Country-Level Segmentation Using K-Means Cluster Analysis

Cluster Plot

Using data from the International Monetary Fund World Economic Outlook 2009 and The Global Enabling Trade Report 2009 from the World Economic Forum, I thought that it would be interesting to develop macro-level country clusters based on three variables: percent growth in gross domestic product from 2008 to 2009 (GDPG), percent population growth from 2008 to 2009 (POPG) and an assessment of the openness of the country to business endeavors (OPEN: the Enabling Trade Index – ETI) from 2009. Complete data for 119 countries is available, attainable and used in this example.

Cluster analysis is a statistical technique that groups individuals (in this case, countries) into clusters so that the objects in the same cluster are more similar to one another, based upon the characteristics investigated, than they are to individuals in other clusters. All data was analyzed using the open source R statistical package. One of the limitations of cluster analysis is identifying the optimal number of clusters to develop. Using model-based cluster analysis resolves this problem by mathematically determining the optimal number of clusters. For this project, the optimal number of clusters is determined to be five.

On the basis of the k-means cluster analysis, with a target of five clusters as suggested by the model-based cluster analysis, distinct clusters emerge from the data with sizes of 25, 17, 17, 28 and 32. In theory, the opportunity for marketing and business in the countries contained within each cluster should be similar based on the growth (or constriction) of the overall economy, population growth and the overall business climate within the country.

Cluster Results

Cluster 1: Armenia, Austria, Bosnia and Herzegovina, Cote d’Ivoire, Cyprus, France, Gambia, Greece, Indonesia, Jamaica, Lesotho, Macedonia, Madagascar, Namibia, Paraguay, Philippines, Qatar, Senegal, Slovak Republic, Slovenia, South Africa, Switzerland, Uruguay, Venezuela, Zambia

Cluster 2: Bangladesh, Burundi, Cambodia, China, Eqypt, El Salvador, Ethiopia, Guyana, Honduras, Jordan, Malawi, Mauritius, Mozambique, Panama, Tajikistan, Tanzania, Uganda

Cluster 3: Australia, Chad, Kazakhstan, Korea, Kuwait, Mexico, Mongolia, New Zealand, Nigeria, Norway, Poland, Russia, Saudi Arabia, Sweden, Turkey, Ukraine, United Kingdom

Cluster 4: Algeria, Azerbaijan, Bahrain, Belgium, Brazil, Canada, Chile, Columbia, Croatia, Czech Republic, Denmark, Estonia, Finland, Germany, Hungary, Ireland, Italy, Latvia, Lithuania, Luxembourg, Moldova, Netherlands, Oman, Portugal, Romania, Spain, Taiwan, United Arab Emirates

Cluster 5: Argentina, Benin, Bolivia, Bulgaria, Burkina Faso, Cameroon, Costa Rica, Dominican Republic, Ecuador, Ghana, Guatemala, Hong Kong, India, Israel, Japan, Kenya, Kyrgyz Republic, Malaysia, Mali, Mauritania, Morocco, Nepal, Nicaragua, Pakistan, Peru, Singapore, Sri Lanka, Syria, Thailand, Tunisia, United States, Vietnam

What does this mean for global marketers? The results are mixed. Countries in the first cluster, on average, are growing in population, have respectable levels of openness for business but their macro-level economies shrunk by an average of 10.84 percent.

Countries in cluster 2 show growth in GDP and population, but have the lowest average score on openness for business. Countries in cluster 3 experienced the largest level of economic loss for the 2008-2009 period, on average 26.56 percent. Countries in cluster 4, on average, enjoy the highest levels of pro-international business policies (as measured by the ETI). Finally, the countries in cluster 5 have the second-best macro-level economies (on average), positive population growth but some trade policies that are edging towards protectionist, as indicated by their average ETI.

The benefits of using cluster analysis for macro-level segmentation are apparent. The more difficult task is determining how to use the results in a way that supports your market entry decisions or global business strategy. Clearly, the most telling result is that this type of blog post can only come from a self-confessed nerd with nothing better to do with his time.

Share

Leave a Reply