Large samples? 1 non-branded and 9 branded articles went missing. That's not a huge number of cases to examine.

Actually, it's plenty. The number of lost parcels should be, roughly speaking, a Poisson distributed statistic. If we assume a fixed rate of parcel loss, the number of parcels lost from any given batch of shipments should come in at that rate, plus or minus some noise. For this type of statistical distribution, the standard deviation from batch to batch is approximately the square root of the expectation value.

For 178 parcels (89 under each condition) the observed losses were 1 parcel (1.1%) for non-atheist packages, and 9 parcels (10%) for atheist packages. If we suppose that the actual loss rate is between those two extremes, we get a loss rate of about 5.5%, and an *expected* loss of 5 parcels per batch of 89 parcels.

The standard deviation for that batch size is the square root of 5, or about 2.2; the two observations that we have are both about two standard deviations away from the expectation value. The likelihood of pulling a random value this far from the expectation value by chance is about 5%; the likelihood of it happening twice is 5% squared: about 0.03%.

Feel free to twiddle with different expectation values and expected loss rates; you'll find the odds are strongly against these values coming up by chance.

The "3 days longer" statistic seems to be massively skewed by a single non-representative parcel that took 37 days later than its counterpart.

Roughly 80 parcels and roughly 40 days' delay means that the mean was increased by about half a day. Discarding that one outlying datum, the atheist packages still would have averaged 2.5 days longer for delivery. Among the ones that were delivered at all....