0

I know theres a similar answered question to mine on this forum but I would like to ask a new question for my specific case and would also like to ask further questions for some of the answers that were given there as well.

I have just received the results from a quesionnaire I did where some of the questions were in likert scale, 1 (Strongly agree) to 5 (Strongly disagree). When I tried doing a chi-square test for independence for two question responses, I got the below results and warning message about expected counts being less than 5 for some cells. I've asked in another place about this and one person said to combine columns or eliminate respondents but I really don't want to remove or alter my data. Is there a way of fixing this problem without changing data too much that it'll affect the results?

Also in the other post that I earlier mentioned, someone commented that the chi-square test is "poorly disposed to handling varying numbers of categories within a single test". Is that the case with me? If, so what test should I use?

Further background: For my research, one of the thing I want to do is to compare the relationship between different variables such as recycling attitude and recycling behavior.

enter image description here

David Jr Gbadago
  • 75
  • 8
  • 1
    Since attitudes to recycling is an ordered categorical variable why not use ordered logistic regression? – mdewey May 07 '17 at 14:37
  • 1
    The chi square approximation does not work well when you have such sparse cells. You can use a form of Fisher's exact test. – Michael R. Chernick May 07 '17 at 15:36
  • To mdewey: I'm not familiar with regression but I'll give it a shot. Also, I'm still not sure whether to set my data as nominal or ordinal in my statistical software which is why I posted a question about this: https://stats.stackexchange.com/questions/277931/will-there-be-a-difference-in-my-results-if-i-set-my-data-as-nominal-rather-ordi?noredirect=1#comment533147_277931

    But I think the answer is that ordinal is better since I have more options. So logistic regression a better tool to use in your opinion?

     – David Jr Gbadago May 07 '17 at 15:44
  • To Michael: I'm not familiar with Fisher's tests. What specific form is this? – David Jr Gbadago May 07 '17 at 15:47
  • https://en.wikipedia.org/wiki/Fisher%27s_exact_test – whuber May 07 '17 at 20:53

0 Answers0