I am conducting Spearman's correlation on two skewed continuous variables. I do not wish to log transform them in order to perform Pearson's correlation.

I want to display a scatter plot of the two variables and a line of fit as this

Code:
twoway scatter x y || lfit x y
However, I'm worried that the line of fit drawn on the scatter graph by the -lfit- command will be a false representation since the lfit assumes that the distributional assumption of normality is met??

Does anyone know how I can represent a non-parametric line of fit without log transforming the skewed variables?

Thanks,

Madu