0
$\begingroup$

I have a mixed model with multiple covariates. I want to test if some of the covariates are insignificant. Since I have mixed model the LR-test will result in a p-value which is too low, hence I want to use bootstrap to do the hypothesis test. Can someone explain the principle in how this could be done? Should I simulate new LR-tests statistics and do something with these or..?

I know how to use bootstrap to make confidence intervals for single parameters, but I'm not sure how to do when I want to remove multiple parameters (i.e. a single covariate) like in this case.

Improve this question
$\endgroup$
5
  • $\begingroup$ "Since I have mixed model the LR-test will result in a p-value which is too low" - Why? $\endgroup$
    – Bryan Krause
    Commented Jun 18, 2018 at 21:22
  • $\begingroup$ The LR-test assume that observations are independent and have equal variance, but in my model I also have random effects. Pinheiro et al. write that it tends to be anti-conservative when fixed effects are tested. $\endgroup$
    – J. Jensen
    Commented Jun 19, 2018 at 5:54
  • $\begingroup$ How large is your sample? $\endgroup$
    – Bryan Krause
    Commented Jun 19, 2018 at 6:07
  • $\begingroup$ 5000 approximately $\endgroup$
    – J. Jensen
    Commented Jun 19, 2018 at 6:21
  • $\begingroup$ I believe the anticonservative issues with likelihood ratio tests fall off with larger sample sizes; 5000 is plenty large. $\endgroup$
    – Bryan Krause
    Commented Jun 19, 2018 at 6:52

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.