> isn't it on the border of measurement error?

At least going by that payscale.com link, I don't think so. That is compiled from 600K+ responses, so they have enough data to measure small differences with some confidence, I think. I didn't sign up to download the full dataset though, so I'm mostly going by their claims.

Quoting from the article:

  Although $0.99 may seem very close to $1, the red line in the chart
  below has never crossed the dotted $1 line in blue representing men’s
  pay. Even when women are doing the same jobs, the gender pay gap is
  not zero.

If it were a "lost in measurement error" thing, I would expect that chart to have a lot more noise in it — some years women would be above men, other years below (that said, I do wish the charts had error bars). Instead, it's showing a small-but-consistent difference repeated across the years.