I have done a questionnaire composed of multiple responses for each question. I have two categories which I am comparing (dentists and dietitians) so the tables I got after data entry were 2Xn(the number of responses to a certain question). I then carried out the Chi-squared test for every question and I got a significant result but I couldn't see which responses were the most significant. Another problem I faced was that some frequencies in the table were below 5 so I added low frequencies for more than one response to get a total of 5. I have done the same for both categories in the same column. Can anyone address these problems I am having with my Chi-squared test?
The version of Chi-square I used was from a basic MSDos program. I couldn't do it on SPSS because I haven't entered the data originally there.
Occupation Never Rarely Sometimes Daily
Dietician 1(1%) 1(1%) 15(16%) 76(81%)
Dentist 18(15%) 11(9%) 23(19%) 70(57%)
this is one example of the questions. the responses to this question are single but please not the low frequencies!
Rare events are never informative. I'd suggest to simply dichotomize the data to "daily" vs. "less than daily". Then you can do a binomial test and calculate an odds ratio.
Thanks very much for the reply. It's the same with my other questions, if here I can categorise them in two groups; for the other questions I can't because every single response indicates something for me.
What do you mean by binomial test?
Samiha, you may consider the following:
First you determine the computed chi-square value and solve for the chi square based measures of association such as phi coefficient, contingency coefficient or Cramer's V. These measures of strength of association will tell which pairs have the greatest magnitude of relationship.(SPSS will automatically anwwer some of these)
Second, you "chi-square test" the pairs ( if you use SPSS, it will automatically tell you whether the pair is significant at 0.05, 0.10, 0.001), this will tell you which pairs "are more significant" on the basis of the outputted level of significance.
Jochen, this could be a misunderstanding. I mean alphas. Some books use highly significant for 0.01, and significant for 0.05.
thanks all, i didn't use SPSS because my data are not in the format for SPSS.
however, i ranked the answers from highest frequency to lowest. do you have any ideas which rank test i could use?
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics