CSS Fellowships yield cross-disciplinary insights
One of IRiSS's mandates is to facilitate computationally intensive social science research. IRiSS sponsors a number of trainings and workshops to this end, but a major priority is funding research itself. IRiSS’s Center for Computational Social Science runs a competitive grant proposal program that awards five Computational Social Science Fellowships per year. The funding assists graduate students conducting computational research to acquire data, pay RAs to process and validate data, and other research activities.
The 2019-2020 CSS Fellows conducted innovate computational research that pushed disciplinary boundaries and yielded a trove of rich insights. The research reports provided by the fellows describe their groundbreaking findings and the impact of the fellowships on their research.
The Effects of Slum Clearance on Displaced Residents: Evidence From Victorian London — Yiming He, Economics
My doctoral dissertation studies slum clearance in Victorian England, focusing on slum residents that were displaced after clearance. I combine original government archival data on slum clearance and historical individual census data that have recently become public to researchers. I conducted two archival trips to England in 2019 that allowed me to digitize original slum clearance maps and engage with local historians. I appreciate the funding support from CSS fellowship that allows me to hire freelance workers to assist the census data collection, which speeded up my research tremendously in the past year.
Understanding Live Broadcasting Media Events and Intensified Emotions — Mufan Luo, Communication
Live streaming services allow people to concurrently consume and comment on media events with other people in real time. Durkheim’s theory of “collective effervescence” suggests that face-to-face encounters in ritual events conjure emotional arousal, so people often feel happier and more excited while watching events like the Super Bowl with family and friends through the television than if they were alone. Does a stronger emotional intensity also occur in live streaming? Using a large-scale dataset of comments posted to news and media events on YouTube, we address this question by examining emotional intensity in live comments versus those produced retrospectively. Results reveal that live comments are overall more emotionally intense than retrospective comments across all temporal periods and all event types examined. Findings support the emotional amplification hypothesis and provide preliminary evidence for shared attention theory in explaining the amplification effect. These findings have important implications for live streaming platforms to optimize resources for content moderation and to improve psychological well-being for content moderators, and more broadly as society grapples with using technology to stay connected during social distancing required by the COVID-19 pandemic.
Formal Inclusion, Informal Exclusion: The Gender Differences In Social Interaction At Work — Katarina Mueller-Gastell and Austin Van Loon, Sociology
With funding from the IRiSS Computational Social Science Fellowship, we have been able to make tremendous progress on our joint project on gender differences in workplace social interaction. In this project, we study whether social ties are transposed to formal interaction, and whether this process happens unequally for men and women. We are using the CSS Fellowship funding to pay for the manual classification of email data from two different companies as containing social or work-related content. This manual classification is now complete for one of the two email datasets. Using these manually classified data, we have trained a machine learning model to classify the full set of emails in the corpus. We are currently in the process of doing this for the second company email dataset. We are looking forward to having comparative results at the end of the summer. We hope that our findings on how social ties can exclude women from task-related interactions could provide important insight to devising better workplace diversity policies.
The Effect of High-Skilled Firm Entry on Gentrification — Franklin Qian and Rose Tan, Economics
What happens to low-skilled incumbent residents after the entry of a large establishment employing mainly high-skilled workers? Using 391 large high-skilled firm entries, we follow local residents over a 8-year period after the firm entry announcement using rich micro-data including individual address history, homeownership, and financial records. We compare outcomes for individuals living close to the entry site with those living far away while controlling for neighborhood demographics and access to potential high-skilled firm entry sites. The rise in wages and rents following firm entry primarily accrues to high-skilled owners and renters, who are better off financially. Both low-skilled owners and renters are more likely to move, especially to places farther away outside of the MSA. Low-skilled renters live in slightly worse neighborhoods and live in lower-priced homes, but are not worse off financially. Low-skilled owners cash out and have neutral outcomes in terms of neighborhood quality, homeownership, and financially. Using a structural spatial equilibrium model, we decompose welfare from changes in wages, rents, and amenities for low-skilled owners and renters. We find that the effect of high-skilled firm entry on low-skilled incumbents is overall neutral. Low-skilled owners neither benefit nor are they harmed, while low-skilled renters live in slightly worse neighborhoods but are not worse off financially.
The Impact of Artificial Intelligence on the Labor Market — Michael Webb, Economics
Michael Webb developed a new methodology to address a timely question that is on policymakers’—and workers’—minds: what will the effects of artificial intelligence (AI) be on jobs? While many researchers are working on different facets of this question, Webb’s dissertation research is focused finding a way to make credible predictions about the effects of AI in the future, rather than simply studying the impacts of past automation. Using natural language processing on historical databases of patent records and job descriptions, Webb uses the overlap between these two bodies of text to develop an algorithm that can correctly predict which occupations were impacted by new technologies. Although Webb had developed the fundamentals of the model upon beginning his CSS Fellowship, the painstaking process of calibrating and validating his algorithm required hundreds of hours of ingesting, standardizing, and linking a large number of disparate data sources. He was able to hire a research assistant to carry out this work, thereby refining the predictive capabilities of his model and freeing up valuable time to work on other aspects of his research.