I'm working with survey data and I have asked 7 questions with predefined responses. As an example, "How do you feel about the meal?" Answers: Satisfied, Ok, Not satisfied. I want to measure the strength of association between two variables. As an example: Strength of association between "Meal satisfaction and Overall satisfaction". My data would be as follow.

enter image description here

I know we can use "Chi - squared test" to check whether there is a relationship between the two variables or not. Is there any way that I can measure the strength of the relationship? (Like the association between the variables is positive/negative/no relationship). Thanks in advance.

The data I have is not dichotomous because there are 3 responses for the variables. Hence I think I cannot use either Cramer's V or Polychoric correlation.

  • 1
    $\begingroup$ Possible duplicate of Correlations with categorical variables. Also check this question: stats.stackexchange.com/questions/28094/… $\endgroup$ – T.E.G. - Reinstate Monica Jan 31 '17 at 5:44
  • $\begingroup$ I checked above questions. But the answers are given for 2x2 contingency tables. But I have 3x3 and 3x2 or 2x3 contingency tables. Is there any other method that I can use? $\endgroup$ – user147313 Jan 31 '17 at 6:00
  • 1
    $\begingroup$ Correlation only between a pair of categories can have sign (such as between Satisfied and Satisfied; Satisfied and Ok; etc.). See stats.stackexchange.com/q/74708/3277. $\endgroup$ – ttnphns Jan 31 '17 at 6:52
  • $\begingroup$ @ttnphns: Thank you :). So then do I have to calculate correlation between the variables as "compared to Ok, Satisfaction has a positive relationship, compared to Ok, Dissatisfaction has a ..., etc."? Like we do in multi-nomial logistic regression? $\endgroup$ – user147313 Jan 31 '17 at 21:55