Analysis of Rating Scales for the Measurement of Attitudes and Perceptions

It is still common today to see questionnaires with Likert Scale items concerning very different variables being used to capture data on aspects as varied as possible that are to be investigated by the research work. This is perfectly alright if each of the questions is to be treated as standing on its own and is not intended to add up to a measure of a single variable. This, however, has the problem of inadequate sampling of items to come to any meaningful measure of persons on that set of multiple variables, with as small a standard error of measurement (SEM) as possible. Each variable to be measured is best put on a single rating scale, with items being replicated a sufficient number of times to reduce the SEM. There can be more than one rating scale in one questionnaire, but they should obviously be placed in separate sections, and their analyses done separately. This paper discusses a specific example of the measurement of attitude towards teaching and perceptions of subjects’ own teaching knowledge and skills, and how to measure their changes over time, through the anchoring of item calibrations, using a Rasch model.