I'm not suggesting you do (although I know others on this thread are suggesting that).
Administer sobriety tests the same way you currently administer drug tests -- upon application, and randomly. They're a lot cheaper so you can actually get more data for less $/time.
Administer sobriety tests the same way you currently administer drug tests -- upon application, and randomly. They're a lot cheaper so you can actually get more data for less $/time.