Multicollinear

Part of speech: adjective

Definitions

1. Describing predictor variables in a regression that are so closely linearly related to one another that their individual effects are hard to separate
2. Characterizing a set of explanatory variables in a statistical model that move together so strongly that they nearly duplicate each other's information
3. Referring to independent variables that are highly intercorrelated, which makes estimates of their separate contributions unstable and imprecise

Etymology: "Multicollinear" is about as transparent as technical vocabulary gets: it joins the prefix "multi-", meaning "many", to "collinear", which itself comes from the Latin "com-", "together", and "linearis", "of a line". Put together, the two parts describe several variables that lie along, or move with, a common line, so that they cannot be cleanly told apart. The idea behind the word is that when a statistician fits a model using several predictors, two or more of them may be so closely related that they carry largely the same information. Their effects on the outcome then become hard to separate, and the estimates for each can swing wildly with small changes in the data. The term names that overlap, and the problem it describes is a practical one for anyone trying to read a regression table.