I appreciated the flexibility and 1-line simplicity of lnskew0 to transform any given variable(s) into a normal distribution, however I struggled to back-transform the resulting values into their original form.
Forward transformation: transformedX = ln(S * originalX - K), where S is +1 or -1, and K is a positive or negative continuous number
Backward transformation: originalX = (exp(transformedX) + K) / S, wherein the value of K is stored and recovered from r(gamma) but the value of S is mysteriously not stored
While it would be nice to have r(sign) stored in a future version of lnskew0 (I believe originally written by [email protected]), I found - luckily - that the sign happens to be indicated in the transformed variable's label (!!) so it can be recovered as follows:
I hope this helps someone...
Forward transformation: transformedX = ln(S * originalX - K), where S is +1 or -1, and K is a positive or negative continuous number
Backward transformation: originalX = (exp(transformedX) + K) / S, wherein the value of K is stored and recovered from r(gamma) but the value of S is mysteriously not stored
While it would be nice to have r(sign) stored in a future version of lnskew0 (I believe originally written by [email protected]), I found - luckily - that the sign happens to be indicated in the transformed variable's label (!!) so it can be recovered as follows:
Code:
lnskew0 ln_value = value
scalar k = r(gamma)
local lbl : variable label ln_value
scalar sign = cond(substr("`lbl'", 4, 1)=="-",-1,+1)
gen value = (exp(ln_value) + k) / sign
