Asymmetric uncertainties in input data

[lneumann95]

Is it possible to modify the routine in order to deal with asymmetric uncertainties (xsig=(xsig-,xsig+), ysig=(ysig-,ysig+)), or even better accounting for the resulting uncertainty distribution when going from linear to logarithmic scale (which is not Gaussian anymore)?
The problem is, if you have (symmetric) uncertainties in the (initially linear-scale) data and want to apply LinMix to perform a linear regression of the log-log-scale converted data, you end up with asymmetric uncertainties and the fit will be biased towards larger values in the Gibbs sampler (for the drawn data) and hence biased towards larger values in the intercept and eventually a flatter slope (which I have tested with some data simulations).
This should be a common issue in astronomy because we often go from linear-linear to log-log scale and then perform a linear regression to effectively fit a power law to the data.

[gsrinivasaragavan]

I also am running into this issue currently and was wondering if there is any update.

[paigemkelly]

I am still stuck on this. Has anyone found a solution?

[alessandropeca]

Same here.
I think people usually take the average of the upper and lower error bars. This is not correct, but it is not expected to produce large differences if the asymmetry is not too large.