Looking for a technical answer.
How bad is the bias introduced by just making ln(0)=0


\(\infty\)

Infinity

I've often wondered the same. For real
credit risk bro

Oh my.
\(\lim_{x\rightarrow 0}\ln(x)=\infty \)
That's just a wee bit far from \(0\).

ln(0+) > inf
You should instead replace 0s with a small number (like 0.005) before taking logs. 
ln(0+) > inf
You should instead replace 0s with a small number (like 0.005) before taking logs.Hack.

Nice try, Emil. GTFO.

ask Miles Mathis

ln(0+) > inf
You should instead replace 0s with a small number (like 0.005) before taking logs.Hack.
cubic root is an alternative when you have observations with values 0. but good luck getting it passed reviewers. I had reviwers complain about iut and saying I should replace 0s with small number, which of course is a hack method because there is no technical justification for doing so. Nick Cox has a number of posts on this on stata list.

Better answer: transfer all your data by +1 before logs.

Better answer: transfer all your data by +1 before logs.
hack method

poisson regression

This. Look up the ReStat paper called "The Log of Gravity."
poisson regression

Inverse hyperbolic sine good

pretty sure ln(0) = 1
what am i missing

It's undefined. You are thinking ln(1)=0.
pretty sure ln(0) = 1
what am i missing 
pretty sure ln(0) = 1
what am i missingyou're missing that ln(0) does not equal 1, but that ln(1)=0. Are you in high school?

efd9 has it. Poisson regression is the answer.

You guys are implying that the bias is infinite, but when I remove observations where I have made ln(0)=0 the results don't change by anything worth mentioning.