rules designed for places they don’t live in and problems they don’t have. Urban clustering creates statistical noise. Policy-makers then act on the noise, not the underlying signal.

This isn’t just a technical error but a form of systemic bias, one that can be measured, understood, and corrected. And in another field, it already is.

In the mining industry, decisions about where to dig can cost millions, if not billions. A wrong guess means financial loss, environmental damage, and wasted effort. To avoid this, mining companies use geostatistics to evaluate ore deposits. One of its key tools is called declustering.

When sample holes are drilled, they are often placed in the easiest or most accessible areas. That creates clusters, concentrated pockets of data. If you just average all the sample results together, your estimate will be skewed toward the high-density zones. You’ll think the whole area is richer in minerals than it really is.

Declustering assigns less weight to data from clustered zones and more weight to data from sparse ones. It evens out the influence and gives a more accurate estimate of the overall deposit. The math can range from simple to advanced, but the principle is consistent: reduce bias by adjusting for spatial density. If geologists didn’t do this, mines would open in the wrong places and the industry would lose its economic footing.

To see why clustering matters, consider a simple example from mining.

Imagine a flat area of land that’s 1,000 metres by 1,000 metres. We divide it into 50 by 50 metre blocks, each one metre thick. This gives us 400 blocks in total. To estimate how much gold lies below, we drill one test hole in each block. Almost every test hole shows the same result: 3 parts per million (ppm) of gold.

But in one special block, instead of one test, we drill 25 holes, perhaps because it’s easier to reach or looks promising. Each of those 25 holes shows a high grade of 30 ppm gold. If we just average all 425 test results without adjusting for where they came from, the average grade looks like 4.59 ppm.

That may seem small, but it leads to a major error. Using that average, we estimate there’s about 12,398 kilograms of gold in the ground. But if we correct the calculation by giving each of the 400 blocks equal weight, regardless of how many samples were drilled in each, the real average is only 3.07 ppm. That gives us a much lower and more accurate total of 8,282 kilograms.

In other words, by not correcting for clustering, we overestimate the gold by about 50 percent. In mining, that kind of miscalculation could waste tens to hundreds of millions. Sometimes billions.

The same thing happens in policy-making when decisions are based on data from just a few crowded places, especially polling. If one area produces lots of data points, but the rest of the country is under-sampled, we end up making national decisions based on a distorted picture. That leads to costly and unfair outcomes, just like in mining, but for people, not just rock.

Consider how national housing, health care, and criminal laws are built. Most of the data comes from where people cluster, which is overwhelmingly urban. Yet the policies shaped by that data are applied across the entire country. It’s like drilling one corner of a deposit and assuming the rest looks the same.

No one in government applies a correction for clustering. Government departments release national averages. Media outlets amplify them. Polling firms present surveys as representative even when their samples are concentrated in a few metro areas. The public hears numbers, but this fundamental flaw is never disclosed. The result is policy that fits the cities but fails the rest of the country.

Canada’s federal gun control laws offer a clear example. Most firearm-related crime occurs in large cities. Much of it involves gangs and illegal weapons. Yet rural Canadians, who use firearms for hunting, livestock protection, or defence where police response is delayed, face the same restrictions as city residents. A policy designed to reduce urban gun violence makes it harder for rural citizens to protect themselves. The 2020 mass shooting in Nova Scotia was carried out with smuggled firearms by Gabriel Wortman, a man banned from possessing them. No federal gun law in place or proposed would have stopped it. Yet such events are used to justify sweeping legislation applied uniformly across very different contexts.

Policies like this are built on data that is clustered but not corrected. What’s needed is a shift in how we gather, weigh, and interpret national data. The key is not to treat every data point equally, but to recognize that its weight should depend on where it comes from and how overrepresented that area already is.

This is where a declustered governance model comes into play. It borrows from geostatistics and applies the same risk-reduction mindset to policy. The goal is not abstraction or ideology, but simple fairness by measuring Canada in a way that accounts for its geographic and demographic realities.

The approach begins by rejecting the idea that more data means better understanding. In reality, more clustered data without correction leads to more distortion. A single survey in a remote community may reveal more about national gaps than a thousand surveys from downtown Toronto. Without adjusting for location and density, that one survey is ignored.

Policy must reflect where people live, not just how many live in one place. A rural community in Northern Ontario or a fly-in First Nation cannot be treated as a statistical footnote. When they are, policy is bound to fail them.

Risk must also be grounded in location. Crime in the downtown core should not define laws in regions with entirely different social dynamics. Service access must be counted as part of the cost equation. If two people pay the same tax rate but only one has a hospital, a police detachment, or public transit nearby, then the tax burden is not equal in effect.

Local variation matters. If a regulation about building materials makes sense in southern Ontario but breaks the economy in Nunavut, then it is not a neutral rule. If a staffing ratio for hospitals works in Mississauga but collapses in Moose Factory, then uniformity is not fairness.

These insights form the basis for four principles that could guide national decision-making: represent the full range of geography, assess real risks rather than inferred ones, link burdens to actual access, and let local realities shape implementation. The purpose is not to fragment the country but to make it governable by its truth and not just by its clusters.

The same methodology that prevents billion-dollar errors in mining can be used to prevent system-wide failures in housing, justice, health care, and beyond. But it requires abandoning convenience, media shorthand, and the false comfort of averages produced by a failure to allocate them based on reality.

Views expressed in this article are the opinions of the author and do not necessarily reflect the views of The Epoch Times.