• Graduate program
    • Why Tinbergen Institute?
    • Program Structure
    • Courses
    • Course Registration
    • Facilities
    • Admissions
    • Recent PhD Placements
  • Research
  • News
  • Events
    • Summer School
      • Behavioral Macro and Complexity
      • Econometrics and Data Science Methods for Business and Economics and Finance
      • Experimenting with Communication – A Hands-on Summer School
      • Inequalities in Health and Healthcare
      • Introduction in Genome-Wide Data Analysis
      • Research on Productivity, Trade, and Growth
      • Summer School Business Data Science Program
    • Events Calendar
    • Tinbergen Institute Lectures
    • Annual Tinbergen Institute Conference
    • Events Archive
  • Summer School
  • Alumni
  • Times
Home | Events Archive | Observing the unobservable: Connecting persistent emotional state to out-of-home visits in cities under risky contexts
Seminar

Observing the unobservable: Connecting persistent emotional state to out-of-home visits in cities under risky contexts


  • Series
  • Speaker(s)
    Siqi Zheng (Massachusetts Institute of Technology, United States)
  • Location
    Online
  • Date and time

    June 27, 2022
    12:00 - 13:00

Abstract:

Microeconomic choices are often strongly affected by unobserved factors that determine individual preference orderings and behavior. Economists have employed numerous experiments and survey instruments to measure those unobserved subjective attributes, yet we lack scalable datasets for researchers and policymakers to connect subjective traits with observed behaviors. Here, we use social media platforms as an unsolicited poll to infer individuals' persistent emotional state, and link it to individual behavior during the COVID-19 pandemic. We construct a unique dataset comprising the universe of posts [N=20.24 Million] for a cohort of over 500,000 individuals from 2019 January to 2021 June from Sina-Weibo, the largest social media platform in China. We use a deep learning-based Natural Language Processing (NLP) technique to infer the level of emotions expressed in each post, and we classify individuals based on average fear expression during 2019, prior to the pandemic. We then link individual's fear types to their visitation patterns constructed from the location of their geotagged posts. Employing a difference-in-differences design, we find that high-fear individuals display a 4.2-6.9% greater reduction in out-of-home activities compared to low-fear individuals within the same city. We find that these effects persist for more than 18 months, well beyond the emergency phase of the pandemic response. The behavior gaps between fear groups are largest (20% additional reduction) at high-risk venues and among those individuals that display a low interest in leisure activities in their posts. We provide strong evidence that behavior responds more strongly to persistent differences in fear across individuals, rather than transitory differences triggered by the crisis. Our research shows that using cohort-based social media posts to infer persistent emotional state can be a useful tool for measuring important unobservable individual traits at scale, with potential applications in choice modeling, behavior prediction, and targeted interventions.

Please register here before Thursday June 23, 9 am if you would like to have an online bilateral with the speaker.

To join the seminar, use the following link