Let's start with the preface that I am a linguist, not a mathematican. My knowledge of stats stopped after A-level oh-so-many years ago. So if answers could be written in simpleton terms, that would be much appreciated!

For a linguistics study, I have used a corpus program to calculate keyness. Keyness being the actual frequency of a word within my corpus compared to the expected frequency, based on its frequency in a larger reference corpus. A keyness of 3.8 is significant to p<0.05 and a keyness of 6.6 is significant to p<0.01. The program has two options for calculating keyness- Log Likelihood and Chi Squared. The creator of the program has set LL as the default, and has stated that he feels it is often the better method.

However, I do need to state in my project why this method has been used, even if just a sentence.

My question: What is the difference between the two tests? Why would LL be the preferred method?

  • $\begingroup$ If the manual gives more information you can add it; if not it's most likely that the former option is the likelihood-ratio goodness-of-fit test & the latter Pearson's chi-square goodness-of-fit test. $\endgroup$ – Scortchi - Reinstate Monica Apr 26 '18 at 10:34