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

A Does Delta Method work for asymptotic distributions?

  1. Apr 24, 2017 #1
    So if I have a logistic regression: ##log (\hat {odds})=\hat{\beta_{0}}+\hat{\beta_{1}}x##. How would I find a confidence interval for x if I am given ##odds=5## This is going in reverse, where if I have the outcome, I try to do inference on the predictor.

    We know that ##\hat{\vec{\beta}}##, the vector of ##\hat{\beta_{0}}## and ##\hat{\beta_{1}}##, is distributed asmptotically as ##N( \vec{\beta}, \hat{{Var{\hat{\beta}}}})## for large sample sizes where ##\hat{{Var{\hat{\beta}}}}## is the estimated variance covariance matrix.( similar to how ##\sigma^{2}## is estimated by ##s^{2}## for the basic t test).

    So solving for x using ##log (5)=\beta_{0}+\beta_{1}x## I get: ##x( \beta_{0},\beta_{1}) =\frac{ log(5)-\beta_{0}} {\beta_{1} }## a multivariate function of both parameters.

    So it seems that I would use the delta method but the one problem is the variance covariance matrix is an estimated one, so it doesn't have any actual parameters in it. Would delta method still work for this? Or would I need to try another method?
     
  2. jcsd
  3. Apr 25, 2017 #2

    DrDu

    User Avatar
    Science Advisor

    No, that's ok. The point is that you can Taylor expand ##Var_\beta=Var_\hat{\beta}+(\partial Var_\beta/\partial \beta )(\hat{\beta}-\beta)+\ldots##.
    Now ##\hat{\beta}-\beta## is of order ##1/\sqrt{N}##, so that asymptotically, you can neglect the second term.
     
Know someone interested in this topic? Share this thread via Reddit, Google+, Twitter, or Facebook

Have something to add?
Draft saved Draft deleted



Similar Discussions: Does Delta Method work for asymptotic distributions?
  1. Why does it work? (Replies: 4)

  2. How Does This Work (Replies: 3)

Loading...