So, it's pretty clear that a lot of this can be written off as confirmation bias (specifically biased memory). Even if the numbers are wrong, I guarantee that people will think they're worse than they are, simply because you remember the times that were frustrating more than the times that were easy. http://en.wikipedia.org/wiki/Confirmation_bias#Biased_memory
That said, I'm surprised no one's done any actual statistical analysis on the numbers people have been throwing out. Without such analysis, even the most convincing numbers are entirely meaningless, because we don't know how likely those results actually are. For example, Awong, your data on the first page. I did a couple quick (1-tailed) t-tests, and your first test has a .23% chance (p = .0023) to occur if the chance of learning a spell is actually 75%. Since scientific convention generally uses 5% (p = .05) as the realistic cutoff point between reasonably likely and highly unlikely, we can tentatively conclude that your first set had less than a 75% chance of success. Your second set, the one without reloading, tells a different story. It has a 10.73% chance (p = .1073) of occurring randomly, which is above our 5% threshold. Therefore, we conclude that your second result is within the realm of reasonable likelihood for something with a 75% success rate. From this data, I surmise that your theory about getting a bad random number seed on the first test was correct, and that there is no reason to believe that the chance isn't 75%.
Meanwhile, Troodon, I also ran the numbers for your sample of 3000, and found that your results are within the 54.32% range (p = .5432). In other words, if the true probability is 85%, random chance is actually more likely to give you a lower result than a higher one. Of course you knew that, because you learned 85.1% of spells correctly. So, basically, what I'm saying is that I agree with your interpretation of your results.
My F/M with 18 INT just failed 5 times in a row trying to learn "Grease"...
At least it wasn't Stoneskin... Mind you for something as useful as that, I'd quaff a potion, even with 18 Int...
I just save before trying and reload if it fails. To me this whole issue is just an annoyance. Having 100% learn spells is pretty much the only thing I like from "Normal" difficulty, and would use it if it's available as an option on "Core".
Comments
http://en.wikipedia.org/wiki/Confirmation_bias#Biased_memory
That said, I'm surprised no one's done any actual statistical analysis on the numbers people have been throwing out. Without such analysis, even the most convincing numbers are entirely meaningless, because we don't know how likely those results actually are. For example, Awong, your data on the first page. I did a couple quick (1-tailed) t-tests, and your first test has a .23% chance (p = .0023) to occur if the chance of learning a spell is actually 75%. Since scientific convention generally uses 5% (p = .05) as the realistic cutoff point between reasonably likely and highly unlikely, we can tentatively conclude that your first set had less than a 75% chance of success. Your second set, the one without reloading, tells a different story. It has a 10.73% chance (p = .1073) of occurring randomly, which is above our 5% threshold. Therefore, we conclude that your second result is within the realm of reasonable likelihood for something with a 75% success rate. From this data, I surmise that your theory about getting a bad random number seed on the first test was correct, and that there is no reason to believe that the chance isn't 75%.
Meanwhile, Troodon, I also ran the numbers for your sample of 3000, and found that your results are within the 54.32% range (p = .5432). In other words, if the true probability is 85%, random chance is actually more likely to give you a lower result than a higher one. Of course you knew that, because you learned 85.1% of spells correctly. So, basically, what I'm saying is that I agree with your interpretation of your results.
I just save before trying and reload if it fails. To me this whole issue is just an annoyance. Having 100% learn spells is pretty much the only thing I like from "Normal" difficulty, and would use it if it's available as an option on "Core".