Choose which dummy to drop in feols - fixest package

Question

I have two fixed effect variables and I'm interested in estimating coefficients of one fixed effect, but feols automatically drops the first one always.

For example, in my toy data like below:

library(data.table)
library(fixest)
set.seed(1)

test_dt = data.table(
  y = rnorm(12, 15,20),
  x = rnorm(12, 25, 5),
  TIMESLOT = rep(c('BEG','MID','END'),4),
  FIRM = rep(c('A','B','C','D'), 3)
)
print(test_dt)

            y        x TIMESLOT FIRM
 1:  2.470924 21.89380      BEG    A
 2: 18.672866 13.92650      MID    B
 3: -1.712572 30.62465      END    C
 4: 46.905616 24.77533      BEG    D
 5: 21.590155 24.91905      MID    A
 6: -1.409368 29.71918      END    B

I'm trying to estimate coefficients of TIMESLOT fixed effect (removing intercept), and I'm not interested in coef of FIRM.

So my first approach was to run

feols(y ~ 0 + i(TIMESLOT) +  i(TIMESLOT, x) + FIRM, test_dt) # WORKS, and drops one of firm

Which yields what I need - all of TIMESLOT coefficients.

The variable 'FIRMD' has been removed because of collinearity (see $collin.var).
OLS estimation, Dep. Var.: y
Observations: 12 
Standard-errors: IID 
                  Estimate Std. Error   t value Pr(>|t|) 
TIMESLOT::BEG   -31.539770  107.34824 -0.293808  0.78806 
TIMESLOT::END    16.660959   45.20014  0.368604  0.73690 
TIMESLOT::MID    72.309104   64.56608  1.119924  0.34432 
TIMESLOT::BEG:x   2.672511    4.23945  0.630391  0.57319 
TIMESLOT::END:x   0.477604    1.97274  0.242102  0.82431 
TIMESLOT::MID:x  -1.121334    2.42085 -0.463199  0.67476 
FIRMA           -17.185688   19.01731 -0.903687  0.43277 
FIRMB           -35.705475   25.40943 -1.405206  0.25461 
FIRMC           -17.706701   21.30094 -0.831264  0.46680 
... 1 variable was removed because of collinearity (FIRMD)

My real data has lots of unique values of FIRM so I'd like to put it as fixed effect variable. Unfortunately, feols removes one of my dummy variable of interest.

For example,

feols(y ~ 0 + i(TIMESLOT) +  i(TIMESLOT, x)| FIRM, test_dt)

yields

OLS estimation, Dep. Var.: y
Observations: 12 
Fixed-effects: FIRM: 4
Standard-errors: Clustered (FIRM) 
                  Estimate Std. Error   t value Pr(>|t|) 
TIMESLOT::END    48.200729   63.44549  0.759719  0.50267 
TIMESLOT::MID   103.848874  123.20970  0.842863  0.46120 
TIMESLOT::BEG:x   2.672511    3.39202  0.787882  0.48828 
TIMESLOT::END:x   0.477604    1.75570  0.272030  0.80325 
TIMESLOT::MID:x  -1.121334    2.21219 -0.506888  0.64714 
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
RMSE: 9.82348     Adj. R2: -0.468187
                Within R2:  0.403153

So now TIMESLOT::BEG is dropped. I would rather want one of firm dummies to be dropped, but I had no success.

Finally,

I saw y and X, and fixef_df arguments in feols function and was trying to use this path as my last resort.

However, I'm getting some error messages that I can't interpret. My trial is like below.

feols(y= test_dt$y, X = test_dt[,.(i(TIMESLOT), i(TIMESLOT,x))], fixef_df = test_dt[,.(i(FIRM, ref ='A'))])

#Error in get("isFixef", env) : argument "env" is missing, with no default

What could be the way to manually drop one of dummy variable here? I'd appreciate your answers.

(fixest package is not tagged in SO currenlty)

Answer 1

Response from the dev: there's no way to manually drop one.