I have carried out the following two spearman partial correlations using the "ppcor" package in R:

-- Coefficient 1 = Correlation of A1 with B controlling for C;

-- Coefficient 2 = Correlation of A2 with B controlling for C.

I have a set 234 words. For each word I have different measures (A1 and A2 are two versions of the same measure, and B and C are in common as you can see. I would like to compare the two Spearman partial Rhos but I am not sure how to proceed.

My idea would be to use bootstrapping to generate a distribution of absolute difference between the two partial correlation coefficients, to then extract the 95% confidence interval and see if it includes 0.

Do you have any suggestions?

here is the distribution of B outcome:

enter image description here

here is the scatterplot of the relationship between A1/A2 and B:

enter image description here

And here is the scatterplot of the relationship between A1/A2 and B with A1/A2 transformed for graphical purposes (Note, this transformation does not affect Spearman or Kendall tests as they are based on ranks):

enter image description here

  • $\begingroup$ Perhaps regression and compare betas, this will allow you to examine the magnitude of the association and control for C. $\endgroup$ – coconn41 Jan 6 at 21:10
  • $\begingroup$ thanks coconn41. After taking the log10 of the A1/A2 (for graphical purposes), the relationship between A1/A2 and B seems something like y=1/x. More importantly, the error around this potential model is much larger for smaller values of A1/A2. Also, there are outliers in the data which it does not make sense to exclude for now. Do you think regression can still be applied? I used Spearman (and Kendall gives same results) because it makes no assumptions about the distribution of the data. $\endgroup$ – Francesco Cabiddu Jan 7 at 15:18
  • $\begingroup$ Can you post a graph / histogram of the distribution of your dependent variable? And also a scatter plot of your A1 against B and A2 against B? $\endgroup$ – coconn41 Jan 7 at 16:31
  • $\begingroup$ Or if you know how they are distributed you can just state it. $\endgroup$ – coconn41 Jan 7 at 16:38
  • $\begingroup$ Sure thanks, I have edited my original question adding the plots you asked. $\endgroup$ – Francesco Cabiddu Jan 7 at 18:26

Your Answer

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

Browse other questions tagged or ask your own question.