MODULE #3 - CLUSTER ANALYSIS - MSA SEGMENTATIONThe goal of this deliverable is to perform both methods of Cluster Analysis.A Metropolitan Statistical Area (MSA) is a designated area outlined by the Census. It is a commonway that marketers with multiple markets look to advertise, in the how, the where, and how much,specifically around media and media mix.For marketing managers and media planners, trying to build a strategy for every single MSA theyserve is sub-optimal. Specifically, for media purposes, there are key considerations made around adplacement - radio, TV, billboard, on local transportation, etc. And it is widely accepted thatgeographic features about MSAs tend to be key factors in the decisions made in where, when, andhow to place media.For this module, you have a data set of Metropolitan Statistical Areas. You should recognize most ofthem and understand why some major areas are grouped together. Cluster Analysis allows a mediaplanner to plan a few separate, and distinguishing, strategies, rather than going in with a shotgunapproach.The Tasks:1. Run a Hierarchical Cluster model of all Continuous Variables (those with a Measure if "Scale"in SPSS) with the MSA in "Label Cases by" using the default settings. Add the Dendogram.Using the Rescaled Distance of 10, how many clusters does this suggest? Why is thispotentially problematic? (i.e., not useful)2. Why would even cutting the Rescaled Distance criteria of 5 be problematic?3. Now run a HC model the same way, but this time select the Within-groups linkage as theCluster Method. At a Rescaled Distance of 5, how many clusters does this suggest? Is thispotentially more useful? Is there potentially a more important way to build a meaningfulcluster model?4. NOW, try the HC model again as the default setting (Between-groups linkage), but this time,Transform Values to Z-scores. What is different about this model, according to theDendogram at a Rescaled Distance of 5?5. Do the model the same way, but run Within-groups linkage. Does this model perform betterthan Between-groups? (hint, the Rescaled Distance to create clusters is a good way to tell)6. Shifting course, run a k-Means Cluster model with 5 clusters. Again, use MSA as the label.Also, run the ANOVA table. What are the sizes of each cluster (i.e., how many cases in each)?What does this suggest about Clusters 1 and 5?7. Run the model again, but this time, save the Cluster membership number. Check the citiesfor clusters 1 and 5 (Hint: Crosstab Cluster ID by the MSA label). What can you infer aboutthese two cities?8. Describe what differentiates each cluster (hint, the variables that came up statisticallysignificant in the ANOVA table should tell you what is important in distinguishing theclusters).9. Select a city for each that might best summarize the cluster. (Note, there is literally no way tobe wrong on this without saying that a cluster with only one city in it is better represented bya city from another cluster)