This afternoon we studied hypothesis testing. In some ways it sounds so basic, but the method of getting the answer is rather involved.
For example:
You want to know if something made a change or not. So you have two statements:
There was no change.
There was a change.
You could take a guess, but why do that? Let's make some signal noise for numbers! Conceptually, all hypothesis tests are the same in that a signal (lambda)-to-noise (sigma) ratio is calculated (lambda/sigma) based on the before and after data. This ratio is converted into a probability, called the p-value. So now, calculate the p-value and compare it to the alpha. (The alpha is usually 0.05).
If:
p-value < alpha, go with statement 2 (there was a change).
p-value >= alpha, go with statement 1 (there was no change).
But only do this if your confidence level is 95% or higher.*
*Confidence levels depend upon each situation's set of circumstances. 95% is only a guideline.
So during my first quarter, I took a statistics class and learned all about p-variables and confidence levels and all of that. My professor was a very eloquent speaker, but had a seriously thick Chinese accent.
ReplyDeleteSo when we started talking about the bell curve on a graph and Z-statistics, she started talking about how the curve was "magical," and I'm like, yeah, I can see that statistics are kinda magical.
Then she says that in order to make them magical, you have to scandalize them.
It was at this point I realized that she was talking about the graph being symmetrical once you standardize the figures.
Oh, that was a fun class.
HAHA! What a great story. I will think about this throughout the day, and it will make me smile. If only the others in class knew why....
ReplyDelete:D