I have conducted a survey and collected 182 responses from 6 different groups and I am about to analyze my data:

  • group1: 101 responses
  • group2: 47 responses
  • group3: 15 responses
  • group4: 13 responses
  • group5: 6 responses

Some of in some of questions I used likert scale, for example:

A How much is quality of life important to you?

1 not important at all...2.......3.............4........................5 very important

and three kind of personal questions:

B 1. where did you live prior to relocate to NYC?

1. SF  2. Austin 3. other U.S. areas  4. non U.S. area 

C 2. please indicate your age.

1. below 40  2. between 41 and 50   3. over 50

D 3. were you born in NYC?

1. Yes  2. No

I am trying to:

  1. Do a correlation between each group of responses and each single of likert scale questions.

  2. Do a correlation between likert scale questions( I think in this section I have to use spearman)

  3. Do a correlation between C questions and likert scale ones and D and likert scale ones.

    Also between D and C and B.


Don't do correlations at all. Do ordinal logistic regressions instead and, if the assumption of proportional odds is really violated, try multinomial logistic.

Surely "Quality of life" is the dependent variable and the others are independent.

Correlations (of whatever sort) do not treat one variable as the dependent and others as independent - they treat all variables equally. Also, correlations look at the relationship between two variables. Regression lets you control for multiple variables.

Ordinal logistic regression is good when the DV is ordinal, but the usual method assumes proportional odds. Although there are models that do not make that assumption - or make it partially - they are relatively little known, so multinomial might be better


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.