In my profession, we call that a selection bias.
Now I'm going to show my geek side...
If the population randomly took pregnancy tests - men and women, on bcp and not, ttc and not, any day of the month/cycle - then you probably won't get that trend.
Most tests are probably taken by women who are ttc. (women who are unexpectedly late probably take fewer tests.) To avoid a false negative, these women take the test close to when af is due. Since the odds of getting pregnant are 25% or less, that means that 75% or more of the tests will be negative right before af shows up.