Discrepancy marginaleffects in R and margins command in Stata

[pollyst8]

Hello,

My question refers to the R package #marginaleffects and the Stata command margins, which seem to return different estimates for equivalent functions executed in similar ways. The discrepancy is not too large but is not due to rounding rules. I would like to understand why.

Using the mtcars data available in base R,
car <- lm(mpg ~ wt * drat + am, data=mtcars)
library(marginaleffects)
predictions(car, newdata=datagrid(wt=seq(1, 5, by=.5), drat=c(2, 5)))

R retrurns:

wt drat Estimate Std. Error z Pr(>|z|) S 2.5 % 97.5 % am
1.0 2 21.12 4.70 4.490 <0.001 17.1 11.901 30.3 0
1.0 5 41.68 4.16 10.020 <0.001 76.1 33.530 49.8 0
1.5 2 20.50 3.98 5.156 <0.001 21.9 12.706 28.3 0

Using the same data in Stata, the following command:
reg mpg c.wt##c.drat am margins, at(wt=(1 (.5) 5) drat=(2 5))

---

         |            Delta-method
         |     Margin   Std. Err.      t    P>|t|     [95% Conf. Interval]

-------------+----------------------------------------------------------------
_at |
1 | 20.52733 4.781105 4.29 0.000 10.71731 30.33734
2 | 41.08823 3.768129 10.90 0.000 33.35666 48.81979
3 | 19.90286 4.069634 4.89 0.000 11.55266 28.25306

Thank you!

[vincentarelbundock]

These commands are not equivalent and should not produce the same results. There is an extensive comparison between the two software packages here:

https://marginaleffects.com/vignettes/alternative_software.html

[pollyst8]

Thank you, that's very helpful!