Bayesian statistics for Spearman's

LucieR

New Member
Does anyone know if you can compute Bayesian statistics for Spearman's correlations on SPSS?

hlsmith

Less is more. Stay pure. Stay poor.
Not off hand, but that doesnt mean much. What does your data look like? Were you hoping to use informative priors?

There are lots of Bayesian nonparametrics (e.g., Bayesian quantile regression, etc.).

Karabiner

TS Contributor
Spearman's rho is the same as Pearson's r applied to the rank-transformed variables.

With kind regards

Karabiner

spunky

Can't make spagetti
Does anyone know if you can compute Bayesian statistics for Spearman's correlations on SPSS?
What's wrong with the regular calculation? (SPSS can't do it, btw. You'd need to use Stan or R or a more specialized programming software)

LucieR

New Member
Spearman's rho is the same as Pearson's r applied to the rank-transformed variables.

With kind regards

Karabiner
So will the Pearson's r implementation produce the same Bayes factors to a (hypothetical) Spearman's implementation?

Many thanks

LucieR

New Member
Not off hand, but that doesnt mean much. What does your data look like? Were you hoping to use informative priors?

There are lots of Bayesian nonparametrics (e.g., Bayesian quantile regression, etc.).
I have computed several Spearman's correlations already as my data is not normally distributed. I was hoping to use bayesian statistics to demonstrate the lack of relationship between my variables.

spunky

Can't make spagetti
I have computed several Spearman's correlations already as my data is not normally distributed. I was hoping to use bayesian statistics to demonstrate the lack of relationship between my variables.
Couple of things to keep in mind: The Spearman correlation captures a very specific type of association, monotonic. So a small rank correlation does not imply a lack of relationship. If you have two variables X and Y and define Y = X^2 you can see that the rank correlation is very small (it's 0 in the population) but the variables are strongly related.

And second...yeah. Not something available in the SPSS Bayesian module at the moment. But you can do it in R.

LucieR

New Member
Couple of things to keep in mind: The Spearman correlation captures a very specific type of association, monotonic. So a small rank correlation does not imply a lack of relationship. If you have two variables X and Y and define Y = X^2 you can see that the rank correlation is very small (it's 0 in the population) but the variables are strongly related.

And second...yeah. Not something available in the SPSS Bayesian module at the moment. But you can do it in R.
This is really helpful, thank you!

Karabiner

TS Contributor
I have computed several Spearman's correlations already as my data is not normally distributed. I was hoping to use bayesian statistics to demonstrate the lack of relationship between my variables.
Does the computation of a Bayesian uncertainty interval require normally distributed variables?
I am not sure.

Is your sample size very large? How did you assess degree of nonnormality?