These are aggregate metrics. As such, individuals may have seen their salaries stay the same or even go up, but if the employees at the top of the pay scale are removed (layed off, or even a simple selection bias in the dataset), the median will still drop.