Meet the Data Scientist Revealing How Big China's 'Ghost Cities' Really Are
Baidu is revealing the truth about China's emptiest cities.
Zhengzhou, China. Image: Flickr/Ran Allen
China is changing faster than anyone can can keep up with—or keep track of. The country's bold urbanization plan, an attempt to nudge China's economy from manufacturing to consumption, has prompted untold numbers of government-built concrete cities to pop up out of thin air, complete with towering apartment buildings and grandiose monuments.
It's utopian social engineering on a massive scale. But if you were to judge the plan by the photos in Western media coverage, you'd think China is building gargantuan open-air tombs.
The new towns aren't filling up quickly enough, and those towering apartment buildings are standing empty in areas dubbed "ghost cities."
The government has yet to publish any official information on the vacancy rates in these notoriously underpopulated areas. As far as public stats go, the residents in these cities might as well be ghosts themselves.
"We don't know the type of data the government has, but at least we can use our data to penetrate this problem," Haishan Wu, a data scientist at Baidu's Big Data Lab, told me in an interview.
Baidu, China's Google-like Internet search company, is using location data collected from 770 million Chinese users to take the pulse of these cities and quantify who lives there. The results of the study were published this week on the ArXiv preprint server, and has not been peer-reviewed.
"It covers most smartphone users, and we can do very detailed research," Wu said. "Maybe the government has their own data, but at least we can use our own data to assess the state of ghost cities in China."
A team from Baidu's Big Data Lab—where China's largest search engine company figures out ways to parse its massive corpus of data—and Peking University in Beijing used location data from Baidu Maps to put together what they claim is the most accurate picture of China's ghost cities to date.
The results showed that some of the most famous ghost cities, the ones most covered by the media like the Kangbashi New Area, are actually fairly well populated now, and the older suburbs have emptied. They also revealed previously unknown ghost cities, like Dongying, an overbuilt oil boom town.
The researchers' data set contained billions of locational data points for 770 million Baidu users—China has a population of 1.36 billion—captured between September 2014 and April 2015. The results were used to make a handy interactive map of China's ghost cities.
Of course, the sample size is biased towards younger people who own smartphones, and to Baidu users more generally, the paper notes. But it is a substantial sample size, and offers a rare look at how younger people in China are responding to the government's urbanization initiatives.
But location only tells part of the story—the "what." There's also the "why," and Wu believes Baidu only has enough information to venture a guess right now.
"I believe the most important reason is a lack of basic infrastructure like hospitals and shopping malls," Wu said. "At first, for new cities, only a few people live there, so developers are reluctant to build big shopping malls there."
It's a chicken and the egg situation, Wu speculated—there's little infrastructure because not enough people live there, and not enough people live there because there's little infrastructure.
There are no doubt myriad reasons to explain the complex dynamics of migration in modern China, however. The government will sometimes offer apartments in exchange for rural land rights, for example, but it's a deal some citizens may be reluctant to take. Land rights are one of the last real remnants of the socialist state in China, according to Cornell labour professor Eli Friedman.
Wu and the Big Data Lab at Baidu aren't done with their work, however, and hope to fill in some of the blanks that remain in the story of China's spectral cities. The next step, Wu told me, is to beef up their analysis, so far based on location information, with more data on road work and building projects that can give some context to their results.
"We don't expect a response from the government, but we hope that this work is helpful for them in terms of urban planning," Wu said. "If they come to us and read our paper, I think that would be a very good thing."