3

A program I have used for a linguistics analysis offers the option of using log likelihood or chi-squared to calculate the keyness of a word (Keyness being the actual frequency of a word in a text in comparison to the expected frequency based upon its frequency in a reference text) The program has a default setting of using log likelihood, so I happily left it at that and let it compute the statistics I needed. I do however, need to briefly explain the difference between the two methods, and possibly why LL is suggested as the better option.

I haven't studied maths in a long time. Can someone explain the difference between chi-squared and log likelihood, in simpleton terms for me?

Eloisa
  • 31
  • "Chi-squared" probably means minimum-chi-squared estimation and "log-likelihood" probably means maximum-likelihood estimation. And maybe this question would get better answers at stats.stackexchange.com . $\qquad$ – Michael Hardy Apr 26 '18 at 00:37
  • Your brief description does not say much about the data. The distinction here may be between the Pearson chi-squared test using statistic $Q = \sum \frac{(X_i -E_i)^2}{E_i},$ where $X_i$ are category counts and $E_i$ are corresponding expected values under the null hypothesis. $Q$ is approximately CHISQ$ with degrees of freedom depending on linear constraints. The likelihood ratio test statistic also has a chi-sq distribution, has a messier computational form, and is thought to be more accurate. // Maybe see Wikipedia for a start. // Here or elsewhere, pls give more detail. – BruceET Apr 26 '18 at 05:50

0 Answers0