Effect size measures for "Text Dispersion Keyness"

[elco...@gmail.com]

Hi Laurence,

I found something peculiar in calculating keywords. As a likelihood measure, I selected "Text Dispersion Keyness (4-term)" because I was interested to get keywords that are representative of all the texts in my corpus. However, I noticed that the effect size measures take word frequency, and not dispersion as their base. That is, the effect size won't tell me how big the difference is between the amount of texts where the keyword appears. It continues to tell me the effect of the difference between the frequencies of the word in the target and reference corpora.

What was your rationale for keeping it like that? It seems to me that we should have the option of deciding if the effect size measure should be applied to word frequency or (text) dispersion scores.

Thanks again for the great work!

Best,

Elvis

[Laurence Anthony]

Hi Elvis,

Sorry for the super slow response. This is a really good question. I implemented all the effect size measure as they are defined in the literature. This is so that people get the value that they would expect. But, as you say, if you are using text dispersion as a keyness measure, it would be useful, interesting, novel to generate effect sizes using range over frequency. I would strongly recommend you try this manually (e.g. by exporting the results to Excel and doing the calculation there). Depending on the results, you might have a research paper there worthy of publication. I think it would be a very new idea.

I hope that helps!

Laurence.