Strange group columns with subset and data.table in feols()

[adamaltmejd]

No idea whats going on here....

> modelsummary(feols(c(mpg, disp) ~ hp + drat + qsec, data = mtcars, subset = mtcars$gear > 3), output = "markdown")


|            |   mpg    |   disp    |
|:-----------|:--------:|:---------:|
|(Intercept) |  19.649  | -123.965  |
|            | (23.205) | (205.069) |
|hp          |  -0.053  |   1.021   |
|            | (0.022)  |  (0.195)  |
|drat        |  3.263   |  20.100   |
|            | (3.535)  | (31.242)  |
|qsec        |  -0.152  |   3.705   |
|            | (0.717)  |  (6.334)  |
|Num.Obs.    |    17    |    17     |
|R2          |  0.566   |   0.810   |
|R2 Adj.     |  0.466   |   0.766   |
|AIC         |  100.1   |   174.2   |
|BIC         |  103.5   |   177.6   |
|RMSE        |   3.64   |   32.14   |
|Std.Errors  |   IID    |    IID    |

> modelsummary(feols(c(mpg, disp) ~ hp + drat + qsec, data = dt, subset = dt[, gear > 3]), output = "markdown")


|            | group.x |   mpg    | group.y |   disp    |
|:-----------|:-------:|:--------:|:-------:|:---------:|
|(Intercept) |         |  19.649  |         | -123.965  |
|            |         | (23.205) |         | (205.069) |
|hp          |         |  -0.053  |         |   1.021   |
|            |         | (0.022)  |         |  (0.195)  |
|drat        |         |  3.263   |         |  20.100   |
|            |         | (3.535)  |         | (31.242)  |
|qsec        |         |  -0.152  |         |   3.705   |
|            |         | (0.717)  |         |  (6.334)  |
|Num.Obs.    |         |    17    |         |    17     |
|R2          |         |  0.566   |         |   0.810   |
|R2 Adj.     |         |  0.466   |         |   0.766   |
|AIC         |         |  100.1   |         |   174.2   |
|BIC         |         |  103.5   |         |   177.6   |
|RMSE        |         |   3.64   |         |   32.14   |
|Std.Errors  |         |   IID    |         |    IID    |

Especially since

> identical(mtcars$gear > 3, dt[, gear > 3])
[1] TRUE

Maybe its a fixest bug because:

> identical(
          feols(c(mpg, disp) ~ hp + drat + qsec, data = mtcars, subset = mtcars$gear > 3),
          feols(c(mpg, disp) ~ hp + drat + qsec, data = dt, subset = dt[, gear > 3])
      )
[1] FALSE
``

[vincentarelbundock]

Thanks a lot for the useful report. This should now be fixed on Github.