Predicting individual-level income from Facebook profiles

Area of Science:

• Social Computing
• Computational Social Science
• Machine Learning Applications

Background:

• Personal income data is valuable for business applications like targeted advertising and salary negotiations.
• Individuals often perceive income as private, leading to reluctance in sharing such information.
• Digital footprints, particularly social media activity, offer a potential alternative data source.

Purpose of the Study:

• To investigate the predictability of individual income using digital footprints from Facebook.
• To quantify the accuracy of income prediction based on Facebook Likes and Status Updates.
• To assess the incremental predictive value of social media data beyond traditional socio-demographic variables.

Main Methods:

• Utilized a machine learning approach to analyze Facebook Likes and Status Updates.
• Applied the method to a representative sample of 2,623 U.S. Americans.
• Correlated social media data with reported income and socio-demographic variables.

Main Results:

• Facebook Likes and Status Updates alone predicted income with an accuracy up to r = 0.43.
• Social media data provided incremental predictive power (ΔR2 = 6-16%, r = 0.49) over socio-demographic factors.
• The study demonstrates significant predictive capability of digital footprints for income.

Conclusions:

• Digital footprints on Facebook are strong predictors of personal income.
• These findings present opportunities for businesses but also highlight substantial privacy risks.
• Ethical considerations regarding consent and data usage are crucial for such predictive models.