Power analysis - Partial Spearmen correlation

#1
Dear experts,

I conducted several partial Spearman correlations to test association between two variables while controlling for several variables using a script because SPSS does not provide this option directly.

However, I would also need to report the power analysis for the non-parametric partial Spearman correlations but SPSS does not provide this option i.e. by default the power analysis for partial correlation is for Pearson partial correlation.

Does anyone by any chance know how to do power analysis for nonparametric partial Spearman correlation? Or perhaps if a script could be provided this could be run on SPSS.

Thank you in advance
Jose
 

spunky

Can't make spagetti
#2
Dear experts,

I conducted several partial Spearman correlations to test association between two variables while controlling for several variables using a script because SPSS does not provide this option directly.

However, I would also need to report the power analysis for the non-parametric partial Spearman correlations but SPSS does not provide this option i.e. by default the power analysis for partial correlation is for Pearson partial correlation.

Does anyone by any chance know how to do power analysis for nonparametric partial Spearman correlation? Or perhaps if a script could be provided this could be run on SPSS.

Thank you in advance
Jose
The sampling distribution for the Spearman correlation under the alternative hypothesis is quite complex, and not usually available except for very limited, simple and usually unhelpful cases (e.g., ranking variates from a normal distribution).

Your best bet is to approximate power through a Monte Carlo simulation, but you'd need to code that.
 
#3
The sampling distribution for the Spearman correlation under the alternative hypothesis is quite complex, and not usually available except for very limited, simple and usually unhelpful cases (e.g., ranking variates from a normal distribution).

Your best bet is to approximate power through a Monte Carlo simulation, but you'd need to code that.
 
#4
Thank you very much for the reply much appreciated.

Unfortunately I am still quite new coding is there a possiblity to provide code for this? Or suggestion on how to code for this? Thank you again
 
Last edited:

spunky

Can't make spagetti
#5
Thank you very much for the reply much appreciated.

Unfortunately I am still quite new coding is there a possiblity to provide code for this? Or suggestion on how to code for this? Thank you again
Are you familiar with R? Or does it have to be in SPSS?

Also, for which test of the correlation would you like to calculate power for?
 
#6
Unfortunately I have never used R just SPSS - perhaps a script could be made to calculate power based on the n, number of covarites and p-value

But I can provide the information regarding the partial Spearman correlations. In total, I want to calculate power for 4 partial Spearman correlations (below)


Correlation 1 (r = 0.76, p = 0.004, n = 17, covariates used = 5)

Correlation 2 (r = 0.7, p = 0.01, n = 17, covariates used = 5))

Correlation 3 (r = -0.62, p = 0.03, n = 17, covariates used = 5)).

Correlation 4 (r = 0.82, p = 0.001, n = 17, covariates used = 5)).

Does this help?
 

hlsmith

Less is more. Stay pure. Stay poor.
#7
It is a faux pas to calculate power after you have data. It is unreliable, since you have already seen the results.
 

spunky

Can't make spagetti
#8
Unfortunately I have never used R just SPSS - perhaps a script could be made to calculate power based on the n, number of covarites and p-value

But I can provide the information regarding the partial Spearman correlations. In total, I want to calculate power for 4 partial Spearman correlations (below)


Correlation 1 (r = 0.76, p = 0.004, n = 17, covariates used = 5)

Correlation 2 (r = 0.7, p = 0.01, n = 17, covariates used = 5))

Correlation 3 (r = -0.62, p = 0.03, n = 17, covariates used = 5)).

Correlation 4 (r = 0.82, p = 0.001, n = 17, covariates used = 5)).

Does this help?
No, it doesn't help. You can't do what you want in SPSS. You'll need a more general programming language like R or SAS or Stata to be able to do what you want to do.

Also, what @hlsmith said is correct. Doing a power analysis after the fact (e.g., "post-hoc power analysis" or "retrospective power analysis" among others) is neither useful nor helpful. Reference to get you started:

https://rss.onlinelibrary.wiley.com/doi/abs/10.1111/1467-9884.00139
 
#9
Hello all thank you for the sugestion. Actually it was the reviewers who asked me this power analysis but I will use this reference to first read and they report why it was opted not to include this analysis. Have a good day