"option mlabformat() not allowed" error after boottest

[Sergey Alexeev]

Hey there,

I have a question on a user-written package boottest.

when I run a regression with clustered error and then perform a boottest

reg charlindex treat, cluster(StudyWard)
boottest treat=0

I get an error
option mlabformat() not allowed
which seems unusual considering that this a figure option

. reg charlindex treat, cluster(StudyWard)

Linear regression                               Number of obs     =         36
F(1, 16)          =       0.27
Prob > F          =     0.6101
R-squared         =     0.0102
Root MSE          =     1.3216

(Std. Err. adjusted for 17 clusters in StudyWard)

Robust
charlindex       Coef.   Std. Err.      t    P>t     [95% Conf. Interval]

treat      -.2625   .5047524    -0.52   0.610    -1.332527    .8075274
_cons       .8125   .4625186     1.76   0.098    -.1679956    1.792996


. boottest treat=0

Wild bootstrap-t, null imposed, 999 replications, Wald test, bootstrap clustering    by
>  StudyWard, Rademacher weights:
treat=0

t(16) =    -0.5201
Prob>t =     0.6146
option mlabformat() not allowed
r(198);

The data

* Example generated by -dataex-. To install: ssc install dataex
clear
input int StudyWard byte charlindex float treat
 19 0 1
142 0 1
114 0 0
119 0 1
157 0 0
114 3 0
119 0 1
143 0 0
216 0 1
157 0 0
143 0 0
211 0 1
142 0 1
215 1 1
217 0 0
 19 0 1
142 2 1
213 2 0
120 0 0
216 0 1
119 0 1
113 0 0
112 0 1
217 0 0
143 0 0
 15 6 0
 11 0 1
215 2 1
114 2 0
211 3 1
 16 0 0
113 0 0
112 1 1
211 0 1
119 0 1
112 2 1
end
label values StudyWard labels2
label def labels2 11 "1-7AS", modify
label def labels2 15 "1-8AS", modify
label def labels2 16 "1-8BN", modify
label def labels2 19 "1-9AN", modify
label def labels2 112 "2-TW", modify
label def labels2 113 "2-W1A", modify
label def labels2 114 "2-W1B", modify
label def labels2 119 "2-W2B", modify
label def labels2 120 "2-W2C", modify
label def labels2 142 "3-L5W", modify
label def labels2 143 "3-L6W", modify
label def labels2 157 "4-2A", modify
label def labels2 211 "5-X10N", modify
label def labels2 213 "5-X7N", modify
label def labels2 215 "5-X8N", modify
label def labels2 216 "5-X8S", modify
label def labels2 217 "5-X9N", modify

[Sergey Alexeev]

Oh gosh, all I needed was nograph option. I am terribly sorry for spamming the forum. I could not find an option to delete the original post.

[William Lisowski]

Oh gosh, all I needed was nograph option. I am terribly sorry for spamming the forum. I could not find an option to delete the original post.

Your posts aren't spam. Statalist isn't structured as a "help desk" but rather as a community forum where we learn from each other's experience. Your post and this discussion now are part of that searchable knowledge base. Your discussion of what worked for you is important and appreciated.

With that said, the error message

option mlabformat() not allowed

describes an option you did not specify. So I infer that somewhere in the depths of boottest.ado the code supplied an incorrect or inappropriate option to a graph command.

The boottest package is a community contributed package authored by David Roodman and available through SSC (an older version remains available from the Stata Journal). After installing the version from SSC today I see

. which boottest
/Users/lisowskiw/Library/Application Support/Stata/ado/plus/b/boottest.ado
*! boottest 4.1.0 13 July 2022
*! Copyright (C) 2015-22 David Roodman
*! Version history at bottom

When I run your code on your data with this version, the error does not occur.

Perhaps you have an older version installed and the current version includes a fix for the problem. The version history at the bottom of boottest.ado is extensive: TL;DR so I can't tell you if that is indeed the case.

You might try

ado uninstall boottest
ssc install boottest

and see if your luck is similar to mine.

Or if you have the version from SJ installed

ado uninstall st0549
ssc install boottest

In any event, this workaround is useful for those who aren't easily able to update their community-contributed commands.

[David Roodman]

What William said. Sergey, regardless of whether you want a graph, that crash should not happen. However, the immediate question is whether you have the latest version. If the problem persists with the latest version, let me know.

[Lawrence LaPlue]

I am receiving the same error message with a recent install on a new machine (running STATA 15.1).

I first installed boottest on this machine via st0549
After running boottest and receiving the message that the installation was likely out of date, and I ran:

ssc install boottest, replace

The boottest command began failing with the same error message Sergey noted.

Running the following:

use https://raw.githubusercontent.com/droodman/boottest/master/data/collapsed
regress hasinsurance selfemployed post post_self, cluster(year)
boottest post_self=.04

The boottest command was failing with the following error message: "option mlabformat() not allowed"

I uninstalled st0549 AND uninstalled boottest and then installed boottest again

ssc install boottest

and received the same error message (also restarted PC to clear memory).

The boottest installation appeared to be up to date:

. which boottest
c:\ado\plus\b\boottest.ado
*! boottest 4.3.0 17 November 2022
*! Copyright (C) 2015-22 David Roodman
*! Version history at bottom

I uninstalled boottest and installed st0549 again, and the code exercept example from above now, again, runs without issue (except for the prompt noting that it is likely out of date).

[David Roodman]

So if I understand right, the issue is not with boottest per se but with the installation, which should be controlled by the built-in ssc and ado commands. Clearly old and new versions are being mixed somehow. Not sure how to help...

[Sergey Alexeev]

Dear David Roodman,

Thank you very much for your careful reply. My apologies for being silent.

After updating Stata and boottest, the issue was resolved.

Regards,
Sergey

[Anupam Ghosh]

Hi, I am facing the same problem. I have a bigger chunk of code as given below. Please do let me know where I am erring

ssc install boottest, replace

program drop _all
program define cluster, rclass
syntax[, obs(integer 1) states(integer 1) ]

*** PART I: Create the simulated dataset ***

drop _all // Drop all variables
set obs `states' // Create one observation for each simulated U.S. state
gen state = _n // Give each simulated state an ID number
gen x = runiform(0.04,0.12) // Draw a "state unemployment rate" from a uniform distribution
gen eta = rnormal(0,1) // Grouped error term; notice it will be uncorrelated with x
expand `obs' // Expand the dataset so there are `obs' people in each state
* Now draw the outcome variable: "Log of earned income" for each person
* Notice that x is not in this data-generation process, so the true coefficient on x is 0
gen y = log(40000) + eta + rnormal(0,2)

* Run the regression with OLS variance estimator
reg y x
return scalar beta = _b[x] // Tell program to return the estimated coefficient on x
return scalar se_ols = _se[x] // Tell program to return the estimated OLS standard error
matrix A=r(table)
return scalar rej_ols = (A[4,1]<=.05) // This will be 1 if p-value less than 0.05, 0 otherwise

* Now run the regression with clustering at the state level
reg y x, vce(cluster state)
return scalar se_clu = _se[x] // Tell program to return cluster-robust SE
matrix B=r(table)
return scalar rej_clu = (B[4,1]<=.05) // =1 if p-value less than .05, 0 otherwise

* Now run the regression with a block bootstrap
reg y x, vce(bootstrap, reps(10) cl(state)) // I reduce the bootstrap reps to make it faster!
return scalar se_bbs = _se[x] // Block-bootstrap standard error
matrix C=r(table)
return scalar rej_bbs = (C[4,1]<=.05) // =1 if p-value less than .05, 0 otherwise

* Now run the regression with a wild bootstrap
reg y x, cluster(state) // I reduce the bootstrap reps to make it faster!
*waldtest x, cluster(state)
boottest x, reps(10)
return scalar se_wild = _se[x] // Wildcluster bootstrap
matrix D=r(table)
return scalar rej_wild = (D[4,1]<=.05)
end


* Now that the program has been created, this portion of the code uses
* the "simulate" command to run the "cluster" program 1,000 times. This
* simulates the sampling distribution of the OLS estimator of the coefficient on x.
* Note that the "rej_ols" and "rej_clu" variables should be very close to 0.05
* if my variance estimator is unbiased
set seed 12345
simulate beta=r(beta) se_ols=r(se_ols) se_clu=r(se_clu) se_bbs=r(se_bbs) se_wild=r(se_wild) ///
rej_ols=r(rej_ols) rej_clu=r(rej_clu) rej_bbs=r(rej_bbs) rej_wild=r(rej_wild), reps(1000): ///
cluster, obs(100) states(6)

* Now I'll summarize the results
su beta se_* rej_*

[Xin Nong]

I have Stata 14.2 and have run "ssc install boottest, replace," but the question "option mlabformat() not allowed" still exists.

ssc install boottest, replace
which boottest
webuse nlsw88, clear
regress wage tenure ttl_exp collgrad, cluster(industry)
boottest tenure, svmat

I then used remote desktop to connect to a University computer which has Stata 15, but the same question persists.

From discussions in this post I understand that it may be a version issue, but I still don't know how to solve it after trying all the recommendations.

Could anyone help me with this? Many thanks!

[David Roodman]

Thanks for reporting this. Yes, the code is using the mlabformat() option of the -scatter- command, which wasn't introduced till Stata 16.

I've just posted a new version of boottest on Github that I think prevents this crash. It should be on SSC soon.

Unfortunately, there is not a neat command to install from Github into Stata 14. Until the new version of boottest is on SSC, I recommend you manually download boottest.ado from github.com/droodman/boottest and save it over your current boottest.ado.

[Xin Nong]

The problem has been resolved! Thank you, David!

[Evelyn Intan]

Hi all,

Does anyone here also have the problem with boottest for the interaction between two categorical variables with the following error message?

"Null constraint applies only to omitted variables or base levels of factor variables."

Thank you in advance.

Best regards,
Evelyn