Pearson's Partial Correlation - Transforming one non-normally distributed variable

Hi all,

I would be really grateful if someone could advise me on an issue I am having. I am carrying out research looking at the relationship between the neural metrics of decision-making and accuracy on the decision-making task and I want to partial out the effect of group (control versus ADHD).

1. I know in an ideal world all 3 variables in the analysis should be continuous. Is it ok to use a binary co-variate such as group (1 = control, 2 = ADHD) in a Pearson's Partial Correlation?

2. Accuracy is not normally distributed (data are at ceiling levels) but, the neural metrics are normally distributed. I reflected the accuracy variable and log10 transformed it which renders it normally distributed. As Andy Field suggests both variables in analysis should be on the same scale I log10 transformed the neural metric variable but, as it is normally distributed the transform renders it non-normally distributed thus, unsuitable for the correlation analysis. Is it ok (meaningful) to correlate one transformed and one non-transformed variable?

Many thanks in advance.

Best wishes,


Less is more. Stay pure. Stay poor.
Re: Pearson's Partial Correlation - Transforming one non-normally distributed variabl

What is your research goal? Please, provide more information on the final purpose of your analyses. I feel inclined to ask why you aren't just using regression analyses, but I don't know what you are trying to do in particular.