I know that some researchers, when taking the log of a variable with some nonpositive values (like wealth), simply replace the nonpositive values with a small positive value. I also know that this approach has been strongly criticized, but I've been unable to turn up any refereed papers that discuss the dangers of the replacement method. I'd be grateful for any suggestions. I need to make the case to an editor that this is a bad idea and would like to have the citations to back it up.