That's fair enough, but the numbers you give are neither the absolute nor the relative consequences. You can't draw any conclusions from simply comparing the m

That's fair enough, but the numbers you give are neither the absolute nor the relative consequences. You can't draw any conclusions from simply comparing the medians. On one hand that completely ignores any confounding variables which could lead to either over- or underestimation of the true difference. And on the other hand the medians may not be particularly meaningful. For example, if the effect only occurred in the top 25%ile of employees, the median would stay the same even if all of the ones in open plan offices were sick for the entire year. If you took the mean instead of the median the difference would already be 8.1 - 4.9 = 3.2 days.