reghdfe
reghdfe copied to clipboard
sergiocorreia
assert(): 3498 assertion is false
I'm not sure what has changed, because I ran this code earlier and it worked, but now I get an error with following syntax (there are five f_stacked* and five lambda_* variables)
reghdfe ln_E_rest L1_ln_E_rest L2_ln_E_rest L6_ln_E_rest ln_MW L1_ln_MW L2_ln_MW L6_ln_MW ln_pop ln_E_tot, absorb(County#c.f_stacked* month#c.lambda_* County month)
assert(): 3498 assertion is false
fixed_effects(): - function returned error
<istmt>: - function returned error
I also tried reghdfe ln_E_rest L1_ln_E_rest L2_ln_E_rest L6_ln_E_rest ln_MW L1_ln_MW L2_ln_MW L6_ln_MW ln_pop ln_E_tot, absorb(County##c.f_stacked* month##c.lambda_*)
But then the interactions are only kept for the first instance of each variable (output below explains this better than I can in words)
Absorbed degrees of freedom:
-----------------------------------------------------------------+
Absorbed FE | Categories - Redundant = Num. Coefs |
-------------------------+---------------------------------------|
County | 1030 0 1030 |
County#c.f_stacked1 | 1030 0 1030 ?|
f_stacked2 | 282 1 281 |
f_stacked3 | 282 282 0 ?|
f_stacked4 | 282 282 0 ?|
f_stacked5 | 282 282 0 ?|
month | 282 282 0 ?|
month#c.lambda_stacked1 | 282 0 282 ?|
lambda_stacked2 | 1030 1030 0 ?|
lambda_stacked3 | 1030 1030 0 ?|
lambda_stacked4 | 1030 1030 0 ?|
lambda_stacked5 | 1030 1030 0 ?|
-----------------------------------------------------------------+
Is there something I'm doing wrong? I'm using Stata 14.2, reghdfe version 4.3.6 28jun2017
PS: This seems to be where things go wrong:
- mata: HDFE = fixed_effects(`"`absorb' `comma' `options'"', "`touse'", "`weight'", "`exp'", `drop_singletons', `verbose')
= mata: HDFE = fixed_effects(`"County#c.f_stacked* month#c.lambda_* County month , "', "__000001", "", "", 1, 0)
assert(): 3498 assertion is false
fixed_effects(): - function returned error
<istmt>: - function returned error
---------------------------------------------------------------------------------------------------------------------------------------------- end reghdfe.Parse ---
--------------------------------------------------------------------------------------------------------------------------------------------- end reghdfe.Estimate ---
- Cleanup `c(rc)'
= Cleanup 3498
PS2: Manually expanding the * seems to work:
reghdfe ln_E_rest L1_ln_E_rest L2_ln_E_rest L6_ln_E_rest ln_MW L1_ln_MW L2_ln_MW L6_ln_MW ln_pop ln_E_tot, absorb(${panelvar}##c.(f_stacked1 f_stacked2 f_stacked3 f_stacked4 f_stacked5) ${timevar}##c.(lambda_stacked1 lambda_stacked2 lambda_stacked3 lambda_stacked4 lambda_stacked5))
-----------------------------------------------------------------+
Absorbed FE | Categories - Redundant = Num. Coefs |
-------------------------+---------------------------------------|
County | 1030 0 1030 |
County#c.f_stacked1 | 1030 0 1030 ?|
County#c.f_stacked2 | 1030 0 1030 ?|
County#c.f_stacked3 | 1030 0 1030 ?|
County#c.f_stacked4 | 1030 0 1030 ?|
County#c.f_stacked5 | 1030 0 1030 ?|
month | 282 1 281 |
month#c.lambda_stacked1 | 282 0 282 ?|
month#c.lambda_stacked2 | 282 0 282 ?|
month#c.lambda_stacked3 | 282 0 282 ?|
month#c.lambda_stacked4 | 282 0 282 ?|
month#c.lambda_stacked5 | 282 0 282 ?|
-----------------------------------------------------------------+
I don't recall if the code in absorb allows * expansions; at least I haven't used them before.
Also, I would strongly suggest if you replace "country#var country" with "country##var), it makes a large difference on speed and numerical accuracy.
Here's an older log (I don't know which reghdfe version I used back then)
. reghdfe $depvar ${Lx_$depvar} ${Lx_$indepvar} $controls, absorb(${panelvar}##c.f_stacked* ${timevar}##c.lambda_*) cluster($clustervar) residuals(res_`name') // Clustered
(converged in 3 iterations)
Absorbed degrees of freedom:
----------------------------------------------------------------------------+
Absorbed FE | Num. Coefs. = Categories - Redundant |
--------------------------+-------------------------------------------------|
County | 0 1030 1030 * |
County#c.f_stacked1 | 1030 1030 0 ? |
County#c.f_stacked2 | 1030 1030 0 ? |
County#c.f_stacked3 | 1030 1030 0 ? |
County#c.f_stacked4 | 1030 1030 0 ? |
County#c.f_stacked5 | 1030 1030 0 ? |
month | 281 282 1 |
month#c.lambda_stacked1 | 282 282 0 ? |
month#c.lambda_stacked2 | 282 282 0 ? |
month#c.lambda_stacked3 | 282 282 0 ? |
month#c.lambda_stacked4 | 282 282 0 ? |
month#c.lambda_stacked5 | 282 282 0 ? |
----------------------------------------------------------------------------+
For now I just fixed it by using the ds command, but I don't know how convenient that is in an ado/mata context ...
ds f_stacked*
local fvars = "`r(varlist)'"
ds lambda_*
local lambdavars = "`r(varlist)'"
reghdfe ln_E_rest L1_ln_E_rest L2_ln_E_rest L6_ln_E_rest ln_MW L1_ln_MW L2_ln_MW L6_ln_MW ln_pop ln_E_tot, absorb(${panelvar}##c.(`fvars') ${timevar}##c.(`lambdavars'))
By the way, the code that parses absorb() is actually in ftools and you can access it directly from Stata:
sysuse auto, clear
ms_parse_absvars foreign##c.*h*, noisily
ms_parse_absvars foreign##c.(*h*), noisily
It's in it's own .ado so it can be called from other programs (e.g. for IV regression), but that also makes it easier to fix.
In this case, with parenthesis it works correctly but without them it forgets to add a parenthesis; fixing it might be fairly straightforward.
Ah, good to know.
Just FYI, this also causes issues with the regife command.
regife $depvar $indepvar $controls, f(lambdai = $panelvar ft = $timevar, 7) a(uPart = $panelvar tpart = $timevar) vce(cluster $clustervar) residuals(res_`name')
internall call to reghdfe failed (error code: 3498). Returning the estimate without standard errors.
REGIFE Number of obs = .
Panel structure: County, month Wald chi2() = .
Factor dimension: 7 Prob > chi2 = .
Converged: true Iterations = 1681
last estimates not found
r(301);
TBH the new version causes issues with every command that depends on reghdfe (regife, poi2hdfe, etc.), which is one of the reasons I haven't updated the SSC version.
Oh the joys of software development... :D
On 14 July 2017 at 03:33, Sergio Correia [email protected] wrote:
TBH the new version causes issues with every command that depends on reghdfe (regife, poi2hdfe, etc.), which is one of the reasons I haven't updated the SSC version.
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/sergiocorreia/reghdfe/issues/104#issuecomment-315245262, or mute the thread https://github.com/notifications/unsubscribe-auth/AUBCSjEJ8FuUaKu3Ou8-UOEkkZ_7T4TRks5sNsWAgaJpZM4OVb04 .
Hi guys, FYI - I have the July 13, 2017 version installed, and it crashes Stata every time I use reghdfe - unless using the option "old". System: OSX El Capitan (both on iMac and on macbook pro). On the windows server it does not seem to cause problems.
Hi Glenn,
What happens if you type "reghdfe, compile"? (And then maybe close and open stata)
Hi Sergio,
thanks, works after a reboot of the system! I must have missed that information, apologies.
Hi Glenn,
I haven't heard of that problem before, but restarting usually solves most problems :) (although I am worried that when I upload to SSC I could get many issue reports)
I had to compile and restart the system in the end... Happy to send feedback. cheers Glenn