Hi all,
I am running a partial correlation analysis on mood data. I want to see if there is a correlation between scores on a novel hypomanic mood scale and existing measures of hypomania - the Mood Disorder Questionnaire and Hypomanic Personality Scale-6, respectively, while controlling for BDI (depression inventory) and STAI-T (trait anxiety measure) scores. If it helps, we have data from 182 participants with no missing data points.
I have done a preliminary analysis to test for parametric assumptions - the data is non-normally distributed. It does not meet parametric assumptions. And from scatterplots, it is also non-monotonic.
Is it still ok to run a spearman's correlation in this case? I have found contradictory evidence online e.g. the advice on the SPSS website is that an assumption of Spearman's correlation is that the relationships between our variables are monotonic. They are not at all, even after ranking the data it still looks random/no relationship.
Any help would be much appreciated! Thanks!
Yes, there is nothing wrong with using Spearman correlation coefficient.
Of course, I think it would be helpful to identify and possibly delete outlier data before starting a correlation analysis.
Hello Tara,
Yes, you may certainly compute and report Spearman correlations; the interpretation is still valid--the degree to which ranked scores are linear in form across the set of paired cases.
However, you might consider polymomial regression instead, especially if your data set suggests obvious non-monotonic relationships. Doing so would allow you to characterize the way(s) in which non-linear relationships are operating, at least for your data set.
Good luck with your work.
You can still use the correlation coefficient, but if the data has some obvious pattern that is not linear, then you should try to fit whatever pattern you observed in the data (polynomial, power, exponential, logarithmic, etc.) instead of using a linear regression model. If there is no correlation at all, then it would show in the R^2 or R values you calculated for any of the patterns that you try. If it makes a significant difference between different patterns, then you can pick the one that has the highest correlation.
- Badges
- Science topic
- Similar topics
- Biological Science
- Kinesiology
- Running