Part 2. MM4XL Tools > 1. Strategic Tools > Brand Mapping > How To Interpret Brand Mapping
Brand Mapping How To Interpret Brand Mapping Brand Mapping produces positioning maps applying correspondence analysis, which is a broadly used multivariate technology. Users of Brand Mapping are strongly encouraged to refer to Benzcri (1973, 1992) and Greenacre (1984, 1993) for a comprehensive review. How does Brand Mapping work? According to statistics, Brand Mapping positions in a low dimensional space one bubble for each column and each row of any rectangular table of values. The position of the bubbles in the space is derived measuring the variation of each profile from the average profile. The variation is measured with the Chi squared statistic and the bubble position corresponds to a Chi squared distance. In other words, Brand Mapping performs a basic task of great help to marketers: it pulls apart items according to their difference, and it makes all differences visible at once, thanks to a clear and easy to interpret output. Now, given that product positioning is largely based on segmentation and differentiation, it goes without saying that Brand Mapping is the tool of preference for visualizing the competitive structure of markets as perceived by customers and also: - to exploit product (re)positioning concepts
- to circumscribe the competitive environment and select direct competitors
- to find new product opportunities (market gaps)
- to evaluate new product concepts
- to foster strategic reasoning.
The fostering of the strategic thinking today is a major issue in many large companies. Brand Mapping in many cases can be a great starting point. Interpreting Brand Mapping We state here some general principles, which should be applied with caution: - The closer two bubbles, the higher their association.
- The more one bubble moves away from the center of the map, the more one or more elements in its profile, characterize the profile itself.
- A bubble tends to be positioned in a space corresponding to the attribute category prominent in its profile.
The route we suggest when looking at brand maps is as follows: - Check the amount of inertia covered by each axis and figure out how much variance exists in the data. The lower the inertia, the smaller the variability, hence the less differentiated the profiles. This may have strategic implications, e.g., when reasoning is applied to strategic positioning.
- Identify any outlier points, as they can falsely affect the dimensionality of the map.
- If needed, focus the map by rescaling the axes (simply reducing the max and min values of each axis).
- Plot row points and column points on two separate maps.
- Look at the squared cosines to identify any poorly displayed bubbles.
- Identify any evident bubble segments on the 2D and 3D space.
- Assign a name (label) to axes and regions of the map, when possible.
- Double-check the accuracy of the raw data.
- Use your prior knowledge to interpret solid maps with strategic eyes.
Note: A map is the closest picture to reality that the analysis can display in two dimensions, and it is not perfect unless it displays 100% of data variance (inertia). The amount of variance displayed should always be kept in mind when interpreting positions on a map, and marketers should look at inertia with interest. Indeed, inertia gives an idea of the level of spread across bubbles, so the further apart the bubbles lay, the broader the market space available for new offers. The bullets in the outline above and more are discussed in the Examples chapter. What is in the output? Depending on the options selected, Brand Mapping prints by default, a map and two or three tables of figures. An explanation of how to interpret both the map and the figures follows. The numerical output The option that re-scales the raw data as a percentage, using either row or column, prints a table at the top of the output region, as shown below. This new table is the input to Brand Mapping. The next table shows the amount of inertia that each principal axis accounts for. Brand Mapping transforms and reduces the dimensionality of the original data set, and makes it possible to display the largest amount of variance on a two-dimensional map. The preceding eigenvector is always larger than the next, and the inertia % makes these values more readily comparable. In the case below, the first axis accounts for 60% of inertia, most of the 'meaning' in the distribution of points on the map. The first two axes account for 87% of total inertia, which is a very good map to work with. Unfortunately it is not always easy to achieve such good results, and much depends on the input data set. The larger the number of points to display, the lower the variation accounted by the first two dimensions. Tip: Remove unnecessary columns and rows of data in order to improve the quality of the analysis. The table below contains all necessary information to make a critical evaluation and an objective interpretation of the bubble points. The upper part of the table above shows the coefficients for the column points and the lower part shows the coefficients for the row points. Each row describes one point. The mass can be interpreted as a relative frequency, and the sum of all masses in each direction, rows or columns, adds up to 1000. According to Greenacre (1993), the original coefficients are weighted times 1000 by Brand Mapping, so to make the report more readable. The mass of products can be seen as the market share, although this interpretation does not work when the raw data is re-scaled as a percentage. Re-scaling the mass may be useful to reduce the effect in orientating the map of large, quasi-monopolistic market leaders. The inertia values of one point measure its contribution to the total inertia of the low-dimensional space. The lower the inertia the lower the variance in the data, and the higher the inertia (up to a maximum number of axes, three in our example) the more distant are the points from the origin of the map. The column Inertia shows the relative frequency of the inertia values. In our example, Colgato absorbs 41% or 410 of total inertia, and as expected on the first axis, it lies well away from the other bubbles. Next to the inertia values are the 3C's: Coordinate, Contribution, and Squared Cosinus. All points have one of each for each of the principal axes or eigenvectors. The coordinates are needed to position the bubbles in the low-dimensional space, or map, and are obtained with a process of value decomposition in basic roots for which we refer the reader to the relevant bibliography in appendix. The points placed on a map exert a kind of magnetic attraction of the orientation of the principal axes, which is measured by the contributions. The higher the mass of a point, the stronger its influx in orientating the axes. In our example Colgato is the major contributor (727) to the orientation of the first and most important axis. This impact can sometimes impair the efficacy of the visual display and may be rectified by computing either row or column percentages. The squared cosinus measures the correlation between point and axis. In a two-dimensional display, the Quality of the representation of a point on the map is given by the sum of its first two squared cosines. We deliberately chose this example to highly correlate Colgato and FreshM, respectively, with the first and second axis. In this analysis, only the 2D quality of Good price is not well represented. The sum of all squared cosines of one row is equal to 1000, when all axes are displayed. The Map Brand Mapping produces a scatter diagram and plots points in the form of bubbles. It is the bubble distribution on the chart that gives a meaning to the axes, and not the other way around. The axes have no meaning unless we interpret the point distribution. The map produced with Brand Mapping is also called dual display, as row bubbles and column bubbles are displayed in two different low-dimensional spaces, which are then combined in order to show all bubbles on the same map. The analysis puts data in corresponding rows and columns. This is the reason why the distance between one product and one diagnosis should not be interpreted, and the only meaningful element is that of the angle created by the two bubbles. Only with one unit in one space (e.g. products, corresponds to one unit in the space of diagnosis) is it safe to interpret the distance between the two. The level of association between rows and columns is computed as follows: In the picture below, the yellow bubble (product A), forms an angle with each of the three diagnosis types. Although diagnosis 1 lies further apart from product A than any other diagnosis, the angle formed by the two, is the smallest. This implies that the association between product A and diagnosis one is the highest among the three diagnoses. The angle q in degrees of one point versus the abscissa axis is computed with: With x and y being the first two coordinates of point i. The angle q between a product and a diagnosis can be computed by subtracting the angles of the points with the x-axis. There are three cases (see picture above: q1, q2 and q3): - Case q1: The angle between points is acute, q < 90. Points are positively correlated and highlight the over-representation in the raw data. Leaders are usually over-represented in the market segment they dominate.
- Case q2: The angle between two points is more or less squared, q 90. The points do not interact and show very dissimilar profiles.
- Case q3: The points form an obtuse angle, q > 90. The points are negatively correlated, which means that a product is under-represented (low share) in a given market segment.
The picture above lets us flag one other aspect related to the interpretation of brand maps: the further the distance of one bubble from the origin, the more differentiated its profile from the average profile. If the bubble is a diagnosis, this is a very notable difference, meaning that only one or few products are associated with it. If the far spaced bubble is a product, this means it has quite a different profile from the other products. Sometimes analysts look at wide-ranging market environments, and others tend to work with closely related ones. It is down to the goal of the particular analysis. Tip: When your analysis is done, take a minute or two to refine it. Rescale the axis to make the map more easily read. When working with time series, connect the years with a line. Play with background colors and bring labels with shadows and borders to the front. Insert labels with prior knowledge, such as launch dates, company names, price levels, etc. Excel allows changing virtually each element of a map. These are the basic tools one needs to start interpreting Brand Mapping. The topic however is much more complex than this short summary. For this reason we encourage users of Brand Mapping to read Greenacre (1993). |