Comment by ksbakan

If the methodology is good, the chance of a fluke is pretty low, given the 2% difference and large sample size.

It is much more likely that the result would be thrown off by a sampling error.