No, the global hashrate is extrapolated from the rate that blocks are found. A lucky streak will make it look like the hashrate is higher than it actually is.
Beyond that, the sample period is too small immediately following a difficulty change so the numbers are always way off for a few days.
I agree with you first point - it is true that it's block discovery rate, not hashrate, that is directly measured, and that it's possible we're just seeing a lucky period. However, your second point doesn't make any sense: what do you mean by the sample period? The block discovery rate is calculated by adding up N blocks over the past T minutes/hours, and then calculating N/T. The calculation of block discovery rate should be completely independent of both the difficulty, and where we are within a given 2016 block period. So, why should we inevitably see more variation in hash rate at the beginning of a 2016 block period? What you're suggesting is that T (the 'period' you mentioned) automatically resets to zero at the end of a 2016 block period. Well, I was taking the rate value from bitcoincharts.com, so if that's the way they do it, then yes I would expect a large over or under-estimate *just after* a change of difficulty. For a few days though? That makes no sense at all - think about it.
