Hacker News Reader

Comment by ghshephard | Hacker News Reader

I'd like to see the actual survey data. I'm not even sure how to interpret this:

"Thus the ad hoc study: The company says it sent two packages each to 89 people (178 packages total) canvassing nearly every U.S. state — one package with the Atheist-branded tape, one without. And this is where the results suggest something fishy: According to Atheist Shoes, company-branded packages took on average three days longer to reach their destination and were 10 times more likely to disappear outright."

What does "10 times more likely to disappear outright" mean? Are they saying that 10 of the 89 Atheist packages disappeared, and only 1 of the 89 non-atheist packages disappeared?

View on HN

I can't find the previous submission but someone pointed out that one of the Atheist labelled packages was lost for almost 40 days: that alone dragged the Atheist package average down so much it can almost account for the entire difference.

edit: I misremembered about the difference it made, here's the submission: https://news.ycombinator.com/item?id=5442728

Check out the comments on their own web site: http://www.atheistberlin.com/study

Not sure it will completely sate you but it's a start.

Thanks - their comments (below the infographic) are awesome - very level headed and it looks like they are going to publish a peer-reviewed paper.

(https://news.ycombinator.com/item?id=5442728)

(http://www.atheistberlin.com/study)

Agreed - the sample size is also really small to be interpreting slower arrival times as well, if only 1 or 2 packages took a long time it's likely they just found an outlier.

According to http://www.atheistberlin.com/study they got a significant Wilcoxon Signed Rank Test at p < 0.01. The Wilcoxon Signed Rank Test is nonparametric (i.e. sacrifices some statistical power in order to not make assumptions about the underlying distribution of the data) so is meaningful even if there is a long tail of packages that take longer due to circumstances outside of the study variable.

Sample size is something that must be interpreted in the presence of power. You can make a solid conclusion with a very small sample size if the true difference in arrival times is very large, given that the assumptions of the hypothesis test hold (t-test can be a little ridiculous with some of its assumptions sometimes).

In the original article, one of the footnotes mentioned that they tested the data using Wilcoxon's Signed-Rank test, which mitigates a lot of the impact of single outliers.

I'd love to see the raw data though, to see an even less-sensitive method to outliers (sign test). If the difference between the groups is as large as the article would lead us to believe, the loss of power should not present any problem.

Yeah, that 3 days could either be damning, or absolutely nothing. Until we can examine the data this is a non-story despite how juicy we may want it to be.

Yep pretty much. In the original article (not from time.com) it was 9 and 1. So I am not sure where the number came from.

Reply on news.ycombinator.com