[AtheistBerlin]

You're welcome to see the data if you like - we ran it by two independent statisticians to make sure our sample size was large enough... n = 178, in a 2 condition study, really isn't too bad. The important point here is that the variance should have been equally spread across both conditions... and it wasn't.

[ethomson]

My understanding is that you ran this test once. You sent all the packages - labeled and not - on the same day. And you did this one time. This obviously doesn't control for things like "half the packages just fell off the back of the truck" or "the packages at the bottom of the bin got delayed for extra screening".

[dthunt]

So the real question is, where to go from here with a reasonable amount of certainty?

Does USPS have an office that conducts quality control experiments?

Will you continue collecting data on this front at some lower frequency (with consenting customers)?