This month we continue our series of looking at the world of all things quantitative. In this edition, Vicky McLellan looks at rating scales
Vicky McLellan 
18 Dec, 2007
Quantitative Corner... A quick guide to using rating scales
Rating scales are commonly used by quantitative researchers to measure many aspects of respondent behaviour and attitude.
There are a number of established scales used, such as the Likert scale to measure attitudes, the Semantic Differential scale to rate attributes or products, or the Juster scale to measure intent. These established scales are useful as they have been validated in published studies and, in certain cases, normative data can be created to benchmark the results.
On a scale of 1–5, where 5 = full of Christmas spirit and 1 = bah hum bug, how much Christmas cheer do you have?
However, more commonly, rating scales are tailor-made to meet the specific needs for each study. There are a number of things to think about in the design and analysis of these scales.
- To avoid respondent fatigue, do not have too many attributes tested together within one rating scale set (sometimes known as an “attitude battery”); try and keep it to a maximum of around 8 attributes.
- Make sure that all the attributes, when measured together, are of a similar type. If they are not, you should split them into separate questions.
- Make sure that the scale really matches the attributes to be tested. For example, with satisfaction surveys, it is very important to separate out those attributes that you should expect, and could only be dissatisfied with (e.g. punctuality, cleanliness, things working) from “performance” attributes you might be motivated by (e.g. quality of insight/food/atmosphere).
- Be creative! For example, use insights from qualitative research to build your own rating scales which meet the study objectives and engage the respondents.
- Think about whether neutral mid-points are relevant for your scale, and how they differentiate from “don’t know”. Also, if you do introduce neutral mid-points, think about whether that would compromise applying a mean score to the data (for example, for the scale “disagree strongly, tend to disagree, neither agree nor disagree, tend to agree, agree strongly”, I would think that the three middle statements are pretty similar and the scale would not warrant a standard mean score allocation of 1 to 5).
- Think about having as many points as is reasonable for a respondent to consider – or even use a sliding scale with online research, to generate a highly graded score. This extra detail can help the statistical accuracy.
- When it comes to the analysis of rating-scale data, take into account that different people, and different cultures, can have different frames of reference as to what is a “good” score and what is a “poor” score. This is important for performance/satisfaction rating scales in particular – for example, Italians tend to give very high scores; the Dutch give very low scores. There are ways around this, either at the questionnaire design stage (e.g. by using trade-off approaches) or at the analysis stage (e.g. by normalising responses).
Your response:
Login to respond to this article: