marginal effects after STINTREG

[Shulav Neupane]

Hi,

I am using STINTREG to analyze interval data and as expected, the output is hazard ratios. I want to find the marginal effects instead and I used the following command for it.

stintreg <independent>, interval (lower upper) distribution(weibull)
margins, dydx(*)

I don't have much experience and was wondering if the estimates from the margins command could be interpreted as "a 1 unit change in <independent> causes a <coefficient> change in <dependent>. Thanks!

[Clyde Schechter]

First, unless there is information outside your data (or if the independent variable is, in fact, a random assignment to a treatment) you should not use causal language here. Say "is associated with" rather than "causes."

Next, different estimation commands use different default options for exactly what outcome statistic they calculate the predictive margins or marginal effects of, so it probably makes sense to be explict. In the case of -stintreg-, the default predictive outcome is median time.

Finally, there is a difference between marginal effects of discrete variables and marginal effects of continuous variables.

So, if your independent variable is discrete I would interpret it as "Compared to the reference category of <independent> being in the <other category of independent> category is associated with a <coefficient> difference in the median value of <dependent>."

If your independent variable is continuous, I would say: "The rate of change in the median value of <dependent> per unit difference in <independent> is <coefficient> <dependent units per independent units>."

[Shulav Neupane]

Thank you so much! That is very helpful.