When it comes to understanding public well-being, planners and policymakers have a limited set of tools at their disposal – most commonly Census data, surveys and polls that are often conducted at great cost, and economic indicators such as unemployment rates. With the rapid adoption of social media, some researchers see a new opportunity to measure sentiment in a more rapid, cost-effective, and targeted way.

In a new Lincoln Institute working paper, authors Justin Hollander of Tufts University and Henry Renski of the University of Massachusetts use Twitter data to compare public attitudes in 50 shrinking, mostly postindustrial cities with those in 50 stable, growing cities.

After carefully selecting locales with similar demographics, weather, topography, and other characteristics, the authors collected more than 300,000 tweets over two months. They analyzed the contents using the AFINN sentiment dictionary, which rates some 1,500 English words based on their degree of positive or negative sentiment on a scale of -5 to 5 (e.g., “abusive” is scored -3, while “satisfied” is scored +2 ).

The authors found no significant1 difference in the attitudes of resident in shrinking cities and growing cities.

Measuring Urban Attitudes Using Twitter: An Exploratory Study (Working Paper)

Author(s): Hollander, Justin B. and Henry Renski

Publication Date: December 2015, 25 pages; Inventory ID WP15JH1; English 

Abstract:  The goal of this working paper is to introduce a new breed of powerful software tools and social media data that can be used to study the attitudes of people in urban places. In particular, the paper reports on the work of the Urban Attitudes Lab, where a research project experimented with using microblogging data in conjunction with a mix of quantitative and qualitative methods, including content analysis and advanced multivariate statistics, to study the urban experience and draw implications for public policy. The research used propensity scoring to develop matched pairs of mid-sized U.S. cities in the Northeast and Midwest, where the most significant difference between each pair is that of population decline. This resulted in a group of 50 declining cities matched with 50 growing/stable cities. Over 300,000 Twitter posts were collected over the course of two-months, each analyzed for either positive or negative sentiment. After running difference of means tests, we found that sentiment in the declining cities does not differ in a statistically significant manner from stable and growing cities. These findings suggest that real opportunities exist to better understand urban attitudes through sentiment analysis of Twitter data.

• 1. “The research has important implications for public policy,” Hollander and Renski write in the paper, Measuring Urban Attitudes Using Twitter: An Exploratory Study. “It suggests that population decline itself may not contribute to lower overall sentiment levels, which means local, state, and federal agencies ought to better explore how decline does impact neighborhoods and overall community well-being.”

  This type of research comes with caveats, such as the difficulty of distilling human language down to numerical scores, and the skewed sample represented by Twitter users, who are more likely to be young and male, with greater access to smartphones and computers. But more traditional data also has limitations.

  Social media also has its advantages, from the timeliness and sheer quantity of the data – unmatched by even the most thorough surveys – to the availability of precise geographic targeting.

  The authors plan to continue their research, partnering with a linguist to refine their interpretation of language. Next up is a study for the New York City Department of Design and Construction to explore the sentiments of people who live and work near public buildings and plazas.

  “When we think about the best way to do public policy and planning, you always want to start with the people who are most affected,” Hollander said. “What Twitter allows us to do is zoom in on very small geographies, and over time you can collect enough data to get some insight.”