Announcement

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

  • Cox-Model with very few observations

    Hello everybody,

    I am researching peace agreements and their durability if they include gender provisions. I also control for many control variables, such as if the UN took part in the negotiations or gender equality before the conflict.
    Consequently, because of the control variables, my observations are only 37 in total.
    I have the following questions:

    What does this mean for the robustness of my result?
    What does it mean for generalization?
    How should I interpret my results?

    Thank you,
    Theresa

    Click image for larger version

Name:	Stata forum28.11.PNG
Views:	1
Size:	45.9 KB
ID:	1583755

  • #2
    Theresa:
    the only clear information is that you have a too small sample of observations, that you tortured with too many predictors.
    If you do not have the chance to collect more oservations, I'm afraid that any inference is unfeasible.
    Kind regards,
    Carlo
    (Stata 18.0 SE)

    Comment


    • #3
      Carlo is right, the more predictors you use, the less degrees of freedom remain. You have several options here:
      - only include the most relevant ones, based on theoretical assumptions
      - test them separately and build a model in a stepwise fashion by keeping ones that display statistical significance
      - try to combine predictors, maybe by using a principal component analysis or something similar

      And even if this works well, 33 cases is really low and it will be difficult to generalize these findings to a wider population.
      Best wishes

      (Stata 16.1 MP)

      Comment


      • #4
        To add to this, the number of events is the more relevant sample size here since they determine the shape of the survival curve. You only have 10 observations, and I'm always that inference may only be possible with a single predictor at best.

        Comment

        Working...
        X