A surrogate variable is a variable that is used to fill in missing values during the data cleaning stage. Suppose we are considering the variable y as a candidate surrogate variable for attribute x. The variable y then must be:
Strongly correlated with attribute x
Independent of attribute x