2

Background*strong text* Am analysing results from a questionnaire (n=130). The respondents fall into four categories (A / B / C / D), and each question/statement has 5 possible answers (Likert Scale so Disagree/Agree etc).

I want to see if there are any significant differences or similarities between the response and the group that they belong too.

If I use the chi-squared test, approx 75% cells have a expected count LESS than 5 - even when I pool the responses from 5 to 3 by merging Strongly Agree and Agree, then it is still 40% of cells have an expected count less than 5 (the threshold is 20%). Conclusion = can't use chi-squared for stat analysis

Question

Is an acceptable alternative the Fishers Exact Test? http://en.wikipedia.org/wiki/Fisher's_exact_test

And because the the contingency tables is 4x5 (my understanding is that Fishers Exact test should be used for 2x2), is it okay to use the Monte Carlo method: http://en.wikipedia.org/wiki/Monte_Carlo_method

Any other suggestions welcome. Thanks! (PS I'm a biologist so my stats know-how is a little limited)

Harriet
  • 21

1 Answers1

1

Yes, Fisher's exact test is an acceptable alternative. Note the test will simply report whether the entire table deviates from independence, and not tell you which entries do. If you use a statistical package like R, I believe it will do exact counting of tables larger than $2 \times 2$ for small entries, otherwise it will do Monte Carlo.

user2566092
  • 26,142