Everything big data claims to know about you could be wrong

When it comes to understanding what makes people tick — and get sick — medical science has long assumed that the bigger the sample of human subjects, the better. But new research by UC Berkeley, Drexel University and the University of Groningen suggests this big-data approach may be wildly off the mark. That’s largely because emotions, behavior and physiology vary markedly from one person to the next and one moment to the next. So averaging out data collected from a large group of human subjects at a given instant offers only a snapshot, and a fuzzy one at that, researchers said.

The findings, published in the Proceedings of the National Academy of Sciences journal, have implications for everything from mining social media data to customizing health therapies, and could change the way researchers and clinicians analyze, diagnose and treat mental and physical disorders.

“If you want to know what individuals feel or how they become sick, you have to conduct research on individuals, not on groups,” said the authors. “Diseases, mental disorders, emotions, and behaviors are expressed within individual people, over time. A snapshot of many people at one moment in time can’t capture these phenomena.”

Moreover, the consequences of continuing to rely on group data in the medical, social and behavioral sciences include misdiagnoses, prescribing the wrong treatments and generally perpetuating scientific theory and experimentation that is not properly calibrated to the differences between individuals.

That said, a fix is within reach: “People shouldn’t necessarily lose faith in medical or social science,” lead-author Aaron Fisher said. “Instead, they should see the potential to conduct scientific studies as a part of routine care. This is how we can truly personalize medicine.”

Plus, he noted, “modern technologies allow us to collect many observations per person relatively easily, and modern computing makes the analysis of these data possible in ways that were not possible in the past.”

HOW THEY CONDUCTED THE RESEARCH

Aaron Fisher (University of California - Berkeley), John Medaglia (Drexel University) and Bertus Jeronimus (University of Groningen, the Netherlands) used statistical models to compare data collected on hundreds of people, including healthy individuals and those with disorders ranging from depression and anxiety to post-traumatic stress disorder and panic disorder.

In six separate studies they analyzed data via online and smartphone self-report surveys, as well as electrocardiogram tests to measure heart rates. The results consistently showed that what’s true for the group is not necessarily true for the individual.

For example, a group analysis of people with depression found that they worry a great deal. But when the same analysis was applied to each individual in that group, researchers discovered wide variations that ranged from zero worrying to agonizing well above the group average.

Moreover, in looking at the correlation between fear and avoidance – a common association in group research – they found that for many individuals, fear did not cause them to avoid certain activities, or vice versa.

More information:

• Journal Reference: Aaron J. Fisher, John D. Medaglia, Bertus F. Jeronimus. Lack of group-to-individual generalizability is a threat to human subjects research. Proceedings of the National Academy of Sciences, 2018; 201711978 DOI: 10.1073/pnas.1711978115
• Profile Bertus Jeronimus
• Source: http://news.berkeley.edu/2018/06/18/big-data-flaws/