Supervised machine learning methods are increasingly employed in political science. Such models require costly manual labeling of documents. In this paper, we introduce active learning, a framework in which data to be labeled by human coders are not chosen at random but rather targeted in such a way that the required amount of data to train a machine learning model can be minimized. We study the benefits of active learning using text data examples. We perform simulation studies that illustrate conditions where active learning can reduce the cost of labeling text data. We perform these simulations on three corpora that vary in size, document length, and domain. We find that in cases where the document class of interest is not balanced, researchers can label a fraction of the documents one would need using random sampling (or “passive” learning) to achieve equally performing classifiers. We further investigate how varying levels of intercoder reliability affect the active learning procedures and find that even with low reliability, active learning performs more efficiently than does random sampling.

The purpose of this study was to compare the clinical outcome, union rate, and complications of a consecutive series of Scaphoid excision and limited wrist arthrodesis performed by a single surgeon using distal radius bone graft and K-wires or circular plate fixation. A sequential series of ten patients(11 wrists) who were stabilized with temporary K-wires were compared to 11 patients (11 wrists) who were stabilized with a circular plate. Minimum follow-up was 1 year. One patient in the Kwire group was converted to a wrist fusion. Six of the remaining ten patients in the K-wire fixation group and 8 of the 11 patients in the circular plate fixation group returned for the following blinded evaluations: Quick DASH, analog pain scale, range of motion, grip and pinch strength, plain x-ray, and multi-detector computed tomography evaluation. One non-union occurred in the K-wire group. There were no non-unions in the circular plate fixation group. There was no difference in any of remaining measures or rate of complications. This study shows that equivalent results can be obtained using circular plate fixation compared to Kwires when equivalent bone graft source and fusion technique are used. If K-wire removal requires a return to the OR, circular plate fixation is more cost-effective.

Despite massive investment in China's censorship program, internet platforms in China are rife with criticisms of the government and content that seeks to organize opposition to the ruling Communist Party. Past works have attributed this "openness" to deliberate government strategy or lack of capacity. Most, however, do not consider the role of private social media companies, to whom the state delegates information controls. I suggest that the apparent incompleteness of censorship is largely a result of principal-agent problems that arise due to misaligned incentives of government principals and private media company agents. Using a custom dataset of annotated leaked documents from a social media company, Sina Weibo, I find that 16% of directives from the government are disobeyed by Sina Weibo and that disobedience is driven by Sina's concerns about censoring more strictly than competitor Tencent. I also find that the fragmentation inherent in the Chinese political system exacerbates this principal agent problem. I demonstrate this by retrieving actual censored content from large databases of hundreds of millions of Sina Weibo posts and measuring the performance of Sina Weibo's censorship employees across a range of events. This paper contributes to our understanding of media control in China by uncovering how market competition can lead media companies to push back against state directives and increase space for counterhegemonic discourse.

No abstract

In this paper, we examine how the Chinese state controls social media. While social media companies are responsible for censoring their platforms, they also selectively report certain users to the government. This article focuses on understanding the logic behind media platforms’ decisions to report users or content to the government. We find that content is less relevant than commonly thought. Information control efforts often focus on who is posting rather than on what they are posting. The state permits open discussion and debate on social media while controlling and managing influential social forces that may challenge the party-state's hegemonic position. We build on Schurmann's “ideology and organization,” emphasizing the Party's goals of embedding itself in all social structures and limiting the ability of non-Party individuals, networks or groups to carve out a separate space for leadership and social status. In the virtual public sphere, the Chinese Communist Party (CCP) continues to apply these principles to co-opt, repress and limit the reach of influential non-Party “thought leaders.” We find evidence to support this logic through qualitative and quantitative analysis of leaked censorship documents from a social media company and government documents on information control.