The sample is allways unbiased unless proven otherwise.
I'm sorry, you've got that backwards. A sample is only useful if the margin for error is known. The margin for error can only be known if all possible measures to eliminate selection bias have taken. Ideally that means a fully blind selection of the full population.
Since this survey is taking place on a fansite forum, you've inherently created a sub-set of the population that has different characteristics than the target population. We have no data on the characteristic differences of the population of this forum from the population at large, so our margin for error is undefined. Data=garbage.
As a profesional statiscian, this is a really good answer. It will be as this guy pointed out till somebody demonstrates that this fansite forum represents the diablo 3 population. By the way, same would happen with the Swedish survey
Actually the size of the population is not terribly important when determine margin for error and sample size. 1000 people should get you around or under a 3% margin of error for a poll like this.
The real problem, is that the participants in the survey are not representative of the population in question, so none of the data turned out by the poll is useful for any question beyond determining opinions of D3fans.com forum users.
Just as a simple lesson. With an infinite population (> ~1.000.000) and taking into account the worst scenario possible (p=q=0,5) and a confidence margin of 95%; the error of having a sample is, as the guy before said, around 3%. This is quite good. We only have the epistemological problem of: ¿Is this fansite representative for the population of d3 players?
And I can assume with almost certainty that this sample is not a good representation of the population of players.
No, no you can't. That's the entire point, the sample doesn't give you the right to any non-trivial knowledge claims.
Nice argument but you can't also claim that its a good representation.Nevertheless the point is that we can't take that survey as valid for whole Diablo 3 gamers population because, as you said, the sample doesn't gee you the right to any non.trivial knowledge claims.
PS: Maybe i should include a David Hume's quote as signature
266 is not a decent sample size for a study of this magnitude. I doubt even 3300 is, but at least it's somewhat closer to the truth. Also, none of these polls will reflect the truth because a majority of players doesn't frequent forums, so taking people from only one place can result in some unwanted variables.
I found the Wizard to be kinda fun in beta, but I don't feel he have anything very funny endgame. Meteor was nerfed to pieces, teleport is a joke compared to other displacers... everything fun with the d2 sorc has gone out the window. Still looking to be my 2nd/3rd class though, but only cause I dislike WD and DH and don't want to play two melees in a row.
Sorry but 3300 is enough and 266 is not bad for a good precision.
On the other hand the problem relies in who are the voters, if your 3300 voters are a very specific population and doesn't represent the overall population then its useless no matter if you have 3300 or 300 000 voters.
And I can assume with almost certainty that this sample is not a good representation of the population of players.
As i said in my post before, this is exactly what i meant
Really interesting thread. I work as statistic so maybe i can help.
a) 2xx sample size has a huge experimental error because population is nearly infinite (some millions)
b )Swedish sample looks fine but only can represent swedish population, you cannot asume that Swedish population = World population.
Now as a gamer:
I will play a Sorc (sorry preffer "Sorc" to "Wizard") and i dont care about underpower. Moreover, in many games, sorc class normally is at lower possitions in early levels but are very devastating in high levels (even in Diablo 2!)
Rollback Post to RevisionRollBack
To post a comment, please login or register a new account.
As a profesional statiscian, this is a really good answer. It will be as this guy pointed out till somebody demonstrates that this fansite forum represents the diablo 3 population. By the way, same would happen with the Swedish survey
Totally failed statistics.
As this guy said:
Just as a simple lesson. With an infinite population (> ~1.000.000) and taking into account the worst scenario possible (p=q=0,5) and a confidence margin of 95%; the error of having a sample is, as the guy before said, around 3%. This is quite good. We only have the epistemological problem of: ¿Is this fansite representative for the population of d3 players?
Nice argument but you can't also claim that its a good representation.Nevertheless the point is that we can't take that survey as valid for whole Diablo 3 gamers population because, as you said, the sample doesn't gee you the right to any non.trivial knowledge claims.
PS: Maybe i should include a David Hume's quote as signature
As i said in my post before, this is exactly what i meant
a) 2xx sample size has a huge experimental error because population is nearly infinite (some millions)
b )Swedish sample looks fine but only can represent swedish population, you cannot asume that Swedish population = World population.
Now as a gamer:
I will play a Sorc (sorry preffer "Sorc" to "Wizard") and i dont care about underpower. Moreover, in many games, sorc class normally is at lower possitions in early levels but are very devastating in high levels (even in Diablo 2!)