Ed Yong asked on Twitter “Explain principal component analysis to a schoolchild in a tweet.” Since I can’t explain PCA eloquently, I found this interesting and wanted to keep a record of the replies for future reference. Here are some of the modified replies, with my favourite first (and the rest in no particular order):

• Something nerds do to show off and that the rest of us secretly don't understand but always look knowingly when it's mentioned. @richboden

• Compare the height, eye color, hair length, etc. of the people around you. Where do you notice the most difference? @holtchesley

• It's a way to take any clues or observations you have about something and use math to see how much you can figure out about it. @ShipLives

• Capturing the essence of a data set in a reduced form. Think of describing the premiership in terms of Man U, Chelsea and Arsenal @fastconvergence

• Imagine lining the crayons in a box up based on their color similarity. Now, in the other direction, separate them by the wear. @APV2600

• "If you have to talk about something really complicated but you can only use two or three numbers, what do you do?" @blakestacey

• PCA is like picking cookies with the most chocolate chips and seeing where they came from in the jar. @holy_kau

• If you have a jellybean-shaped cloud of points, PCA figures out long side and short side of cloud and squishes it to get a ball. @geomblog

• PCA is a way of finding basic similarities between many different things - linking complexities by their simplest similarities @aimeemax

• PCA is a math method that converts data into a map. Objects (countries) that are closeby have similar parameters (coordinates) @scimomof2

• It's like representing a road with a line on a map. The road has width and length, but the length is sufficient for most purposes @CaldenWloka

• PCA is like taking all your responses to 'what is PCA' and condensing their essence into a single retweet. @ErikJCox

• A colleague once summed up PCA as a way of letting data decide for itself which patterns best describe it. @J_Liptak

• PCA tells you which features best describe the differences between items in a group, like colour in a box of lollies @pickleswarlz

• PCA of leaf shape would first give size, then fatness (width:length), wiggles of leaf edge, etc. 1st parts give largest variation @mickresearch

• Imagine collecting all the rocks in the yard, but you can only keep 10. PCA picks the ones with the most common color & sizes @kristenobacter

• Best soccer teams: do they score more or have the best defense ? PCA ranks values of teams based on multiple data of this kind. @tomroud