It would be helpful to have more information on the data collected.

1) Do I understand it correctly that you have a survey where each participant answered a number of questions and you want to see if the responses to two of the questions correlate to each other? And that not everyone answered both questions? If so, those that did not answer both questions do not contribute any valuable data and can be excluded (but you should be transparent with this and potential biases this may cause).

2. If the ordinal variable has only two possible values (i.e. dichotomous; e.g. "good" vs. "bad"), Fisher's test works well. However, if there are multiple possible values (e.g. "very bad", "bad", "good", "very good") it is not applicable.

Jon Edman-Wallér Ajit Singh thank you both for your responses.

In relation to 1) I would like to look at the relationship between an "age group" variable with 4 options against another ordinal variable, likewise another variable named "year of qualification" which was the optional question. I have spoken with another colleague who has suggested looking at using Chi-square for this question as opposed to Spearmans rank. What would your thoughts be on this? In terms of the outliers, yes I think I will remove all cases where respondents did not answer both questions and like you say, be transparent with this in my discussion.

Jon Edman-Wallér - in relation to 2), there are multiple possible values for the ordinal variable - would you suggest using Chi-square as opposed to Fisher's exact?

Any further advice would be greatly appreciated.

Hi Ajit Singh

Thank you again for your response.

Apologies, I'm not sure I have made my intentions clear regarding Spearmans rank.

I want to examine the relationship between "age group" and variable X (ordinal). Then separately examine the relationship between "year of qualification" and variable X.

Fisher's exact is no longer appropriate for 2) above as variables are not binary, but not all cells have a count of over 5, therefore I am unsure which test to use in this instance.

I do hope this makes sense, please let me know if I can clarify anything further.

Thank you for providing more details about your question. To clarify, you want to examine the relationship between variable X (ordinal) and two different categorical variables: age group and year of qualification.

Since Fisher's exact test is not appropriate for analyzing the relationship between variable X and year of qualification, you may want to consider using a non-parametric correlation test such as Spearman's rank correlation coefficient. Spearman's is appropriate for analyzing the relationship between two ordinal variables or an ordinal variable and an interval or ratio variable.

To examine the relationship between variable X and age group, you could use a chi-square test or a similar test for independence. This test would allow you to determine whether there is a significant association between the two categorical variables.

When conducting hypothesis tests, it's important to ensure that you meet the assumptions of the test and that the test is appropriate for your research question and data. I hope this helps! Please let me know if you have any further questions or if there's anything else I can assist you with.

HI Ajit Singh

Thank you again for clarifying.

I am looking to use Spearmans rank for age group and variable X instead of chi-square provided this is appropriate.

However, for 2) in my original question (comparing one ordinal and one nominal variable) I would use chi-squared although not all cells have a count of >5. Fishers exact is not applicable though as the variables are not binary - would you have any other suggestion for this situation?

@Stacey Provan

Spearman for case 1 and Chi2 for case 2 stunds good. However, the low counts is problematic for case 2. See if you could deal with this by combining groups for the ordinal variable. If nothing else works, you can dichotomize it (i.e. sort the population into just two groups above or below a certain value). Then you can use Fisher's test.