Announcement

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

  • ordinal logistic, ologit postestimation, margins, probabilities, odds, odds-ratios

    Hello,
    my apologies for such a basic question. Tried to find the answer throughout Statalist without success.
    I am using ologit with the following expression:

    svy: ologit var_ord i.catvarA

    where var_ord has 5 levels, and catvarA has 4 levels

    I am able to obtain the predicted probabilities for each level of var_ord using the postestimation command margins either for all the levels of var_ord, or for a specific level

    svy: ologit var_ord i.catvarA
    margins catvarA


    or

    svy: ologit var_ord i.catvarA
    margins catvarA, predict(pr outcome(#))


    My understanding is that the margins predicted with commands above give me the probability of outcome(#) for each level of catvarA.

    Two questions:
    First, should I be using the atmeans after the margins var, predict command as in

    margins catvarA, predict(pr outcome(#)) atmeans

    Or could I simply use margins without specifying atmeans as below

    margins catvarA, predict(pr outcome(#))

    Second (the main question),
    is there a way to convert the probabilities in odds, and estimate odds-ratios from the probabilities generated by margins command above?

    I realize that I could calculate manually from the probabilities using the formula odds=p/(1-p), and then estimate the odds-ratios by dividing the odds of each level of catvarA by whatever level of catvarA I choose as the baseline. However, I am sure there must be a way to have Stata estimate the odds and odds-ratios, as opposed to probabilities.

    I realize that either of the two commands below:

    svy: ologit var_ord i.catvarA, eform

    or

    svy: ologit var_ord i.catvarA, or

    will give me the odds (exp(b)), as opposed to the coefficient, but I also realize that those cannot be interpreted as we would the outcome of a logistic regression for a binary outcome variable. The application of this regression is in epidemiology, so reporting the results from the ordinal logistic regression using ologit in odds-ratio (95%CI) would be aligned with the standard in the literature.

    In summary, I want to be able to generate odds-ratios as one would get for a binary outcome using the command:

    svy: logit var_bin i.catvarA, eform

    Any guidance is greatly appreciated.

    Thank you.

Working...
X