I've been a software developer of some description for a fair while, and I've authored at least my fair share of bugs. But my new worst bug I discovered yesterday. It was in code for a published economic paper.

This is the original code:

gen w01ceil = w11ceil + (1-mrw_alpha) * pr * (1-coerc_pi01) + (1-pr)*coerc_pi10

This is the correct code:

gen w01ceil = w11ceil + (1-coerc_alpha) * pr * (1-coerc_pi01) + (1-pr)*coerc_pi10

I'm sure I originally had it right and changed it at some later point for testing. I just didn't change it back. Luckily, although the correct¹ results are fairly different, they still support the point we're trying to make.

1. "Correct" as of 22nd May 2008.