Data Mining Bias
Data mining bias occurs when patterns or causal relationships are falsely identified in random or coincidental data, leading to superstitious beliefs and conspiracy theories. It exploits human tendency to see structure where none exists.
THE IDEA
Data Mining, Statistics, and Charlatanism
What is your probability of winning the New Jersey lottery twice? One in 17 trillion. Yet it happened to Evelyn Adams, whom the reader might guess should feel particularly chosen by destiny. For the much broader event of someone, somewhere, in some unspecified way getting extraordinarily lucky, researchers Percy Diaconis and Frederick Mosteller estimated odds of 30 to 1!
Some people carry their data mining activities into theology—after all, ancient Mediterraneans used to read potent messages in the entrails of birds. An interesting extension of data mining into biblical exegesis is provided in The Bible Code by Michael Drosnin. Drosnin, a former journalist (seemingly innocent of any training in statistics), aided by the works of a “mathematician,” helped “predict” the former Israeli Prime Minister Yitzhak Rabin’s assassination by deciphering a bible code. He informed Rabin, who obviously did not take it too seriously. The Bible Code finds statistical irregularities in the Bible; these help predict some such events. Needless to say that the book sold well enough to warrant a sequel predicting with hindsight even more such events.
The same mechanism is behind the formation of conspiracy theories. Like The Bible Code they can seem perfect in their logic and can cause otherwise intelligent people to fall for them. I can create a conspiracy theory by downloading hundreds of paintings from an artist or group of artists and finding a constant among all those paintings (among the hundreds of thousand of traits). I would then concoct a conspiratorial theory around a secret message shared by these paintings. This is seemingly what the author of the bestselling The Da Vinci Code did.
Even in fields that claim to be rigorously empirical, data mining can produce spurious but convincing patterns. Historically, medicine has operated by trial and error—in other words, statistically. We know by now that there can be entirely fortuitous connections between symptoms and treatment, and that some medications succeed in medical trials for mere random reasons. Medical researchers are rarely statisticians; statisticians are rarely medical researchers. Many medical researchers are not even remotely aware of this data mining bias. True, it may only play a small role, but it is certainly present. One medical study cited here reported a link between cigarette smoking and a reduction in breast cancer, which Taleb describes as conflicting with previous studies. Such a result warrants scrutiny for chance findings; disagreement with earlier work does not by itself establish that data mining caused it.
When to use it
Evaluating claims in fields like parapsychology or alternative medicine where pattern-seeking is common.
Identifying and debunking conspiracy theories by recognizing data mining tactics.
Improving statistical literacy when reading about lottery wins, rare diseases, or miraculous events.
Applying rigorous hypothesis testing to avoid false positives in scientific research or business analytics.
Evaluating new medical treatments where early small studies show dramatic benefits.
Analyzing large datasets in any field where many variables are tested against an outcome.
Reviewing scientific literature to identify potential false positives from multiple testing.
Designing clinical trials with proper correction for multiple comparisons.
FROM THEORY TO EVERYDAY LIFE
In practice
Evelyn Adams' Double Lottery Win
For a specified pair of lottery wins, Taleb cites a probability of 1 in 17 trillion. That is different from asking whether at least 1 person across many opportunities will experience extraordinary luck. His discussion cites odds of 30 to one for the broader event, illustrating why the event and population must be defined before calculating the probability.
The Bible Code
Michael Drosnin's The Bible Code claimed to find hidden messages in the Hebrew Bible predicting historical events, such as the assassination of Yitzhak Rabin. Despite lacking statistical rigor, the book became a bestseller by cherry-picking patterns from a vast text, a classic example of data mining bias.
Conspiracy Theories and The Da Vinci Code
Taleb imagines searching hundreds of paintings for a common trait and then constructing a secret-message theory around it. He compares that process to the premise of The Da Vinci Code. The passage offers a hypothetical mechanism for finding patterns after the fact, not evidence of how Brown researched his novel.
Smoking and Breast Cancer Reduction
Taleb cites a study that reported a link between smoking and reduced breast cancer risk, and raises the possibility of a chance finding. Evaluating that suggestion would require examining the study’s design and analysis; a surprising conclusion alone does not establish data mining.
A MOMENT TO REFLECT
Put it to work.
Think of a time you noticed a surprising coincidence. How could data mining bias explain your interpretation of that event?
When you hear a conspiracy theory, what steps can you take to determine whether its evidence is the result of data mining?
How might a data mining bias affect the way people interpret statistical findings in news reports or social media?
How can researchers in medicine reduce the chance of mistaking random noise for a real effect?
Why is it important for medical researchers to have statistical training?
How would you critically evaluate a surprising study that contradicts established knowledge?
References
- The Bible Code — Michael Drosnin (1997)
- The Da Vinci Code — Dan Brown (2003)
CONTINUE THE CONVERSATION
Source texts
- Fooled by Randomness The Hidden Role of Chance in the Markets and in Life (Nassim Nicholas Taleb)
- Fooled by Randomness The Hidden Role of Chance in the Markets and in Life (Nassim Nicholas Taleb)