In science, social science or any other form of endeavor that uses data, especially statistical data, "Cherry Picking" is the term for picking out data that fits a particular hypothesis. Cherry picking can be done through malevolence or accident, and is almost always a bad thing to do. Purposely cherry picking data is usually considered to be unethical. However, it certainly seems to happen often enough.
There are also some times when it is unclear whether a process would be cherry picking or a sensible way to control data. For example, say that a researcher was trying to find correlation between median income and poverty rate in all the communities in a state. In such a case, excluding communities below, for example, 5,000 people might give a better view of the general trend, as well as saving the researcher time and confusion. Or it could be seen as a form of excluding data so that a certain result can be reached.
The other time when it can be acceptable to cherry pick is to disprove something. The fact that Fairbanks, Alaska has a higher percentage of African-Americans than San Francisco, California doesn't prove that the larger a city, the lower percentage of African-American residents it will have. However, it does offer a disproof of the contrary: as long as one example can be given, getting a perfect correlation in the opposite hypothesis will be impossible.