4.8 to 7.1% is quite a bit more significant than your initial claim of 2 to 3%
edit - also, how do you claim to have any kind of confidence in the accuracy of your own estimate, while simultaneously arguing that reaching any sort of accurate value is impossible?
It was someone else who made that initial claim of 2%, however if you read the full summary of my second link, they admit to only accounting for factors they were able to measure. So additional factors they didn't account for, may have influence towards the divide (other than gender-pay), and may account for any number that could further reduce that 4.8-7.1%. I'm not saying it is as low as 2-3%, but its certainly possible.
It was someone else who made that initial claim of 2%
Sorry, complete failure to pay attention on my part.
Though as to your other point, unmeasured factors could push the figure either way, so if you allow the error bars to go as low as 2% purely on a hunch, then you should also consider that it could be up to 9.9%, if trying to work out any sort of reasonable policy.
edit - also, how do you claim to have any kind of confidence in the accuracy of your own estimate, while simultaneously arguing that reaching any sort of accurate value is impossible?