Armorsmith Archive
Thread: Armor Repair question-truth or fiction
QuarEstee wrote:
Let's see...
Sample size is going to depend more on the number of variables than the scale of any one variable. The more variables you can control via experimental conditions, the fewer samples you'll need to tease out statistical significance. (As a psychological researcher, I know how to tease out meaning from the smallest sample sizes) You can always throw the variables into the mix instead of using design controls, but controlling them statistically will definitely drive up the number of samples needed.
I missed this when you posted, but with the recent thread I'll go ahead and respond ...
It's been a while since I've taken a formal statistics class, so bear with my poor terminology. My background is in engineering, so my natural inclination is to do as many tests as possible. I was suggesting a larger sample size primarily for better resolution, and to potentially account for unseen variations in the server's random number generation.
My understanding of the armor repair system, assuming that it operates the way most programmers and RPG creators would design it, is that each armor repair attempt begins with a "dice roll" ... the actual size of the dice being unknown. That base number is then modified by whatever other parameters are included in the design (innate skills? skill bonuses? tool quality? item condition? etc?). The final number is then compared to a table of values, with each value correlating to one of four outcomes.
Because we don't know much of the internal workings, a higher sample size simply seems prudent.
As you mention, the other potential variables on the player's end are easily controllable.
LeviticusD wrote:
I AMcurrently taking the classes and since I won't be graded on it, I have no desire to comment!
Hey! If we do the tests and the data entry, you want to run the stats? Maybe your prof will give you extra credit for it!
Agreed, tho, 100 attempts per subject is fine ... even 50 ... if every other factor is controlled by the design of the test.
Hmmm ... time to craft a bunch of bracers, get corpsed a dozen times, do all those repairs, write it down before I forget....
Upon further thinking, it occurred to me that the only question really worth asking (particularly in light of the original post ) is whether or not a Master Armorsmith has a lower rate of repair failure than a non-armorsmith. In this way, we can reduce the question to a binary form and effectively ignore the internal parameters of repair.
Furthermore, we're not only interested if a Master Armorsmith has a lower rate of failure, but whether that lower rate is significantly different to the failure rate for a non-armorsmith. I would submit that in terms of game-usefulness, this would be resolved in tests with sample sizes of 100-200.
Now I'm having flashbacks of my old graduate statistics textbook, so I should stop right there before I start talking about variances and what not. I hated that class. ::shudder::
Brutus_Krylop wrote:
Upon further thinking, it occurred to me that the only question really worth asking (particularly in light of the original post ) is whether or not a Master Armorsmith has a lower rate of repair failure than a non-armorsmith. In this way, we can reduce the question to a binary form and effectively ignore the internal parameters of repair.
Furthermore, we're not only interested if a Master Armorsmith has a lower rate of failure, but whether that lower rate is significantly different to the failure rate for a non-armorsmith. I would submit that in terms of game-usefulness, this would be resolved in tests with sample sizes of 100-200.
Now I'm having flashbacks of my old graduate statistics textbook, so I should stop right there before I start talking about variances and what not. I hated that class. ::shudder::
I AMcurrently taking the classes and since I won't be graded on it, I have no desire to comment!