1

Please may someone help me understand how to interpret a positive correlation i have found between a variable and the demographic of ethnicity.

Basically i have found ethnicity to be positively correlated with four of my outcome variables subscales. Below are the numbers. im struggling with how to make sense of these, what ethnicity are the subscales correlated with as our questionnaire includes 10 options for ethnicity?

Integration Segregation (r = .179, p = .012), Private Rights (r = .302, p = .000), Social Distance (r = .159, p = .026) and Subtle Derogatory Beliefs (r = .222, p = .002)

EDIT: So basically we computed ethnicity as 0-asian bangladeshi, 1-asian other, 2-black african etc

LCFox
  • 11
  • 2
  • 1
    Since correlation is a mathematical property of sets of ordered pairs of *numbers,* how did you compute correlations for a ten-category variable like ethnicity, which neither is numeric nor can be fully represented by any single number? – whuber Mar 31 '16 at 20:45
  • @whuber: I'd imagine asian=1, black=2, etc. Answering how is much easier than answering why. – Jeremy Miles Mar 31 '16 at 22:33

2 Answers2

2

You can't make sense of those, because they don't make any sense. Correlation is appropriate for numerical data, ethnicity is categorical data.

Jeremy Miles
  • 13,917
  • 6
  • 30
  • 64
  • The idea here is that the categorical data isn't inherently ordered. If you measured height, weight, or something else numerical, you can objectively say that a measurement of 100 is more than a measurement of 90. In your encoding, a measurement of 2 (black) isn't "more" or "less" than a measurement of 1 (asian). You'd be perfectly justified rearranging your arbitrarily assigned category labels, which would give you totally different results for the analysis you (incorrectly) conducted. – Nuclear Hoagie Apr 01 '16 at 10:03
0

@LCFox, check out this answer from a related thread.

As others have noted, correlation is a measure of the strength of the relationship between two numeric variables. More specifically, Pearson's correlation, the one we usually think of when we see correlation, measures the linear dependence between two real valued random variables.

Community
  • 1
Chris K
  • 108
  • 4