- Curricular Initiatives
- The Botwinick Prize in Business Ethics
- The KPMG Peat Marwick / Stanley R. Klion Forum
- The Paul M. Montrone Seminar Series on Ethics
- Military Initiative Programming
- Leadership and Ethics Week
- Diversity and Inclusion for All
- Leadership Conference
- Diversity, Equity, and Inclusion (DEI) Academic Conference
- Restoring Trust: New Realities and New Possibilities for Business Leadership
- Conscious Capitalism: How Ethical Executives Move the Needle Forward, One Business Decision at a Time
- Lucy Quist: A Global Role Model for Business Leadership
- Two Industry Pioneers Lead the Change for Clean Energy
- The Great Debate on the Ethics of Pricing in the Drug Industry
- Leading With Courage: Top Industry Trailblazers Discuss Pathways to Restoring Trust in Business
- Innovation and the Value of Privacy
- Events Calendar
- Ethical Insights
- Support Us
New York – Depression is one of the most common mental health issues in the United States, affecting the lives of over 17 million Americans suffering from it with over 800,000 people dying from suicide annually, according to the World Health Organization (WHO). While long held stigmas around mental health are being torn down, dependency on technology to accurately predict depression is on the rise. Recent advances in mobile sensing technologies and machine learning have sparked hope and optimism among scientists that claim predictive modelling could revolutionize the way depression assessments are conducted. But research from Sandra Matz, the David W. Zalaznick Associate Professor of Business at Columbia Business School, warns that the swift shift to technologies like GPS sensors and machine learning as predictors – without first validating them in the general population – is having a lasting impact on accuracy.
The study Depression predictions from GPS-based mobility do not generalize well to large demographically heterogeneous samples, that was led by Sandrine Müller and Xi (Leslie) Chen from Columbia University, analyses over 57 million GPS data points to show that the same procedure that leads to high prediction accuracy (approx. 80%) in homogeneous student samples, leads to much lower accuracies in a U.S.-wide sample using heterogeneous socio-demographics (approx. 60%). Notable, even training the classification model on more homogeneous subsamples -- including similar age, similar city, and daily schedules -- did not substantially improve prediction accuracy either.
Overall, the researchers’ ﬁndings challenge the notion of applying mobility-based predictions of depression at scale. While the findings suggest that mobility patterns can indeed reveal information about mental health, they suggest that findings generated on small, homogenous student populations might not easily generalize to the broader population. Without proper validation, the implementation of smartphone-based diagnostic tools could, in fact, cause more harm than good: Inaccurate diagnostic output could both prevent individuals from seeking out the right venues for further diagnostic assessment and treatment as well as unnecessarily bind scarce resources.
To learn more about the cutting-edge research being conducted at Columbia Business School, please visit www.gsb.columbia.edu.
About the researcher
Sandra Matz takes a Big Data approach to studying human behavior in a variety of business-related domains. She combines methodologies from psychology and computer...Read more.