Announcement

Collapse
No announcement yet.
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • How to test the equality of the same coefficient in slightly different models

    Dear Statalist,

    We have been facing a truly terrible reviewer in one manuscript about obesity, nutrient intake and use of antidiabetic drugs.

    The bottom line is: one of the reviewers wants an additional "test", which I have no idea how to perform. Even though our manuscript is relatively robust with interaction terms and evidence that more complex models do not fit the data better than simpler models (substantiated via LR-tests), the reviewer wants a "test" between estimates of the same coefficient in correlated models. The main idea (I guess) is given in the example below.

    Code:
    set obs 1000
    gene y = rnormal(100,10)
    gene drug = runiform()>0.7
    gene a = runiform()
    gene b = runiform()
    regress y drug a
    regress y drug a b

    Typically, I compare different coefficients, say, test a==b, but the idea is (somehow) test drug==drug.

    Doest it make sense to compare the coefficient for drug (model 1) to drug (model 2) - considering the same subjects? Everything is equal, but covariate b, which is lacking in model 1.

    Any thoughts?

    All the best,

    Tiago

  • #2
    The circumstances under which I would be interested in the null hypothesis or the alternative hypothesis don't occur to me right now, but
    Code:
    sysuse auto
    sureg (mpg i.foreign c.weight) (mpg i.foreign c.weight c.displacement)
    test [mpg]1.foreign = [2mpg]1.foreign

    Comment


    • #3
      As always, thank you so much, Joseph.

      Not sure if that test follows a standard chi-square distribution under the null, but this exactly the "test" I have been looking for.

      Thanks again.

      Tiago

      Comment


      • #4
        Hazarding a guess, I'm wondering about - nestreg - command. You can get the Wald test for each block. You may also get "the likelihood test for nested model specifications", as well as AIC and BIC for each block. If Joseph's advice won't please the reviewer, you may add this suggestion.
        Best regards,

        Marcos

        Comment


        • #5
          Interesting, Marcos! I will try it. If it works, it will be a clever approach.

          All the best,

          Tiago

          Comment


          • #6
            Shall you decide to use nestreg, please let us know whether it ‘worked’, Thiago.
            Best regards,

            Marcos

            Comment

            Working...
            X