Big Bad Data vs. Bad Big Data

“Investigators will have to focus on the scientific quality of the evidence, and lesson the statistic methods… recalling the old adage that statistics are like a bikini bathing suit: what is revealed is interesting; what is concealed is crucial.”

– Dr. Alvan Feinstein

Even archaeologists are using big data. Centurions and those who followed them seem to have broken pottery all over what is now becoming post-nationalist Europe. I spent a lot of time using big data in the ‘80s, and learned the hard way that no amount of data can make up for lack of experiential learning and wisdom (for a fictionalized account, see ). I was spoiled, then. I had worked with archival data of the highest quality: vital records and satellite imaging that was precise and accurate. But I realize now that they were largely irrelevant, being more social and physical than biological.

A couple of decades ago, when the modern adoration of big data was boosted by security fears, I began to notice the extent to which people believed that quantity could overcome lack of quality. I also began to notice how bad some of the big data were. The most noticeable was the change at the Canadian border. I was held up every time—after having crossed for years without difficulty. Apparently there was a dope smuggler with my name, and my name had been flagged. It took a while for the penny to drop: no middle name and no picture in the database! What could go wrong? A decade on, my son and his friends in IT were selling data to advertisers; the data looked good—and the profit was substantial. Did they do any more than treat client anxiety by beefing up the illusion of control? I think not.

So why do I bring it up? The big British cellphone study. Like most policy-oriented studies of non-ionizing radiation, this one is big and simple—too simple to make biomedical sense. Studies tracking chemical exposures through armbands are likewise limited by crude assessment of a piece of an exposure with no attention to total end-organ exposure in or over time. The assumption that what isn’t measured doesn’t matter has never been so hazardous or obstructive. Small N-of-1 studies would enable patients to protect themselves. And, aggregated high-quality data could yield useful knowledge decades faster, and be far less misleading.