You're correct; but getting hourly compensation data in panel data (especially with consistent methodology) is difficult. I decided wages would have to act as a proxy, even if imperfect