minor things but undocumented

[Al-Khwarizmi]

Roll 1k and we can discuss. Or at l;east 100 and call that pretesting like i do

Actually, a sample size of 10 can be good enough or not depending of the distributions of what you are measuring. If the difference between both samples is very obvious, 10 can be a sufficient size for it to be statistically significant.

You can use the website http://faculty.vassar.edu/lowry/VassarStats.html for significance testing, go to "t-tests and procedures" and to "two-sample t-test for independent correlated samples". There you would have to click "Independent samples", and in Sample A, write the PV's for your 10 dwarven wizzies and in Sample B for the elven ones. Then look at "Results", "P", "two-tailed". If the value of P is less than .01 (or .005, depending on how demanding you are); then your data are good enough and you can already conclude that there is in fact a difference between your two data sets. If the value is more than that, then either there is no difference, or there is a difference but you need to collect more data to get a significant test.

[Soirana]

10 characters

[Al-Khwarizmi]

if variable is between 1 and 4 [hardly remember better PV on robe] ten is not enough.

I have severe opinion about people making conclusions from few cases.

If you have for example (1,1,2,1,1,1,2,1,3,1) vs. (3,4,4,4,1,4,3,2,3,4) it is enough. If the values are more "interspersed" than that, it's possibly not enough. The test I pointed to tells you if it's enough or not in a particular case.

[Soirana]

10 characters

[Dudley]

where's that?

[Al-Khwarizmi]

Edit:Since you all will call me troll again
I entered your sequence into your calculator

result is...
0.130609

Um. I actually tried my example (I perfectly knew you were going to troll me if it wasn't correct :P), and what I got is p = 0.000243. One of us is doing something wrong.

[Soirana]

10 characters

[meh]

Technically, you need to have normally distributed variables for these kinds of tests. Of course, they produce a good indication for other distributions as well, but you shouldn't rely on the results too much in that case.

I haven't checked out that site yet, but performing standard t-tests on the data here tells that the means are different with any commonly used confidence level if we assume the data to be normally distributed.

[gut]

> Roll 1k and we can discuss.

When writing my druid guide, I did. Not 1000 GeWi and
1000 DwWi, but at least 1000 in total, that is when I
discovered it. I wasn't expecting anyone to take my
'roll' 10 PC's and see as statistical fact : ) I was expecting
someone who follows code diving stuff to chime in with
something like "That's been known about for a while..."

Since I now remember from where I did originally find
this quirk, I will post a section from my druid guide:

* Special note - In testing the starting equipment, I generated tons of characters. I noticed
(yet didn't record exact results) that some races seem to have a MUCH better chance, of
starting out with better than average items. Dwarves in particular, seem to have the best
chance, almost always starting with a few extra points of DV and PV on something in their
inventory. Drakelings are not far behind, and orcs also have higher than normal luck. I don't
know exactly what the average gains are, but I will give an educated guess. Dwarves average
about +2 points of DV or PV. Drakelings average about 1 point of DV or PV, as do orcs.

[Soirana]

10 characters