Sweet tweets! Evaluating a new approach for probability-based sampling of Twitter

Area of Science:

• Social Sciences
• Data Science
• Computational Social Science

Background:

• Rising survey costs and declining response rates necessitate cost-effective data collection methods.
• Social media and sensor data are emerging as viable alternatives for public opinion research.
• Current methods for accessing and sampling social media data, particularly Twitter, are still under development.

Purpose of the Study:

• To address the gap in understanding how to randomly sample Tweets for representative datasets.
• To develop and test a novel algorithm for probability-based sampling of Tweets.
• To enable quality estimates of public opinion from social media sources like Twitter.

Main Methods:

• Proposed and tested the Velocity-Based Estimation for Sampling Tweets (VBEST) algorithm.
• Compared VBEST sample estimates against Twitter Search API access methods.
• Evaluated performance based on total Tweet distribution, COVID-19 keyword incidence, and frequency.

Main Results:

• VBEST samples demonstrated consistent performance across various experimental conditions.
• VBEST exhibited relatively low overall bias compared to common Search API access methods.
• The algorithm proved effective in producing representative daily discourse datasets, even within specific geographical regions.

Conclusions:

• The VBEST algorithm provides a robust and less biased method for sampling Tweets.
• This approach supports the expansion of the fit-for-purpose paradigm to include social media data for public opinion research.
• VBEST facilitates more accurate and cost-effective analysis of public discourse from platforms like Twitter.