So what we're going to talk about in this session is how we can go about examining market structure using perceptual maps. And the perceptual maps that we're going to generally generate are going to be done using a technique I refer to as multidimensional scaling. So going back to the framework that we looked at in the previous session, what we're interested in understanding is not only how our company is seen, how our products and other offerings are seen, but also how we're seen relative to our competitors. Now are we seen as similar to our competitors, who exactly are our competitors, and along what dimensions. That's what perceptual maps and multidimensional scaling can help us answer. So the general question that we're dealing with, I want to know I am seen, I want to know who my competition is, and I want to know on what dimensions that we're seen as similar to those competitors. Do I score well on the quality dimension? On the value dimension? Or where do I score in those two different dimensions. Who's similar to me on one of those or on both of those. And also where are the potential opportunities in the marketplace. So I'm going to show you a couple of examples of perceptual maps to give you an idea of what this will allow us to do. Now these perceptual maps, there are a couple different algorithms that can be used, the most common being multidimensional scaling, that's built into most statistical packages. But what we're going to do is we're going to try to represent the location of brands in these perceptual spaces. So we're not dealing with physical space in this case, we're dealing with perceptual or mental space. All right, so what we're going to focus on in this session is, what's the data that's appropriate? How do we go about getting it? And then to some degree, what's the analysis that we need to do in order to produce these maps? All right, so if I'm interested in understanding how I'm seen, how my competition is seen, one way for me to go about doing that is to build a survey. Let's begin with qualitative research. Let's use the insights from qualitative research to construct a survey. I'm going to administer that survey. I'm going to get a lot of those responses. I'm going to run factor analysis, like we've already discussed. And I'm going to report out how do I score on different dimensions, how do my competitors score on different dimensions. So your output might look something like this, right? Which is fine, but if I'm dealing with trying to understand the entire industry and what does the marketplace look like, how do consumers see us relative to our competitors. This is not the most compelling output for me to try to understand that from. All right, so let's take a slightly different approach to this. Let's start with the same foundation. I want to begin with qualitative research to build my survey, build my survey, then once I collect those responses, I can construct a measure of similarity, where the inverse of similarity being the difference between two different brands or two different products. Well, if I have a set of differences or set of similarities, I can try to arrange the products in a perceptual space based on those distance measures. And if I can do that, I can show you a graphic that illustrates where, which brands are closest to you. The ones that have the smallest differences compared to you. Which brands are furthest from you, those are the ones that are most different from you. And we can also try to understand what are the dimensions on which you're similar and different from each other. And so this is an example of what a perceptual map looks like. This is perceptual map that was done using the auto industry as a context. It's actually produced through brand switching data. So this was produced using actual data, consumer switching from one car brand to another car brand. And so, what we start to see is, Which brands are similar to Porsche? Well, which ones are the closest to it? Well those would be Mercedes, Lexus, Jaguar and BMW. Which brands are most similar to Honda? Let's look for the brands that are closest to it. Volkswagen, Subaru, Mazda, Nissan and Toyota. Well, which brands are most similar to Cadillac? Well, Lincoln is the closest. As far as who's similar to Ford, that'd be Dodge and Hyundai. Now this is based on switching data. So these differences are based on saying, individuals switching from Ford to Dodge. How frequently does that happen? Well, switching from Ford to Dodge happens relatively frequently, compared to people switching from Ford to Porsche. That's why we see such a big difference in terms of the distance between those brands on the map. Now, while this map was built based on brand switching data, we can easily build a similar type of map using the results gathered from survey data. So another common example that's used to illustrate this technique is taking a perceptual map of the beer market. And just looking at where the brands are located, again, some brands closer together, so Budweiser and Miller relatively close together. Beck's and Heineken relatively close together. Coors Light And Miller Lite relatively close together. Those are the ones that respondents to a survey are saying are more similar to each other. Now, if oriented these brands on this particular set of axis one thing that would be natural for us is to say well let's try to name the axis. Let's also try to understand what's driving these different locations, what are the attributes associated with it. Well we could layer on the names of the attributes, so heavy versus light on the vertical axis, premium versus more budget branded, going on the horizontal axis. But if we asked surveys, as if I ask questions about the brand and how the brand is seen on different dimensions such as, being a beer for special occasions, being a beer popular with men, being less filling, being popular with women, being a good value. Imagine, let's overlay these two maps. And I can see, well brands that are in this quadrant are seen as heavy and premium brands. Brands that are in this quadrant are seen as good value blue collar. So they're heavy and budget beers. This quadrant here would be the premium light beers and then the premium budget beers. So we can profile brands based on consumer responses to attributes. That's going to allow us to understand how our brand is seen and if we had that information for multiple brands, we can see how we're perceived relative to how other brands are perceived, using those attributes as the dimension to understand what the axes are. And so, the way that we can go about trying to label these dimensions, we did it in a relatively ad hoc manner. But that works in our car example. You can think of foreign and domestic, budget versus premium vehicles. And we can think of heavy versus light beers, budget versus premium beers, as the dimensions there. But if you wanted a more formal approach for how we could go about doing this, regression does provide us a mechanism for doing that. That is if we look at the location in terms of an X, Y coordinate, a map coordinate, that's the outcome. That's the variable that I'm interested in trying to model using regression. And the way that I do that are using the attributes that I've collected about the brand as the predictors in that regression. Now since I want to see what drives position on the horizontal axis, what drives position on the vertical axis, and that would allow me to put names on to the different axes. So how do we go about building these maps? Well as I have mentioned, there are two different ways that we can collect the data necessary to do this. One is an attribute based approach, where I collect attribute information for each of the brands. The other is a similarity based approach. Now at their core, the data that we need is going to be the same. We need information about how similar or how different pairs of brands are. That's going to be the input when we're trying to produce these maps, right? So ultimately for brand I and brand J, I need the inter-brand distance. So I need a distance measure, D, telling me how different are brands i and j from each other. Where the smaller the difference, the more similar they are. The bigger the difference, the more dissimilar they are. Well, depending on which approach we're going to take, that's going to dictate how we construct this similarity measure. If we're doing an attribute-based approach, I'm going to collect information on multiple attributes that correspond to brand i, I'm going to collect information on those same attributes that correspond to brand j. And I'm going to calculate a distance measure. That is, how different is brand i and brand j on attribute p, in this case, it's the Euclidean distance, so I'm taking the square distance for attribute p. I'm going to calculate that square difference for each attribute, add it all up, and then take the square root. And that's going to give me a composite measure of the overall difference between brands i and j. All right, so the way this survey is constructed would be to say, literally, for each brand, we're collecting a set of attribute perceptions, right? The other way that we can get these dissimilarity measures is to directly ask for them, present to consumer respondent with here's brand i, here's brand j, on a scale of one to ten, for example, how similar are these two brands. So I'm directly eliciting them. So in the similarity-based case, the similarity-based approach, I'm directly asking respondents. In the attribute-based approach, I'm going to construct that difference measure myself. Now, one thing to consider is the kind of product or the kind of market that you're trying to understand. If it's a product or a service that consumers tend to think about very holistically, the similarity-based approach might be an easier one for them to wrap their heads around, because I'm just asking for a holistic impression of the brand or product. But if it's a product, let's say a computer, where people often think about a computer in terms of how fast is it, how much memory does it have, what's the brand, what's the processor, how much storage does it have, what's the video card. If it's a product, if it's a category where we are accustomed to thinking about the individual attributes, then it may be easier for respondents to provide information with regards to how a brand performs on a given attribute. In which case, you would construct that difference measure yourself. Well, once I have this matrix, I essentially have a matrix of distances and that's the input that we are going to use in constructing our perceptual map. Multidimensional scaling is going to take those distance measures that you provide in that matrix format. And it's going to figure out, how can I represent the distance measures that have been provided in a low-dimensional space? With most maps, you typically want to try to restrict yourselves to two or three dimensions, simply because once you get beyond that, it becomes very difficult to visualize.