Correlation between nominal variables

#1
Hi all,

I'm currently working on a dataset extracted from a questionnaire and entered and coded in SPSS. Im trying to show correlation/relationships between a range of demographic dependant variables and one outcome variable. The data is nominal and often on a scale (ie.how important would you rate the following in your decision to do X or at what level are you working etc). There seems to be a lot of conflicting advice on how to go about this, even in the simplest way possible on SPSS so I would really appreciate any help anyone could give me! I have descriptive statistics done but I'm struggling to move past there. I realise this is pretty simple stuff but i'm still quite confused. Thanks!
 

noetsi

No cake for spunky
#2
You aren't going to find consensus on this or any other point in statistics, you will save a lot of pain if you learn to accept that :p (I still am bothered by it after years of doing this).

If the variables are entirely nominal Phi and Cramer's V is often used. This is a link discussing them. I am sure SPSS generates them although I have not worked with it recently enough to know where.

http://www.harding.edu/sbreezeel/460 files/statbook/chapter15.pdf

Some of the variables you mentioned above seem ordinal/interval as importance or level. Other methods such as polychoric correlations or point biserial would apply in that case depending on the exact nature of your data.
 

Karabiner

TS Contributor
#3
Im trying to show correlation/relationships between a range of demographic dependant variables and one outcome variable.
So if simply you want to show relationships (at least, you
did not explicitly describe a need for calculating
correlation coefficients here), you can just perform
some common tests. According to the thread title,
I suppose that your outcome variable is categorical with
> 2 categories; so you could use it as grouping factor
and perform a Kruskal-Wallis test (for ordinal demographic
variables), Chi² (for categorical demographic variables),
or Kruskal-Wallis or oneway-ANOVA for interval scaled
demographic variables.

With kind regards

K.