![]() |
|
Basic stats help - Printable Version +- Southperry.net (https://www.southperry.net) +-- Forum: Social (https://www.southperry.net/forumdisplay.php?fid=14) +--- Forum: Rubik's Cube (https://www.southperry.net/forumdisplay.php?fid=58) +--- Thread: Basic stats help (/showthread.php?tid=25421) |
Basic stats help - Takebacker - 2010-05-06 I need to understand the process for these problems but i have little clue on how to do any of it. >_> The chapter is on hypothesis testing.
Spoiler
Thanks a million guys. ;~; Basic stats help - Dusk - 2010-05-06 Looks like you should Google level of significance, seeing as how that's in every problem. (Sorry, never took stats, can't help) Basic stats help - Takebacker - 2010-05-06 The level of significance is supposed to be what you compare the p-value to at the end to reject or not reject the null hypothesis. If i can get the p-value for each problem then i can solve them all, but getting the p-value is the hard part because finding it depends on the problem... I solved one of them, so i'll take it out. Basic stats help - Kaasoljoyyx - 2010-05-06 I can go through one detailed later if you don't understand, but this is basically what you do Note: I'm using notation and terminology from my class, so if something is unclear, ask Ho = Null hypothesis Ha = Alternative hypothesis These all sound like LRT (Likelihood ratio tests) tests, also known as Nayman Pearson, and not GLRT (Generalized LRT). Basically you're checking 2 different types of data and comparing them, seeing if they are statistically different with a certain percent of confidence, and if you're checking 1-sided or 2-sided i.e. You have a coin and you flip it 10 times. Coin = 7 heads With 95% confidence (1-alpha, so alpha is 5%) can you say it isn't fair? Probably with instinct, you can see 7 heads happening, it's not that unlikely and it will still be a fair coin. But say it was 70 heads out of 100, then you will probably say it's unfair. The main thing you're looking for is the Rejected region (RR), and this is where 2-sded and 1-sided come up in. Ho = " U = Uo" Ha = " U !=Uo" That's a 2 sided test. A 1 sided test would look something like Ho = " U < or = Uo" Ha = " U > Uo" Anyway, I'm assuming you're doing LRT, so you need this formula (I can't really write it out so i'll describe it). Numerator and denominator as follows On top: The product (Capital Pi symbol) of f ( paramater | x) under Ho The f(p|x) is the distribution function using the paramter involved with Ho The bottom: the same thing but it's under Ha (Usually you need to use MLE as your best guess) Call that quantity lambda. Now you check to see if lambda is less than a constant C, which if it is, you're in the RR, and therefore reject your Ho. C will depend on the level of alpha you chose If you are not using LRT, you usually just assume that lambda is your best guess, which is X bar. Then you just do this as follows alpha = P ( |Xbar| < c | Ho) <-- Notice the absolute value around Xbar, this is a sign of a 2-sided test Now normalize it alpha = P ( { |Xbar| - Uo } / (sigma / SQRT)n) < {c - Uo} / {sigma / SQRT (n)} ) Now that first quantity you normalized it to the stardard normal Z, so you just have alpha = P ( Z < c*) where c* is that mess above Then you get Z (a/2) = c* You can then get c*, obtain c, and see if your statistic falls in the RR ----------------- I know that's really confusing, so here are some helpful wiki things to explain it http://en.wikipedia.org/wiki/Likelihood-ratio_test Look for: Simple-versus-simple hypotheses This is when you have Ho = X Ha = Y Where both tests are defined for a single number. When you use LRT, for Ha, just use the value given instead of the MLE like I used in the above example. After you understand 1 of these (Coins usually easiest), the rest are easy to follow. Same process over and over, just different numbers. |