A frequent objective with brand and communication studies is to measure brand image: that is, the perceptions that people hold of the main brands, how they compare and how they might occupy different positions in customers’ minds, either as having functional differences or differences in emotional positioning. Two ways to measure brand image are the use of rating scales and brand-image association.
1. Scalar approaches
With a rating scale approach, each brand is evaluated on a number of dimensions defined as those that are the key dimensions that discriminate between brands. Each brand is evaluated monadically, with the sequence of evaluating rotated between respondents. The rotation of the sequence order is important here as the way in which respondents rate one brand can affect the way they rate any following brands. How they rate the first brand on, say, ‘quality’ sets a benchmark for all subsequent brands. A slightly generous rating for the first brand, even though respondents think it might only be of average quality, requires increasingly positive ratings for any subsequent brands thought to be of better quality.
Respondents are only asked to evaluate brands that they are aware of from a preceding or earlier prompted (aided) brand-awareness question.
Figure 6.2 is typical of the self-completion question to evaluate brand image using an agree-disagree scale. Note that this is technically not a Likert scale. As we are not measuring attitude but perception, there is no necessarily positive or negative position for each dimension, only different brand positionings. The individual respondent scores cannot be summed in order to provide an overall attitude score.
The question in Figure 6.2 could equally have been posed as a bi-polar semantic differential scale. Care then has to be taken in defining the pairs of statements so that they have truly opposite meanings. For example is ‘traditional’ the opposite of ‘modern’, or should it be ‘old fashioned’?
The scalar approaches to measuring brand image provide strong interval data that can be used in a variety of ways, including the calculation of mean scores and standard deviations and the analytical techniques such as correlation, regression and factor analyses.
They do though suffer from two drawbacks. First, because they are completed monadically it is difficult for respondents to reference brands against each other. As discussed earlier, respondents may rate a brand for a particular attribute, only to find that for following brands they have not left themselves sufficient space on the scale to properly express the differences that they perceive between them.
The second disadvantage is that they can take a long time for respondents to complete. A list of 20 attributes for each of six brands requires respondents to complete 120 scales if they are aware of all six brands. At an estimated 15 seconds for each attribute for the first brand, and 10 seconds for subsequent brands, this can take over 20 minutes to complete. This adds to the potential fatigue and boredom of the respondents, the length of the interview and the cost of the study.
2. Attribute association
An alternative approach is the brand-attribute association grid. Here respondents are shown a list of brands and asked to say which brand or brands they associate with each of a series of image attributes. The image
attributes are either read out by an interviewer or appear on the questionnaire or screen for self-completion.
This is quicker because respondents only have to go through the list of attributes once. They also do not have to make such complex decisions about how well each brand performs on each attribute, only that it applies or that it does not.
Brands of which they are not aware will usually not be nominated as possessing any of the characteristics. Some respondents may nominate brands that they have previously said that they are unaware of to have certain characteristics (particularly for attributes such as ‘not well known’) but these can be identified at the analysis stage. If respondents really are responding with an image of a brand of which they are hearing for the first time, that can tell the researcher a great deal about the image attributes of the name alone.
Another advantage is that respondents can assess the full set of brands together. This makes it easier for them to make comparisons between brands, and determine that an attribute is or is not associated with one brand rather than another.
Figure 6.3 is taken from an interviewer-administered questionnaire from which the data has to be manually entered, but the arrangement of the layout could equally be from a self-administered questionnaire.
The coding numbers here have been arranged vertically rather than horizontally. This is for two reasons. First, if respondents should see the questionnaire, there is no suggestion of an order of priority amongst the brands. A horizontal arrangement would have Brand A always as code 1 and Brand F as code 6. Where coding is shown on self-completion questionnaires this can be a potential source of bias.
Second, it helps the researcher to think in terms of brand image profiles for each brand, and the data-processing spec-writer to write tables to produce that. It is more likely to be of value to the analyst to be able to see the image attributes associated with each brand rather than the brands associated with each image attribute. It also makes it easier to be able to analyse by respondents who have heard of the brand, brand users and non-users, those aware of the advertising, and so on.
The disadvantage of attributing image statements in this way is the loss of the degree of discrimination that would have been obtained had scales been used. It may be found, for example, that most respondents think that all brands possess certain attributes, whereas a scalar approach would have shown variation in the strength with which each brand is seen to possess them.
The level of discrimination can be increased by including opposite expressions of an attribute. Both ‘High quality’ and ‘Poor quality’ could be asked; ‘For younger people’ and ‘Not for younger people’. (‘For older people’ is not necessarily the opposite of ‘For younger people’ as the brand could be seen to be for both.) This doubles the number of attribute statements that need to be included, although it probably does not double the time taken to administer them. It effectively creates a three-point scale, with each brand nominated either for the point at each end of the scale, or not mentioned at all, which can be taken as the mid-point of the scale. The relationship of the association between the two end points is sometimes referred to as the ‘quality of the brand image’ and the extent to which the brand is associated at all with the dimension ‘the strength of the brand image’.
An alternative way to increase discrimination is to ask which brand or brands the respondent would choose if they were looking for one that possessed the successive image attributes. Respondents then tend to nominate only brands that are strongly associated in their minds with the attribute. This reduces the number of brands associated with each attribute, and demonstrates ‘ownership’ of attributes more clearly.
A disadvantage of the technique is that the levels of association are dependent on the brand set shown. This acts as the reference set against which each brand is judged. The choice of which and how many brands
are included is thus an important decision that can affect apparent brand positionings. Should the number of brands or choice set change over time, on repeat studies or tracking studies, there is a danger that comparability will be lost. A study may, for example, ask respondents to associate brands from a set of five airlines. If the number of airlines was to be increased to six in a later study, then we should expect to see the levels of association for all brands decrease. This is because the average number of brands associated with each attribute tends to remain reasonably constant, so that with more brands the average number per brand decreases.
Had one of the attributes been ‘innovative’ and the new brand introduced been Virgin Atlantic, a brand known for its innovation, then a substantial change in association for the remaining brands should be expected on this attribute. The frame of reference on this attribute will have changed, and brands that were previously thought to be innovative, in the context of the set asked about, will now appear to be less so. A similar change on this attribute would have been expected had Virgin Atlantic been substituted for another brand in the set, so that the total number remained the same. The levels of association recorded are not absolute, but are relative to both the number of brands asked about and the actual brands in the set.
When deciding upon the brands to use, it can be important to relate them to the attributes to be asked about. Thus, an attribute should not be included without very good reason if the brand set does not include the brand that has the strongest associations with the attribute. The false conclusion that a brand performs strongly on that attribute could easily be arrived at, because it only does so in the context of worse performing brands.
The data generated by this approach allow correspondence mapping, as well as correlation analysis and, with some transformation, regression analysis.
3. Measuring attitudes
Probably the most common way to measure attitudes is to use rating scales, whether it be to measure attitudes to brands, products, social issues or lifestyles.
Formulating the attributes or dimensions and statements used to measure attitudes can be a difficult task. In comparison, brand or product attributes or service attributes in customer satisfaction research are frequently easier to arrive at than are the appropriate set of attitude and lifestyle dimensions. The brand attributes to be measured are often very specific and easily identified (eg modern, value for money, effective).
Brand positioning or image attributes may be less tangible, but are often well defined within the brand positioning statement. In customer satisfaction research the dimensions to be measured are defined largely by operational factors, such as the cleanliness of a room, or a call centre operator’s ability to answer questions. These are usually capable of being expressed in a straightforward and succinct way.
Measuring less tangible attitudes, however, presents a number of other considerations that the questionnaire writer must take into account. Respondents may never have considered the issues that they are being asked about. They may therefore be more open to influence from the question wording or the inferences that they draw from the statements. Some of the issues that must be taken into account when compiling the statements to represent the attitudinal dimensions are:
- whether or not the statement is balanced;
- whether it leads the respondent to a specific answer;
- how the addition or removal of wording may affect how respondents answer.
In addition, the question or questions that are to be asked need to be considered in relation to:
- whether acquiescence or yea-saying is likely to occur;
- whether this is an issue that the respondent has given conscious thought to before being asked about it in the questionnaire;
- the optimum method of measuring the importance of the dimensions to the respondent.
Balance in attitudinal questions is generally achieved by presenting all aspects of the dimension as being equally acceptable. This is important because there is a tendency for people to agree with any proposition that is put to them. With this type of attitudinal question it is important to avoid writing questions where the answer is simply ‘yes’ or ‘no’ and to force respondents to make a choice between a number of options. There may be two aspects to the question:
Do you think that voting in general elections should be made compulsory or not made compulsory?
Or there may be more than two:
Do you think that women are better suited to bring up children than are men, or that men are better suited than women, or that both are equally suited?
The unbalanced version of these questions would be:
Do you think that voting in general elections should be made compulsory?
Do you think that women are better suited than men to bring up children?
These unbalanced versions are likely to lead to a higher proportion of the sample agreeing than would have chosen that option from the balanced questions. The evidence for acquiescence is strong. Schuman and Presser (1981) demonstrated it by asking the balanced and unbalanced version of the same question regarding the roles of men and women in politics in four separate surveys. The unbalanced version produced agreement with the proposition of between 44 per cent and 48 per cent across the four surveys. The same proposition was chosen by between 33 per cent and 39 per cent where the balanced question was used. Thus the use of the unbalanced form of the question added in the region of 10 percentage points. Differences of such magnitude were not found with other topics, so acquiescence would seem to vary between subjects and possibly between individual items within a topic. Questionnaire writers rarely have the luxury of being able to test each topic and item to determine whether or not it is likely to be susceptible to acquiescence. It is therefore good practice to treat all questions as if they are, and to write the question in a balanced format.
Whether or not the question is balanced, expression of the attitude must not lead the respondents to a particular point of view. A hypothetical example of such a question is:
Homeless people in our cities are a major problem and deter people from coming here. Do you think that the state should support homeless people or not?
The position of the question writer is quite clear. Only one aspect of the issue of homelessness has been highlighted, and this would be likely to lead respondents to a particular answer. The questions could as easily have been put as:
Some people find themselves without a home through no fault of their own, and then find it difficult to get back into work. Do you think the state should support homeless people or not?
The actual question is the same, but the information given to ‘assist’ the respondent in coming to an answer is biased in the opposite direction and is likely to lead to the opposite response from the first version.
With complex subjects such as this, the question writer has the choice of rehearsing all the pertinent issues as fairly and as equably as possible, or to ask the respondent to base their answer on what they already know about the subject:
From what you know about the issue of homelessness, are you in favour of or against the state supporting homeless people?
The extent of the wording change does not need to be as drastic as in this example in order to change the response. Schuman and Presser (1981) showed that a relatively small addition of a few words can change the response. In 1974 they asked the question:
If a situation like Vietnam were to develop in another part of the world, do you think the United States should or should not send troops?
To this question, 18 per cent answered that the United States should send troops. When the five words ‘to stop a communist takeover’ were added to the question, that proportion increased to 36 per cent. Similar increases were seen when the experiment was repeated in 1976 and again in 1978.
The additional words, in this case, were more than just a rhetorical flourish; they clearly led to a significant proportion of respondents assessing their position differently because they highlighted one particular aspect of the issue being asked about. It is unlikely that most market research questionnaires explore such emotive issues as was the prospect of communist takeovers in the United States in the 1970s, but the example clearly serves to show how small additions to the question can change the response, and the care that must be taken with wording the question. Just a few words can alter the tenor of the question or crystallize an attitude that was previously only vaguely held. Question writers should be constantly asking themselves whether the inclusion of particular words or phrases help the respondent, are just an embellishment, or in fact alter the basic question.
The case for piloting the questionnaire (see Chapter 10) is clear and should allow for alternative versions of attitudinal questions to be examined and tested whenever there is any uncertainty over them.
The extent to which responses are changed by an additional phrase or a small change in wording may depend on the extent to which the opinion had already been formed in the mind of the respondent prior to the question being asked, and how strongly that opinion is held.
Source: Brace Ian (2018), Questionnaire Design: How to Plan, Structure and Write Survey Material for Effective Market Research, Kogan Page; 4th edition.