Product Positioning: Multidimensional Scaling/ Market Baskets with Python
We’ve all seen perceptual maps—a rendering of consumer perceptions of products or brands. Usually companies ask consumers to rank the products/brands and place average ranks in the dissimilarity matrix. Similarity judgements are useful in product categories for which attributes are difficult to identify or describe such as style, look, odor, or flavor. The resulting dissimilarity matrix then can be used as our input to cluster analysis as well as product positioning. What I love about marketing analytics is that it’s always about people. We cannot possibly steer our analysis away from consumer preferences.
To practice product positioning, I found a perfect data set from the 1995 Wisconsin Dell Case. This study draws data from 1698 face-to-face interviews conducted with visitors to this entertainment resort. We have the demographic data of these visitors and data about whether they had participated in or were likely to participate in any of 33 entertainment activities. The response for these activities constitute a binary response matrix, which is perfect for our multidimensional scaling. Obviously for this specific data set, there are various methods of analysis and presentation such as market basket analysis and hierarchical cluster analysis, but I need some practice on multidimensional scaling so that’s that.
Multidimensional scaling (MDS) takes messy dissimilarities as inputs and provides a clear map as an output. I call it the essence of data visualization. When we try to identify underlying or latent dimensions in the data, for clarity, we really want to reduce the amount of data and the number of dimensions being considered. It is hard to see relationships in large matrices of correlations or dissimilarities, so let’s appreciate the simplicity of MDS maps. We can totally do this with excel, but since we’re not, we got to be careful with typos while entering that thirty three binary variables.
Wallah! Here we get our product positioning map. As we can see from the map, there’s a group of activities in close proximity: Boating, shopping, swimming, and casual dining. These are the most popular activities. In many research contexts, activities closer to each other are similar to one another, hence, potential substitutes for one another. This makes sense as we can see there seems to be nothing close to Bungee Jumping.
Multidimensional scaling is just one of the multivariate methods (Factor analysis, the Language of principal components…) we use to study relationships among many variables. When various methods yield decidedly different results, which is unfortunately common, we will have to choose the method that makes most sense to management. This is why I believe business literacy is sometimes more important than one’s technical skills.
