The Quantified Country: China runs on numbers — even if the numbers lie | SupChina

The Quantified Country

China is the quantified country. It has long constructed metrics and set targets, tying numbers into both the engineering of policy and the governance of individual lives.

But what if its trust in statistics is misguided?

The reality is, data is tangled up with perverse incentives, human fallibility, and behavioral feedback cycles. Numbers lie more frequently than the country’s leaders are willing to admit, with implications that are far-reaching.

Illustration by Marjorie Wang


By now you’ve probably heard about China’s social credit system, including the many concerns over its opacity and complexity. The EU Chamber of Commerce in China, for instance, published a report on the system last month, noting that there are some 300-odd rating requirements for large companies, with little clarity about how all those factors interact (yes, companies are to be rated, just like individuals).

The social credit scheme has gotten a lot of attention, maybe deservedly so. But its emergence shouldn’t have been a surprise. The Chinese government has long been enamored of metrics of all kinds. Officials have been judged by whether or not they hit certain quotas, whether for GDP growth or population control. Numeric targets are declared for whole industries. Great store and pride is set by indices and international rankings. And as developments of the last few years have shown, Beijing is firmly invested in incorporating big data and algorithms into government operations. Earlier this year, Shanghai even experimented with using AI in the judicial system for assessing evidence.

Just as there is a “quantified self” movement, China long ago embraced a kind of quantitative governance: constructing metrics, setting targets, and tying all of it directly into both the engineering of policy and the governance of individual lives. But as China moves to adopt social credit scoring and invests ever more into “big data” and “AI solutions,” it may be a good time to take a look at the risks — not just with algorithms, but with the very nature of using (and misusing) metrics for governing people.


Problem No. 1: Numbers that are unknown

This must be acknowledged upfront: China has issues with data transparency. For years, basic facts and figures collected by the government — such as the number of executions or certain data about natural resources — have been sealed away as state secrets. In cases where data is not explicitly classified, it may not be exactly accessible, either; official data on migration into and out of the country, for instance, is minimal and difficult to track down. Such virtual redactions have long caused consternation to observers both in and outside of China, leaving critical blindspots in our understanding of the country.

To be sure, China has made a bit of progress in this area the last few years, making more of its statistics publicly available, no doubt influenced by the trend toward open data that has been championed in so many other countries, such as Singapore and Sweden. Still, Open Data Watch, a nonprofit, rates China only 39 out of 100 for data openness.

But the data that actually does get published is often just as problematic. China’s GDP figures are famously “man-made,” according to none other than the country’s sitting premier, a point that several studies have since backed up. And so it is with other official figures, whether crime or unemployment or the national GINI figure (which has only been sporadically published in recent years). When a government invests so much interest in quotas and metrics, it also gives itself an interest in obscuring them if they prove less-than-ideal.

This dark side of Chinese statistics is already well-known. It might seem ironic that a country that puts so much stock in its own numbers should produce so few that are reliable, but in fact there’s no contradiction. Even without interference at the top, the passion for quantitative governance itself warps and undermines much of China’s data.

Sometimes the central government obscures its own data, but sometimes the data it solicits from others is obscured by the very way in which it is measured. And so…


Problem No. 2: Numbers that lie

China’s habits of governance have long run afoul of Goodhart’s Law. A 20th-century economist, Charles Goodhart formulated his “law” while critiquing a monetary policy of the UK in the 1970s: “Any observed statistical regularity will tend to collapse once pressure is placed upon it for control purposes.” It’s a fairly dry expression of a profound point: In measuring and monitoring a part of the economy, and then setting policy around it, the government was creating incentives that would tweak people’s behavior in such a way as to distort the very “statistical regularity” they were depending on.

But Goodhart’s Law is not so much a “law” of statistics as it is one of human behavior, and the implications reach far beyond anything to do with monetary policy, or even economics. It is more like a social science equivalent of the observer effect: measuring anything that people do almost invariably warps what they do, reducing the value of the measurement itself. And this is all the more the case in China: government seeks a certain result, and so sets a numerical goal while brandishing carrots and sticks. The governed respond by meeting the target by whatever means they can, even if that invalidates the very purpose of the metric on which the target is based.

The best example of this is Mao Zedong’s disastrous plan to fast-track China into the industrial age via the Great Leap Forward, from 1958 to 1962. Lower-level leaders felt compelled to report exaggerated and fabricated numbers in order to meet unrealistic targets for output, all while diverting a vast amount of resources from food-production into industry. Millions starved in the countryside as a result.

You would think the central government would have learned its lesson, but no. Decades ago, government policy again prioritized economic development, and incentivized it by rewarding officials who were able to post strong GDP growth rates for their jurisdictions. The results have been well publicized by now. China’s economy has indeed grown, but many officials over the years engaged in “creative accounting,” if not outright fraud, to punch up their local statistics. That is not in the interest of the country, and the National Bureau of Statistics has tended to shave down the GDP figures it’s received from provincial agencies, knowing full well that they are overblown — as some have admitted.

Other examples abound:

Until recently, the need to keep birth rates in line with the one-child policy — combined with the threats of cut wages, demotion, or worse for cadres who failed to keep birth numbers in check — drove local authorities to sometimes compel abortions. That in turn drove many parents with out-of-quota births to hide them from local authorities, with the result that birth numbers as reported to the State Family Planning Commission were regularly 25 to 35 percent below what later census takers found.

The pressure for law enforcement to deliver the right stats has resulted in a justice system with an implausible 99 percent conviction rate.

The determination to see the country’s profile in science and technology elevated has filtered through to universities and research centers, where researchers are rewarded for the number of patents filed and papers published — and so that it is exactly what they do, produce quantities of patents and publications, even if they may be compromised in quality.

(The effect is hardly limited to government, either. Click farms, “water armies,” and even special cradle devices to falsify step counts on fitness tracking apps are all examples of how people “maladapt” to the demands of metrics.)

And then, of course, there is the gaokao, the college entrance exam that decides the futures of millions of Chinese young people every year. With such steep consequences hinging upon their final score, it is surely no surprise that some turn to cheating, or go to extreme lengths to get their hukou (household registration) changed to another province where, in the byzantine calculations of exam scores and admissions cut-offs, they can more easily meet the score threshold for their desired school and major.

In 2016, the government decided to clamp down on cheating on the gaokao, even threatening jail time for the worst forms. Just how effective this may have been is hard to say, but it repeats a pattern in China’s quantitative governance: When people are measured in something that has significant consequences, they adapt, strategize, and attempt to outwit the measures used over them. In turn, the government tends to respond by doubling down on punishments for those who cheat or defraud the system, but doesn’t rethink the process of measurement and goal-setting itself.

Trying to cheat or game the metrics is one thing. But even sincere efforts to meet quantitative targets can lead to further divergences between statistics and the ground truth, as you’ll see in the next part…


Problem No. 3: Numbers that are misused

Experienced social scientists know that working out how to measure a thing is one of the single greatest challenges in any research effort. The mathematician Cathy O’Neil, in her book Weapons of Math Destruction, gives a critique of the metrics that have come to rule so much of modern life, from algorithms used to predict criminal recidivism to credit scores to even US News’s college rankings system, and offers a reminder of a basic point familiar to all scientists: “There [will] always be mistakes…because models are, by their very nature, simplifications. No model can include all of the real world’s complexity… Inevitably, some important information gets left out.”

And so, too, the measures and targets made by governments are, in the end, just proxies for a more complicated reality.

Every measurement entails implicit or explicit choices about what to count and what distinctions matter. Take urbanization: media reports routinely make note of China’s rapid urbanization, with about 60 percent of the population urbanized as of 2018. But of course, the first question to ask is: How is urbanization defined and measured? In fact, there are several different official measures for urban populations in China, but the figures tallied by the National Bureau of Statistics have to do with population density, complicated judgments involving areas designated a priori as urban, and what areas are linked to municipal services. It’s a bit of a messy definition, but then again so is the problem: cities don’t actually have hard boundaries, and people are forever moving in and out of them. But the measure still returns a simple dichotomy of “urban” or “rural,” and lumps a modestly sized town in rural Shandong into the same category as central Shanghai, rather than the villages that might lie nearby. Drill down into any other metric, and the same issues are there.

The point isn’t just that metrics always compromise on reality. It’s that the unavoidable discrepancy between measurement and reality feeds back into policy and action, and this is the deeper implication of Goodhart’s Law: People trying to comply with quantitative governance will find that some things move the needle (the things that a metric captures) and some things just won’t (whatever the metric discounts). All too easily, then, the human beings caught up in the system can fool themselves into thinking they are succeeding when a number ticks up, while becoming blind to all the things the metric fails to encompass. In other words, even people working in good faith can still wind up gaming the system without ever meaning to.

For decades, Chinese government at every level pushed for “economic development,” largely with GDP as the guide. But GDP isn’t a simple measure. It’s a compound metric with built-in choices, priorities, weightings, exclusions, and biases. It can capture investments, infrastructure projects constructed, asset sales made, businesses lured in — and so those are precisely the things that local officials have pursued, because the way in which GDP is calculated says they are of value. And why question if these things have other costs, such as environmental degradation or the happiness of individuals? Why bother to pursue other benefits? If such things mattered, the thinking goes, wouldn’t they be factored into the calculations?

The results of this decades-long quantitative governance are clear enough now. Yes, GDP has grown enormously, even accounting for fraud and overestimates. But it has also meant ghost cities and white elephant projects, and is oblivious to matters of inequality, employment insecurity, housing affordability, or, for that matter, basic life satisfaction, without which the meaning of economic development is moot. In the end, officials were never motivated to improve people’s quality of life per se, because they were motivated to improve a specific figure. That is the ultimate flaw in any kind of quantitative governance.


The upshot

The problems line up: Much of China’s data isn’t open, but even when it is, it may be meddled with. Data may also be warped at an earlier stage, by those who supply it. Beyond the data, the metrics themselves may be distorting how people act and the decisions they make.

China is not the only country to use quantitative governance (and for that matter, quantitative governance isn’t necessarily something only governments engage in). But China is perhaps exceptional in just how aggressively it imposes and pushes for quantitative targets, how extensively numerical targets are used in so many different domains, how heavy-handed the government can be in its rewards and penalties, and in just how little opportunity there is for correction when the predictable errors do arise.

Nor are the problems mentioned above the only ones — but they are some of the most long-standing and difficult to confront. There is much the government could do, if it wished: opening up more data; deemphasizing quantitative goal-setting; more carefully crafting metrics and validating them with pragmatic, qualitative checks. But with all the plans for big data collection and social credit scores, along with the continuing emphasis on industrial planning, it seems unlikely we’ll see the relaxation of quantitative governance any time soon.


M.E. Strickland

M.E. Strickland is an anthropologist and translator whose work has spanned from rural Shandong to Beijing’s tech industry.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.