Data about which you cannot tenably make the assumptions needed for a statistical analysis you'd like to undertake.
Yes, this definition is very boring, and that is on purpose. A lot of time and energy could be devoted to coming up with an inclusive and elaborate definition of ‘bad data’. That investment would perhaps do little to meet the real challenge manifested by bad data, so let us short-circuit the question of how to define the phrase. This boring definition is serviceable and constructive. It frees us from the definition question so that we might then focus on the greater challenge, which is to develop statistical methods that can handle the data.
Note that this definition does exclude things. In particular, data that are merely inconvenient or unwanted because of their implications are not bad under our definition. Data that disprove your pet theory may be unwelcome, but they are not bad in our sense.
Data are often bad because they are imprecise, qualitative, multivariate, or available in small or inconveniently large samples:
There may be many other reasons that data are bad.