In JMP, after selecting a numeric variable in the data table, Columns > Utilities > Create Categorization Formula can be used to divide the numeric variable into categories based on various rules.
For example, it is convenient to categorize data in which age is entered into teens, twenties, etc., or classify 80 points or more into a "pass" category and less than 80 points into a "fail" category for a test with a maximum score of 100 points.
Various options are available for categorization rules, as shown in the figure below. It is easy to imagine how to divide "categorize by equal width" and "categorize by percentile", but there are options "Categorization by Jenks natural taxonomy" There may be many people who do not know such a classification method.
Therefore, in this blog, I will show in what cases "Categorization by Jenks' natural classification" is used, and examples of classification when compared with other classification methods.
What is Natural Classification in Jenks
Regarding "Categorization by Jenks natural taxonomy", the JMP documentation has the following explanation.
By specifying the number of bins, you create bin cutpoints that minimize the variance within each bin and maximize the variance between bins.
This categorization method is A method that is compatible when classifying a certain numerical value using a map That's it.
In order to visualize differences in numerical values (e.g., population density, CO2 emissions, etc.) for each region, numerical data may be divided into several categories and color-coded maps (choropleth maps) may be created. At this time, the question arises as to how to divide the numerical data into categories, and Jenks' natural classification method is one way to do this.
An example of color-coding a map using Jenks natural classifications
Consider the following example.
example : Calculate the number of doctors per 100,000 people in Japan's 47 prefectures, plot it on a map and compare
The figure below shows when "Categorization by Jenks natural classification" is selected as the cut point and the number of categories is specified as 5. It is divided by halfway values such as 235.30583, but so to speak Places with large changes in value are separated as group boundaries It's going to be.
lower right Create Calculated Column Click the button to create new nominal columns at the end of the data table with five categories. In this example, the yellow column in the figure below shows the number of doctors per 100,000 population, and from this continuous column we created a new column (selected in blue) with five classifications.
*Data source: National Statistics Center SSDSE (Educational Standard Dataset)
Let's use the newly created columns to color code a map of Japan, and compare it with 190 to 390 divided into 5 equally spaced (equal width) increments of 40.
Using JMP's "Graph Builder", the map of Japan is classified into 5 categories using shades of blue.
The lower left figure is Jenks' natural classification, and the lower right figure is a map classified by equal width.
Both maps show that the western regions tend to have more doctors (per 100,000 people) and the eastern regions tend to have fewer.
Also, in the natural classification of Jenks on the left side, you can see that there are more prefectures painted in dark blue than on the right side.
Kyoto, Tottori, Okayama, Tokushima, Kochi, and Nagasaki fall into this category. It is clear that there are many doctors in these prefectures.
As shown earlier, Jenks' natural classifier uses Data with similar values tend to fall into the same category. Therefore, it is easier to grasp the characteristics of the data if you divide it by color intensity as in this example.
This time, I showed the usefulness of "Jenks' natural classification" with map painting data, but it is a useful method when visualizing data, not limited to maps. However, since it is not divided by a good number such as "190 to 230", it is necessary to be careful that people who see the results will have doubts as to why they are divided by such a half-baked number.
Finally, here's a graph plotting the number of doctors per 100,000 when divided by "Jenks' natural classification" and "equal width".
Blue: Jenks Natural Classification, Red: Equal Width
In this example, you can see that the way to divide the larger value is characteristic.
by Naohiro Masukawa (JMP Japan)
