Likert items are often used in social and health sciences. However, the format is strongly affected by acquiescence and reversed items have traditionally been used to control this response bias, a controversial practice. This paper aims to examine how reversed items affect the psychometric properties of a scale. Different versions of the Grit-s scale were applied to an adult sample (N = 1,419). The versions of the scale had either all items in positive or negative forms, or a mix of positive and negative items. The psychometric properties of the different versions (item analysis, dimensionality and reliability) were analyzed. Both negative and positive versions demonstrated better functioning than mixed versions. However, the mean total scores did not vary, which is an example of how similar means could mask other significant differences. Therefore, we advise against using mixed scales, and consider the use of positive or negative versions preferable.
