Introducing the Policy Change Index Newsletter
Originally Posted on December 13, 2018
Dear Human Readers,
As you may know, the Policy Change Index (PCI) for China was launched in October. In a nutshell, it’s a machine learning algorithm that “reads” the People’s Daily — China’s state-run newspaper — and predicts major policy shifts in China using only the information in that publication. It works because of the highly official status of the People’s Daily; the government often alters discourse before changing course.
You can find out more about the PCI on the project website, policychangeindex.org, including the original research paper and a link to the GitHub repository.
For those who prefer things in bite size, Weifeng Zhong presented the PCI for China at the inaugural Policy Simulation Library meetup, hosted by the Open Source Policy Center at AEI.
For the data science folks, Julian TszKin Chan and Zhong have written about how the PCI method applies machine learning with some novel “twists.” In the Berlin-based Dataconomy, they argue that easily available and ostensibly trivial labels can be used to uncover nontrivial patterns. The page numbers of People's Daily articles are a case in point. In addition, machine learning can not only automate the labeling of unseen data but also be useful for detecting structural differences in complex data, such as the changes in the People's Daily’s priorities. They discuss this approach in the San Francisco-based MLconf.
If you’re interested in DIY, the PCI source code has been released on GitHub, where you can report bugs, request enhancements, or propose code changes. If you don’t have the People's Daily text to play with, don’t fret. In a Quantitative Note, Chan and Zhong provide a simulated example in which they produce some “fake news,” manufacture some policy changes, and demonstrate that the PCI method works as intended. The source code of this example is also available in the GitHub repository.
Since the launch, the PCI has gotten quite a bit of media attention, including the coverage by Zhaoyin Feng at the BBC, William Holland at the Asia Times, Tin Cheuk Leung at the Am730, and Jun Mai at the South China Morning Post. But what is particularly interesting is to see it covered by the web version of the Global Times, an offshoot of the People's Daily. So, we now know the people at the People’s Daily know about our study of their daily — and so do you.
The construction of the PCI does not require the researcher to read the text — a nice “language-free” feature. This property has obvious, broad applications. Chan, Zhong, and their collaborators at AEI hope to soon demonstrate that with the PCI for Cuba, based on its official newspaper Granma.
If you have ideas for collaborations in the PCI project or other projects that share the same methodology, feel free to write back. Encouraging contributions and development is the purpose of making the PCI open source in the first place.
WHAT WE'RE READING
“Public Opinion in Soviet Russia” (Harvard University Press 1950) by sociologist Alex Inkeles became an instant classic in Soviet mass communication. Written when little was known about the subject, the book offered a penetrating analysis of its inner workings. As Vladimir Lenin put it, “The whole task of the Communists is to be able to convince the backward elements,” referring to the masses — including us all, of course — who are presumably incapable of progressing without the party’s guidance. It follows that the media must be completely instrumental to the “correct” unfolding of history. Inkeles’ book is so timeless that if you swapped out the Soviet-specific terms, it would fit the modern-day China, Cuba, North Korea, and Vietnam just as well.
Nikolai Bukharin (editor) and Maria Ilyinichna Ulyanova (Lenin’s sister) at Pravda’s editorial office.
© Hulton-Deutsch Collection/CORBIS
Edited by Weifeng Zhong and Julian TszKin Chan