4 ms·

Not enough packages to make a real claim. The variance on package delivery times in the USPS is high. As an example, I recently had two identical packages shi

Not enough packages to make a real claim. The variance on package delivery times in the USPS is high. As an example, I recently had two identical packages shipped to me. They both left the same facility at the same time. They were identical weight(heavy) and dimensions. They were wrapped the same. They were shipped from in the same state as me. One arrived in 2 days. One arrived in 10. Also, they mentioned that one of the packages in Michigan took 37 extra days. With that sample size it explains about 10 percent of the average right there.

Jabbles 14y ago
I agree that a larger sample would be welcomed. In particular, spreading the experiment out over many days would help eliminate a "oh all these packages are the same, they probably go here" mistake. Whilst sending them all out on the same day eliminates some sources of bias, it may have made the experiment more susceptible to other errors.
Retric 14y ago
Having high variance requires a larger difference to pass a statistical test.
pflats 14y ago
n = 85 and p < 0.01. What's inherently wrong with that? (I have my own personal crusade against the 5% p-value, but this is under 1%.) I agree that the sample size is far too small to criticize individual states, but 85 seems plenty large for the hypothesis of "Atheist-branded packages sent by the USPS are being delayed more than normal packages."
- yummyfajitas 14y ago
  How did you compute a p value without knowing the variance? [Edit: my bad, didn't read the fine print which explains statistical methods at very bottom of page.]
  - maxerickson 14y ago
    In this case, I imagine the method involves reading it from one of the images towards the bottom of the linked page.
- AtheistBerlin 14y ago
  In fact n= 178, which is more than enough for a study with only 2 conditions.
  - pflats 14y ago
    I misread your post when I re-skimmed it, and ignored the 2 shoes thing. Whoops. Thanks!
AtheistBerlin 14y ago
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.
- dthunt 14y ago
  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)?
- ethomson 14y ago
  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".

Jabbles 14y ago

I agree that a larger sample would be welcomed. In particular, spreading the experiment out over many days would help eliminate a "oh all these packages are the same, they probably go here" mistake. Whilst sending them all out on the same day eliminates some sources of bias, it may have made the experiment more susceptible to other errors.

Retric 14y ago

Having high variance requires a larger difference to pass a statistical test.

pflats 14y ago

n = 85 and p < 0.01. What's inherently wrong with that? (I have my own personal crusade against the 5% p-value, but this is under 1%.) I agree that the sample size is far too small to criticize individual states, but 85 seems plenty large for the hypothesis of "Atheist-branded packages sent by the USPS are being delayed more than normal packages."

yummyfajitas 14y ago
How did you compute a p value without knowing the variance? [Edit: my bad, didn't read the fine print which explains statistical methods at very bottom of page.]
- maxerickson 14y ago
  In this case, I imagine the method involves reading it from one of the images towards the bottom of the linked page.
AtheistBerlin 14y ago
In fact n= 178, which is more than enough for a study with only 2 conditions.
- pflats 14y ago
  I misread your post when I re-skimmed it, and ignored the 2 shoes thing. Whoops. Thanks!

AtheistBerlin 14y ago

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.

dthunt 14y ago
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)?
ethomson 14y ago
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".