In general, a large enough sample of Likert scale data eventually conforms to the central limit theorem, and both low scores and high scores balance out. But in working with a small sample, what is the guideline for correcting underlying response data? Not talking about response bias, but rather mistakes like reversing scale, e.g. selected all 1's, but the comments clearly indicate it should have been all 5's. Do you throw away the data, include it, correct it, correct with confirmation from respondent?