NewsAndAnalysis.org

Explanation of Ideology Clustering

Return to the survey.

This page attempts to explain Ideology Clustering through a simple example and by contrasting our method with the traditional type of survey (or poll) with which you are likely already familiar.  A more detailed explanation of the tool will be included in our final report.

To begin, consider a simple survey designed to determine the car people prefer:

1. Which color car do you prefer: blue or red?

2. Which style car do you prefer: 2-door or 4-door?

Once the data are collected, traditional statistical survey analysis may find that:

1. 4 of 10 (40%) respondents prefer blue cars, and

2. 3 of 10 (30%) respondents prefer 2-door cars.

From these results, it is not known whether people who prefer blue cars also prefer 2-door cars, etc.  Some surveys make an effort to connect several questions together, but for a complex set of issues, such as those that make up a political platform, the number of options becomes too large for traditional analysis.

Our tool uses a special type of analysis called "clustering" to determine for each of the 10 respondents above whether they prefer blue or red and 2-door or 4-door cars and how important each of these features are to their overall choice.  Our tool focuses on determining each person's complete ideology (in this case, about a car) for comparison with other ideologies, rather than determining an entire population's opinion about a single issue, such as color.  For this reason, we label our survey technique "ideology clustering."

Ideology Clustering would ask:

1. Which color car do you prefer: blue or red?  How important to you is color (low, medium, high)?

2. Which style car do you prefer: 2-door or 4-door?  How important to you is the number of doors (low, medium, high)?

Each of the ten dots in the graphic below represent a single respondent and their answer to the two questions.  The location of the dot on the graphic conveys the person's complete ideology.  People who prefer blue cars are on the left side and red are on the right.  People who prefer 4-doors are on the top and 2-doors are on the bottom.  If a person does not think that color or doors are important, then their answer (dot) is close to the center (note the shading: the colors blue and red, for example, are lighter toward the center to indicate they is less important there).  If they think that color or doors are important, then their answer (dot) is toward the outside of the figure.

Although these are the same results reported above, more information is provided by this technique.  Notice two distinct clusters: one in the upper right and one in the lower left.  We now know that 50% of respondents prefer red, 4-door cars and that 20% of the respondents are near the border line.  Another 20% of respondents seem highly committed to blue, 2-door cars.  If you were a car manufacturer, how many blue, 4-door cars or red, 2-door cars would you make?  Not many, I hope, because neither respondent claiming a preference for blue, 4-door cars was very committed to their answer.

Sample survey results

The above example has two questions, or "issues," and therefore can be depicted on a two-dimensional graphic.  By contrast, our political survey has 14 questions (issues), so it would require a 14-dimensional graphic (which is impossible to display).  Therefore, we will employ a mathematical technique to identify the clusters for us.  The technique we will use is called a Fuzzy c-Means Clustering Algorithm.  Using fuzzy set theory, this algorithm will find the center ("centroid") of the various clusters of results.  These centroids will represent the average opinion of each cluster (a group of likeminded respondents), and thus, the makeup of each political platform.  If you've stayed with me this far, you've earned a treat.

Return to the survey.