Responses on questionnaires are often based on Likert Scale scoring (e.g. answers on a scale of 1-5). In some cases, there is a difference in the direction of these scales, for example an answer of “1” might be a low score on some questions, but a high score on others.
For example, for questions which are “positively phrased”, a strong agreement with the question indicates a high level of the attribute being measured. For example, the item “I like myself” in a self-esteem questionnaire. In contrast, for items which are “negatively phrased”, a strong agreement with the item indicates a low level of the attribute being measured, for example, the item “I dislike myself” in a self-esteem questionnaire.
Before conducting any analyses, including computing reliability statistics on a group of items, adding up, or finding the mean of items, it is important to ensure that all items are consistent with each other in terms of what an agreement or disagreement mean for the attribute being measured. To do this, a good idea is to reverse score all the items which are “negatively phrased”, so that all variables are consistent.
The aim of reverse scoring is to re-code the responses so that a high score is transformed into the corresponding low score on the scale. For example, in a 5-point scale, a 4 is transformed into a 2, and vice-versa.