Scaling in correspondence analysis

[axphil]

Dear Koichi,

I use the KH Coder to run correpondence analysis of words and variables. It works fine but I'm not sure whether scaling in plots is always the same. Especially if I apply for scaling "none" I have the impression that scaling in the plots vary. I mean the distance between 0 and 1 is different on axes X and Y. See:

How to read the plots? Is it a wrong impression on my side? Or how can I generate plots with axes that show the same distance between 0 and 1 (and other values)?

Best,
Axel

The version number of KH Coder: 3.Alapha.14b
My operating system: Mac OS 10.13.6

[ko-ichi-h]

If you want to fix the aspect ratio of X and Y axis to 1, please select "symmetric" or "symmetric biplot" as the "Scaling" option.

Here is a snippet from section A.5.8 of the reference manual.

Conventionally, KH Coder has used [none] as the default. This method uses the standardized scores output by the "corresp" function in R, such that plotting is performed, without fixing the aspect ratio to 1:1. Thus, results are displayed to the limit of the plot area, i.e., enlarged both vertically and horizontally. In effect, this is advantageous to quantitative text analysis that needs to display as many word labels as possible. To create plots called symmetric maps, which are used more traditionally, select the [symmetric] option. This method is the default setting for the statistical discovery software JMP developed by the SAS Institute (NC, USA). Finally, [symmetric biplot] is the default method used in SPSS (IBM, NY, USA). For details on these three methods, the user is referred to M. Greenacre (2007) Correspondence Analysis in Practice 2nd ed.

Regardless of which option you choose, basically, you can read the results in the following way. Here is a snippet from the detailed version of tutorial part 1. So, please also see Fig. 11 in it.

In correspondence analysis, uncharacteristic words uniformly found in all parts are plotted near the origin (0, 0) (i.e., the point at which the ordinate and abscissa are both zero) whereas words having strong characteristics are located away from the origin. For example, "Cuthbert" is plotted far from the origin in the upper left of Figure 11, meaning that the word has strong characteristics. We then ask, what are the characteristics of the word "Cuthbert"? The word is far away in the direction of "01-07" as seen from the origin, which means the word appears especially frequently in part "01-07". Reading the characteristics of each part from the words plotted in a similar direction as seen from the origin as above is the basic way of interpreting correspondence analysis.

In addition, it is also effective to see where each value of the variable is located. In Figure 11, "01-07" and "08-19" are away from other values, but "20-28" and "29-36" are close to each other. This means that frequently occurring words are similar for "20-28" and "29-36", which suggests that these two parts have similar contents.