Thread: likelihood ratio test and AIC

2. Re: likelihood ratio test and AIC

So this is what I've done so far,

for likelihood ratio
model 1:
(Lo+94.27)/Lo

model 2:
(Lo+90.83)/Lo

Lo in the two models are equal, so model 2 is better. (model 2 < model 1)

for AIC:
model 1:
2 *(94.27-5)=178.54

model 2:
2*(90.83-8)=165.66

so model 2 better.

is this correct?

3. Re: likelihood ratio test and AIC

A likelihood ratio test isn't just seeing which likelihood is greater. The full model (model 2) is guaranteed to have a bigger likelihood. You need to do an actual statistical test.

For the AIC calculation you need to be more careful with the signs. Note that your log likelihoods are negative. You can't just ignore that. Also the fact that your AICs don't match the given AICs should hint to you that maybe you messed up the calculation somewhat.

4. Re: likelihood ratio test and AIC

Originally Posted by Dason
A likelihood ratio test isn't just seeing which likelihood is greater. The full model (model 2) is guaranteed to have a bigger likelihood. You need to do an actual statistical test.

For the AIC calculation you need to be more careful with the signs. Note that your log likelihoods are negative. You can't just ignore that. Also the fact that your AICs don't match the given AICs should hint to you that maybe you messed up the calculation somewhat.

What kind of tests do you mean because what we have learned is

if D1<D2 then model 1 better
D=(Lo-Lm)/Lo

ohh yes. The signs. Thank you for pointing that out. I didn't pay attention.

5. Re: likelihood ratio test and AIC

Google likelihood ratio test. Then note that there is more to it than just looking at the values and seeing which one is 'better'. Like I said the likelihood for the full model will always be "better" so just comparing directly doesn't really tell us anything. You need to do an actual test. And if they're asking you to do a likelihood ratio test then I'm assuming that the class covered that.

6. Re: likelihood ratio test and AIC

Originally Posted by Dason
Google likelihood ratio test. Then note that there is more to it than just looking at the values and seeing which one is 'better'. Like I said the likelihood for the full model will always be "better" so just comparing directly doesn't really tell us anything. You need to do an actual test. And if they're asking you to do a likelihood ratio test then I'm assuming that the class covered that.
LRT=-2(94.27-(-90.83))=6.88
Chi Square (0.05) df=3 = 7.815
since
6.88<7.815 not significant
so model 1 is better.

is this correct?

 Tweet

Posting Permissions

• You may not post new threads
• You may not post replies
• You may not post attachments
• You may not edit your posts