Rating Scales: Research Objectives, Analysis Techniques and Scale Selection
Imprinting is a phenomenon best known in geese whereby the first movable object that a newly hatched gosling sees becomes imprinted in their memory. The social bond that develops through imprinting entails an addictive process that is mediated by the release of endorphins. They follow this object around unquestioningly regardless of whether or not it is their mother.
One Scale Does Not Fit All
It appears that in marketing research, a similar phenomenon can be observed in the adoption of rating scales. Once a scale has been selected, it is often employed in every project thereafter, seemingly regardless of the changing research objectives. There is no one scale that is appropriate in every situation.
Selecting a Scale
Certainly, there is a plethora of rating scales to choose from with important issues to consider including the following:
- Number of items to use (7 – 11 most commonly referred to as optimal)
- Which anchor labels to use
- Full or partial labelling of the items
- Numeric labelling of items (studies have shown that respondents have a different interpretation of the same scales when one is numbered 0 – 10 and the other, -5 - +5)
- Whether or not to have a neutral point, and the list goes on.
Objectives and Analysis
Whilst these issues are very important, more important is that they are addressed in light of the research objectives, which should be aligned to the analysis approach. Choosing the wrong scale may limit the analysts’ ability to perform the technique required to achieve the research objectives. For example, a scale such as 3-point Likert scale with labels such as Satisfied, Neutral, Dissatisfied, produces ordinal variables. Where the research objective is to model satisfaction using multiple regression analysis, such a scale choice is problematic. The lack of ‘intervalness’ of such a scale results in the regression interpretation potentially being misleading. Such problems can be easily overcome by simply following rules of thumb relating to the minimum number of classes in the ordinal independent variable (Achen, 1991, argues for at least 5; Berry, 1993, states five or fewer is "clearly inappropriate"; others have insisted on 7 or more.)
Testing the Scale
A final point to consider is the need for an objective testing not linked to any specific project that can be used to assess the validity and reliability of a scale. Having such a tool would provide a guide that could be used in conjunction with consideration of research objectives, to choose the most appropriate scale.
Scale selection is really the precinct of survey designers. Consequently, Forethought looks forward to the forthcoming book of Krosnick & Fabrigar titled, ‘Designing Good Questionnaires: Insights from Cognitive and Social Psychology.’ Oxford University Press, as it plans to examine many unresolved scaling issues.
