A number of people argue that Likert scale responses are properly treated as ordinal data, meaning that 1<2<3<4<5, but not necessarily that |1-2| = |2-3| or avg

A number of people argue that Likert scale responses are properly treated as ordinal data, meaning that 1<2<3<4<5, but not necessarily that |1-2| = |2-3| or avg(1,5)=3. Treating them as numerical data often (depending on what you're doing) implicitly makes some assumptions about calibration of the ordinal scale that might not be warranted (e.g. the distance between a "1" and "2" response might be very different than the distance between a "2" and "3" response).