What is the meaning of fe in "ivcrc y (x=z), fe"? （no explanation in the help document)

[AlanLee1017]

Hi,
I am trying to run this program with fixed effects dummies using the i.identifier notation, and I can achieve it by adding "xi:" before the ivcrc comand. Luckily, I also found that ivcrc can be followed by a "fe" after the comma, like this "ivcrc y (x=z), fe", but there is no explanation in the help document . So, does the "fe" represent the fixed effect?
Best
Alan

[a-torgovitsky]

What type of specification are you trying to run?
Is this a panel with individual fixed effects?

[AlanLee1017]

What type of specification are you trying to run? Is this a panel with individual fixed effects?

Yes， a panel with individual fixed effects

[a-torgovitsky]

Can you describe the specification you want to run?

(Meaning what are the variables, what's endogenous, what's time-varying, what's the instrument, etc.)

[d-benson]

Unfortunately, "fe" is not an undocumented ivcrc option, so running

ivcrc y (x = z), fe

does not do anything with fixed effects. The module ignores "fe" and proceeds on the part it does understand: "ivcrc y (x =z)." The module does not automatically create dummy variables for fixed effects categories (factor variables more generally) or conduct within transformations the way commands like these

areg y x, absorb(id)

or

xtreg y x, fe

would work. However, stata's "xi: ..." wrapper is very flexible and can create temporary dummy variables from categorical data or IDs for most commands, including ivcrc. Other than using "xi: ...", you could manually generate dummies for each fixed effects category, or possibly use stata's within-transformation command "xtdata, fe" if demeaning makes econometric sense given the application.

[a-torgovitsky]

The econometric issue here is the first stage quantile regression, which you cannot apply the within transformation to, and which will be subject to the incidental parameters problem.
However, it's possible you could justify the assumptions on an outcome model that has already been within-transformed.
Depends a bit on the specifics.

[AlanLee1017]

Meaning what are the variables, what's endogenous, what's time-varying, what's the instrument, etc
It's actually an unbalanced panel (5 periods, on average, 5000 individuals/periods)
the specification is as following:
Y_{ict}： the dependent variable ( i-individual, c-county, t-time)
X_{c,t}： the independent variable
Z_{c,t}: the instrument
H_{ict}: the time-varying control variables (including individual\county levels)

[AlanLee1017]

Unfortunately, "fe" is not an undocumented ivcrc option, so running

ivcrc y (x = z), fe

does not do anything with fixed effects. The module ignores "fe" and proceeds on the part it does understand: "ivcrc y (x =z)." The module does not automatically create dummy variables for fixed effects categories (factor variables more generally) or conduct within transformations the way commands like these

areg y x, absorb(id)

or

xtreg y x, fe

would work. However, stata's "xi: ..." wrapper is very flexible and can create temporary dummy variables from categorical data or IDs for most commands, including ivcrc. Other than using "xi: ...", you could manually generate dummies for each fixed effects category, or possibly use stata's within-transformation command "xtdata, fe" if demeaning makes econometric sense given the application.

Thanks！

[AlanLee1017]

The econometric issue here is the first stage quantile regression, which you cannot apply the within transformation to, and which will be subject to the incidental parameters problem. However, it's possible you could justify the assumptions on an outcome model that has already been within-transformed. Depends a bit on the specifics.

Thanks！