Dismiss Notice
Join Physics Forums Today!
The friendliest, high quality science and math community on the planet! Everyone who loves science is here!

Generalized likelihood ratio test

  1. May 3, 2005 #1
    This is the problem (t for theta):
    Code (Text):

    X ~ Expo(t) = t * e ^ (-t * x), x>0, t >0
                              0 otherwise
    Test H0: t <= 1 vs. H1: t >1 using the generalized likelihood ratio test where you have a random sample from X {X1, X2, ... , X50} and the sum of all Xi = 35. Use alpha = .05 (the probability of a type I error)

    My professor actually did much of this in class and I've asked him and he said it's not supposed to be a hard problem. I don't know. In class, this is the test he arrived at for the GLRT:
    reject H0 if

    ((t0 ^ n)* e^(-t0 * (sum over Xi))) / ((n/(sum over Xi))^n * e^(-n)) <= k

    n is the number of observations (50), t0 is... hmm, I think t0 is 1 in this case. k is the number that makes the test agree with alpha = .05.

    So now I need to simplify that expression so that I can find the probability that the resulting distribution is less than k. I can mess with k to make the left side simpler but I think I shouldn't do anything with (sum over Xi)--I think k should not depend on the values of the Xi's. I can only simplify it this far:
    n * ln (sum over Xi) - t0 * (sum over Xi) <= k1

    Do I need to use transformations to figure out how that messy thing on the left is distributed or am I doing something wrong? He said it shouldn't be that hard of a problem so I'm hesitant to use transformations.
  2. jcsd
  3. May 4, 2005 #2
    I did the uniformly most powerful test for that and got a reasonable answer (reject H0 if the sum over Xi <= 38.33). But the GLRT still escapes me... I'll try the transformation method for it later and see if that gets me somewhere.
  4. May 4, 2005 #3
    I'd like to take a look at this and try to answer your question, but I'm pretty swamped right now, so I'm not going to get to it anytime soon.
Share this great discussion with others via Reddit, Google+, Twitter, or Facebook