False Pattern Recognition

Send us Fan Mail With enough data points it is possible to find patterns which do not exist. This is often done by picking certain data points that fit a pattern and arbitrarily ignoring the rest of the dataset. Outliers are a part of statistical analysis. They are data artifacts that don't appear to fit the rest of the dataset. Outliers can always tell us something important about the dataset, sometimes it's just something about the reliability of the measurement system involved. But sometimes it's something more specific about the circumstances surrounding that particular data point. I think people (who have no formal training in statistics) hear about outliers and they think this is a way to fix or shape datasets. They can conjure an image of statistical analyses having useful data points removed as outliers and therefore skewing the entire analysis. This definitely does happen but the dishonesty is easily found in the actual data that was measured. Any other statistician can expose this dishonesty very easily. In real statistical analysis, the removal of every outlier must be justified. If a large amount of data must be ignored to produce a wanted result it's a certainty that mathematical fraud is at play. Some examples of situations in which people have found definitive patterns that weren't intended and don't mean anything at all (and attempting to use them to build some larger narrative will undoubtedly fail) White Static on old analog televisions (almost any shape can be identified if you focus long enough) Faces in clouds (Our brain is alctually engineered to recognize faces so the fact that cloud shapes appear so often as faces shouldn't be surprising to anyone at all) Anagrams of children's books with demonic expressions (I need to find some more specific examples of this but it's a clear example of how looking for an unintended thing can lead to hilarious results) Qanon message links always had a series of numbers and letters in them. Qanon analysts would attempt to find patterns in these sequences. The types of conclusions they drew were always linked to what they awere *looking for* (confirmation bias). More neutral analyses showed other things. Links https://en.wikipedia.org/wiki/Apophenia https://www.straightdope.com/21342381/do-anagrams-in-lewis-carroll-s-poems-prove-he-was-jack-the-ripper https://www.cnet.com/tech/services-and-software/qanons-coded-conspiracy-messages-look-like-random-typing-says-analyst/