I'm not sure I've really got a very good understanding of what a null hypothesis is in the general sense. (And so I'll take this opportunity to ask....)
In the introduction to statistics and experimental design class that I had in high school, the idea of a null hypothesis was introduced in the context of a statistical "does item A have a correlation with item B" yes/no question -- the null hypothesis is essentially the "false" or "no change" one of the two possible boolean answers, in the context of testing this with a large number of random statistical samples.
The way that this related to engineering experiments -- which were, in my experience, generally of the "we measured the drag coefficient of this airfoil and it's 0.05 plus or minus 5%" sort, where the uncertainty was virtually all the sort of experimental bias error that doesn't show up as variance of the measurements -- wasn't really described, and the teacher sort of gave the impression that she thought all experiments ought to follow this pattern because that's how it was done in her field and that anyone who did it differently was simply wrong, and so I perhaps got a rather more limited impression of its applicability than would be accurate.
If I were to force the "Why did this person do that?" question into the mold I understand null hypotheses in, it would end up with something like "People who do 'that' do not have a statistically different frequency of such-and-so indicator of 'this' motivation from the overall population," for some given indicator of a motivation. Which rather twists the question a bit.
![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
no subject
Date: 2006-03-17 03:39 am (UTC)In the introduction to statistics and experimental design class that I had in high school, the idea of a null hypothesis was introduced in the context of a statistical "does item A have a correlation with item B" yes/no question -- the null hypothesis is essentially the "false" or "no change" one of the two possible boolean answers, in the context of testing this with a large number of random statistical samples.
The way that this related to engineering experiments -- which were, in my experience, generally of the "we measured the drag coefficient of this airfoil and it's 0.05 plus or minus 5%" sort, where the uncertainty was virtually all the sort of experimental bias error that doesn't show up as variance of the measurements -- wasn't really described, and the teacher sort of gave the impression that she thought all experiments ought to follow this pattern because that's how it was done in her field and that anyone who did it differently was simply wrong, and so I perhaps got a rather more limited impression of its applicability than would be accurate.
If I were to force the "Why did this person do that?" question into the mold I understand null hypotheses in, it would end up with something like "People who do 'that' do not have a statistically different frequency of such-and-so indicator of 'this' motivation from the overall population," for some given indicator of a motivation. Which rather twists the question a bit.