Announcement

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

  • FE, OLS & negbin

    Hi there,

    I have a dataset which consists of patient characteristics with 13,165 observations. I would like to do a fixed effects OLS and negbin model with my physician characteristics dataset (292 observations).

    Do I need to merge/append both datasets in order to do the regressions mentioned above and if so, how do I do that? Both datasets have one variable in common. How do I merge/append both datasets with the help of that one variable, that both datasets have in common?

    Thank you in advance!
    Thea

  • #2
    Let's start with the merge portion of the question. It will probably be useful to consult the Stata help for merge. You can either type "help merge" or consult the online documentation here.

    Here is a toy example for how to think about the merge:
    Code:
    * Load the physician dataset
    use "physician" // my name for your physician dataset
    duplicates r physicianid // this is an id for the physician (I'm assuming this is the "one variable in common")
    * Merge - my assumption is that there is a one physician to many patients link using physicianid
    merge 1:m physicianid using "patient" // patient is my made up name for the patient dataset
    Once you've successfully merged your datasets, please share more information about your data and the model you'd like to estimate. Providing more complete information will likely lead to faster and better answers!

    Comment


    • #3
      Thea:
      welcome to this forum.
      As an aside to Arthur's helpful advice, please note that OLS with FE and negative binomial regression are totally different beasts.
      Hence, you should have two different type of regressands to apply them properly (continuous if OLS; taking on count values=>0 if negative binomial regression).
      Eventually, please note that oftentimes negative binomial regression is the fix to an overdispersed -poisson- (that should be tried before negative binomial regression).
      Kind regards,
      Carlo
      (Stata 18.0 SE)

      Comment


      • #4
        The only restriction to use Poisson regression, whether the usual pooled Poisson or FE Poisson, is that y >= 0, preferably without a natural upper bound. One can always estimate a linear model, by fixed effects or not, for any y. Whether it's the best model depends on the nature of y. But for comparison it's not a bad idea.

        Thea needs to post much more to get decent advice. I suspect mine will be to use linear FE and the Poisson FE, but I need to know what y is and the exact data structure.

        Comment

        Working...
        X