Author Date

2021-03-18

Degree Name

BS

Department

Economics

College

Family, Home, and Social Sciences

Defense Date

2021-03-12

Publication Date

2021-03-18

First Faculty Advisor

C. Arden Pope III

First Faculty Reader

Robert Reynolds

Honors Coordinator

John E. Stovall

Keywords

sentiment analysis, social media, United States, air pollution, happiness, visibility

Abstract

Objective

This study examines the associations between actual and perceived air pollution (PM2.5, AQI, and ground visibility), weather information, and expressed sentiment via US Twitter. Heterogeneity in the associations across date and county characteristics are also explored.

Methods

A sentiment index was constructed using 27,827,828 geotagged U.S. tweets posted between May 31 and November 30, 2015. Associations between AQI category changes and the sentiment index were estimated using multi-cutoff regression discontinuity models. Associations between same-day and lagged PM2.5, ground visibility, and the sentiment index were estimated using weighted linear regression models. Models include weather variables and county and date fixed effects. Stratified analyses by county type (MSA, urban, rural) and date characteristics (holiday or non-holiday, weekday or weekend) were performed.

Results

Being in the AQI category of Moderate rather than Good is estimated to predict a 1.5 percentage point decrease in the sentiment index. A 1-mile increase in ground visibility is estimated to predict roughly a 0.34 percentage point increase in the sentiment index, while increasing PM2.5 is found to predict a very small increase of about 0.02 percentage points per 1 g/m3. Temperature, pressure, wind speed, and precipitation were all found to significantly affect sentiment. Heterogeneous results were observed across both date and county characteristics.

Conclusion

The findings suggest that air pollution has a short-term psychological effect on expressed sentiment via U.S. Twitter but not a physiological effect. Weather variables are also found to be significantly associated with expressed sentiment.

Share

COinS