Techniques and tools
Brand Mapping

Brand Mapping allows large data tables containing, for instance, product ratings across a number of areas of interest, to be represented in a two-dimensional map. It reduces vast quantities of numerical information into an easy to interpret, visually appealing format, using a data reduction approach which has analogies with traditional factor analysis. Brand Mapping is so named because it is typically used to visually represent the ratings or scores of competing brands across a number of benefits. It enables the correlations between specific brands and benefits to be mapped in a common 2-dimensional space. The information is usually collected using simple multicoded grid questions or semantic rating scales (typically 5 or 10 point scales).

The two most commonly used forms of Brand Mapping are Bi-plots and Correspondence analysis. The use of these techniques is best illustrated with an example[1].

Research was conducted to assess Brand Image across 8 brands in an industry sector using a simple battery of multi-response questions. For instance, respondents were presented with an image attribute (e.g. Business minded) and asked which, if any, of the listed brands they associated with that attribute.

A summary table, generated from a Tabulations package is shown below. This shows, in each column, the percentage of decision makers associating each attribute with a brand.

Notice that both rows and columns have been sorted in descending order of their respective totals. This shows that the brands further towards the left-hand side of the table received more ratings across all the attributes. This is quite common in this kind of research and reflects the fact that some brands are more dominant in the marketplace than others, and therefore have a much higher levels of awareness and usage by decision makers.

Similarly, if you look at the rows of the table, the attributes nearer the top received more ratings across all of the brands, reflecting the fact that some attributes are encountered more frequently in this collection of brands than others. Both of these effects (brand dominance and attribute dominance) may or may not be of interest when we wish to compare the various brands on the attributes. It depends on the studies objectives.

One way of eliminating these effects is to apply a correction to the table, to first calculate an expected value for each brand on each attribute, taking into account the average distribution of responses across brands and attributes and then work out the actual deviations from the expected values (in percentage point terms). This is achievable using a variant of the chi-square calculation, but we will not present this here. An alternative would be to plot the data using Correspondence Analysis.

The main distinction between Correspondence Analysis and Bi-plots is that Bi-plots reflect the absolute differences of brands on each of the attribute scores, whereas Correspondence Analysis reflects the relative differences for a particular brand (i.e. the peaks and troughs on different attributes within brand). Intuitively, Bi-plots correspond more with Table 1, whereas Correspondence Analysis corresponds more closely with the “deviations from expected score” described. Both plots can be directly generated from the table shown.

Bi-plot

A bi-plot of the data in Table 1 is shown below:

The vector lines for the attributes give us several pieces of information. The relative lengths of the attribute vector lines tell us which attributes are better at differentiating the various brands. The longer the line, the more the attribute varies across the eight brands. The angle between the attribute vector lines tells us how correlated the attributes are in absolute terms. An angle of between 0° and 90° means that the attributes are correlated positively, whereas an angle of between 90° and 180° means that the attributes are negatively correlated. An angle very close to 0° or 180° indicates a very high positive or negative correlation (respectively). An angle close to 90° shows that there is no or little correlation between attributes.

The rating of each brand with respect to each attribute can be determined by drawing a perpendicular line from each brand to the attribute vector line we are interested in. This can be done directly on the main Bi-plot or we can produce separate charts such as the one below, which shows the relative ranking of each brand on the attribute “business minded”.

The story told by the main bi-plot chart is as follows:

The bi-plot shows that there are two distinct groups of brands, consisting of three brands (brands 3, 4 and 5) and two brands (brands 7 and 8) respectively. There are also two brands which are distinct both from each other and the others (brands 2 and 6). This grouping is the same as that reflected in the Correspondence Analysis (see next section).

• The attributes business minded, competent, business partner, growing and responsible are highly correlated. These attributes can be hypothesised as representing a composite attribute which we can call “sound business”.
  
• The attributes responsive, supports business and convenient are also highly correlated and can be thought of as representing "sensitive to market needs"; the attributes supports local economy and appropriate fees are also highly correlated; good price and helps rural community are also highly correlated – however inspection of a third axis (not shown) shows that good price is more correlated with supporting the local economy rather than Helps rural community. This gives rise to a category we could call “good local value”.
  
• The individual attributes differentiating the most across all the brands are: business minded (in the sound business group), convenient (in the sensitive to market needs group) and supports local economy (in the good local value category). We know this because they are the longest lines.
  
• The story with respect to how the brands map onto the attributes is the same as can be interpreted from Table 1, but easier to visualise. To give an example, if we just focus on brands 1, 2 and 3 - Brands 1 and 3 rate highly on the “soundness of business” attributes (business minded etc), but low on “sensitive to market needs”. The exact reverse is the case for Brand 2. The can be confirmed by checking the summary ratings in Table 1. The Bi-plot makes these absolute differences on attributes and the relationships between attributes easier to visualise.

Correspondence Analysis

Contrast the above with a Correspondence Analysis of the data in Table 1.

This chart can be interpreted as follows:

The vertical axis on this chart represents, in the upward direction, “good value” (appropriate fees, good price) and in the downward direction “sensitive to market needs” (convenient, business partner, responsive, supports business). The horizontal axis, in the right-half of the chart can be thought of as representing “locally orientated” (supports local economy and helps rural community); in the left half of the chart the axis represents "sound business" (business minded, competent, business partner, growing, safe and responsible). The chart can be divided into 4 quadrants along these lines (Top left = good value and sound business, Top right = good value and locally orientated, Bottom right = sensitive to market needs and locally orientated, Bottom left = sensitive to market needs and sound business).

Notice how the individual attributes fall in the quadrants – for example, the attribute “growing” is equally high on good value and sound business (this makes sense as those that are growing are likely to discount); the attribute “helps rural community” is also high on the being sensitive to market needs whereas the attribute “supports local economy” is higher on good value. We can postulate from this that helping the rural community is more about being sensitive to the needs of this community; whereas supporting the local economy is more synonymous with providing cheaper goods.

The position of the brands with respect to specific attributes represents their perceived ratings in relative terms. In other words, on the Correspondence Analysis plot they tend to be closer to attributes which they peak on relative to other attributes. This closeness does not reflect their absolute score on attributes.

Using the interpretation of the chart quadrants discussed, Brands 7 and 8 peak more on good value and sound business relative to other characteristics; Brands 3 and 4 peak more highly on sound business and sensitivity to market needs; Brand 5 peaks on sound business alone; Brand 1 on sensitivity to market needs alone; Brands 6 on both good value and locally orientated; and Brand 2 on locally orientated and sensitive to market needs. Although I have expressed this by applying my own labels to groups of similar attributes, this interpretation could equally have been made from Table 1.

Notice that the clustering of Brands in the Correspondence Analysis reflects the clustering we found in the Bi-plot (see first bullet point in Bi-plot section). Broadly, we can say that there is consistency between the two plots, though they look at the data in slightly different ways.

Note that there are often addition dimensions (3rd, 4th etc) in both Bi-plots and Correspondence Analysis, though we usually only plot the first two axes as these usually explain the majority of the variation in the data.

Conclusion

These charts can help us visualise the results and their implications from different angles. Along with other analysis, it can help to assess which attributes are dominated by particular brands and which attributes are unoccupied. This helps to inform strategy i.e. where to take the client’s brand in the future.

Additional features of the Bi-plots and Correspondence Analysis not shown are the ability to project onto the chart how changes in a brand’s attribute ratings will effect its position relative to other brands. The interactive software for Brand Mapping also allows us to do the reverse of this i.e. find out by how much and in what direction the brand ratings have to change to get to a desired position on the chart (we can select the position that we want the brand to move to). This obviously helps a great deal with making the findings actionable. We can also map snapshots over different time periods onto each other, facilitating the monitoring of how perceptions are changing between waves of tracking studies.

Please contact Gary Bennett for further information (garyb@logitresearch.com).

[1] The raw data for this illustration is borrowed from Jenni Romaniuk’s and Byron Sharp’s paper in the International Journal of Market Research Vol 42 Issue 2 (2000)

< Back to Techniques and Tools Home Page