P < 0.05 means that the probability of seeing the observed results if there is no difference is less than 1 in 20. So, if you run the experiment 20 times, you'd expect to see a link once, even if there is really no link.