stata-pretrends Multivariate Normal Plugin

Plugin to call Alan Genz's Fortran code

Apr 13 '24 04:04 mcaceresb

@jonathandroth I've implemented this in Unix. It seems the speed gains come at the cost of numerical precision. It's probably the correct trade-off, given how insanely slow this is at higher dimensions.

I'll merge later on. I assume I'll need a bit of time to debug the OSX compilation given the extra step of calling Fortran code.

Apr 13 '24 04:04 mcaceresb

@jonathandroth I got the Linux and windows working, and it compiles on x86_64 on a Mac. Can you test that out? TBD if I'll get the arm64 (silicon) to compile

Apr 13 '24 15:04 mcaceresb

Just to make sure, you want me to uninstall the package, then net install from your github and try running the example?

On Sat, Apr 13, 2024 at 11:12 AM Mauricio Caceres Bravo < @.***> wrote:

@jonathandroth https://github.com/jonathandroth I got the Linux and windows working, and it compiles on x86_64 on a Mac. Can you test that out? TBD if I'll get the arm64 (silicon) to compile

— Reply to this email directly, view it on GitHub https://github.com/mcaceresb/stata-pretrends/pull/6#issuecomment-2053673436, or unsubscribe https://github.com/notifications/unsubscribe-auth/AE6EXFF4D4ABJVJMJFUJSSLY5FDMHAVCNFSM6AAAAABGFEMCLOVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDANJTGY3TGNBTGY . You are receiving this because you were mentioned.Message ID: @.***>

Apr 16 '24 01:04 jonathandroth

@jonathandroth

If you have Intel then yes, and from this branch.
If you have Silicon then you'll have to compile first. You can run do src/install from this wording directory from inside Stata. LMK if it works or not (e.g. you may need to install the compiler from xcode).

Apr 16 '24 01:04 mcaceresb

I tried on Apple silicon but there appears to be a problem with the installation. Below is my log:

  name:  <unnamed>
   log:  /Users/jonathanroth/Library/CloudStorage/Dropbox/stata-pretrends-mauricio/stata-pretrends/ap

ple-silicon.log log type: text opened on: 16 Apr 2024, 18:10:14

. do src/install

. cap noi ado uninstall pretrends package not found

. cap noi net uninstall pretrends package not found

. mata: mata clear

. mata: mata set matastrict on

. mata: mata set mataoptimize on

. cap mkdir src

. cap mkdir src/build

. shell make all

rm -f src/build/pretrends_mvnorm_macosxarm64.plugin src/plugin/mvtdstpack.o gfortran -c src/plugin/mvtdstpack.f -o src/plugin/mvtdstpack.o -bundle -O3 -arch arm64 -std=legacy make: gfortran: No such file or directory make: *** [src/plugin/mvtdstpack.o] Error 1

. cap noi erase src/build/lpretrends.mlib

. qui {

. mata: mata mlib create lpretrends, dir("src/build") replace (file src/build/lpretrends.mlib not found) (file src/build/lpretrends.mlib created)

. mata: mata mlib add lpretrends PreTrends*(), dir("src/build") complete (15 functions added)

. net install pretrends, from(`c(pwd)') replace checking pretrends consistency and verifying not already installed... installing into /Users/jonathanroth/Library/Application Support/Stata/ado/plus/... installation complete.

. * f src/build/pretrends_mvnorm_macosx.plugin . * f src/build/pretrends_mvnorm_macosxarm64.plugin . end of do-file

. use "https://media.githubusercontent.com/media/mcaceresb/stata-pretrends/main/data/workfile_AEJ.dta", c

lear

. reghdfe l_poor_reg_rate Lead_D4_plus Lead_D3 Lead_D2 D0 Lag_D1 Lag_D2 Lag_D3_plus, absorb(v_id year) cl

uster(v_id) dof(none) (MWFE estimator converged in 5 iterations)

HDFE Linear regression Number of obs = 2,654 Absorbing 2 HDFE groups F( 7, 254) = 1.41 Statistics robust to heteroskedasticity Prob > F = 0.2021 R-squared = 0.6527 Adj R-squared = 0.6130 Within R-sq. = 0.0056 Number of clusters (v_id) = 255 Root MSE = 0.5659

                             (Std. err. adjusted for 255 clusters in v_id)

         |               Robust

l_poor_reg~e | Coefficient std. err. t P>|t| [95% conf. interval] -------------+---------------------------------------------------------------- Lead_D4_plus | .0667032 .0943746 0.71 0.480 -.1191533 .2525596 Lead_D3 | -.0077018 .0770514 -0.10 0.920 -.1594428 .1440392 Lead_D2 | -.0307691 .0551237 -0.56 0.577 -.1393268 .0777887 D0 | .0840307 .0626478 1.34 0.181 -.0393445 .2074059 Lag_D1 | .2424418 .0898103 2.70 0.007 .0655741 .4193096 Lag_D2 | .219879 .0887783 2.48 0.014 .0450438 .3947142 Lag_D3_plus | .1910925 .0989365 1.93 0.055 -.0037478 .3859329 _cons | 1.478639 .0811732 18.22 0.000 1.31878 1.638497

Absorbed degrees of freedom: -----------------------------------------------------+ Absorbed FE | Categories - Redundant = Num. Coefs | -------------+---------------------------------------| v_id | 255 0 255 | year | 12 1 11 ?| -----------------------------------------------------+ ? = number of redundant parameters may be higher

. pretrends power 0.5, pre(1/3) post(4/7) warning: option b() detected without option vcov() WARNING: Unable to load mvnormalcv() plugin; falling back on mata. Execution may be excessively slow. If it takes more a few minutes we recommend using the R package. Slope for 50% power = .0520259

. log cl name: log: /Users/jonathanroth/Library/CloudStorage/Dropbox/stata-pretrends-mauricio/stata-pretrends/ap

ple-silicon.log log type: text closed on: 16 Apr 2024, 18:10:41

Apr 16 '24 22:04 jonathandroth

@jonathandroth Can you install gcc (to get gfortran) and clang, then try again? If they're too cumbersome to install lmk.

Apr 16 '24 23:04 mcaceresb

I installed both but I'm getting the same error as before:

make: gfortran: No such file or directory make: *** [src/plugin/mvtdstpack.o] Error 1

However, if I go to the terminal and do

which gfortran

I get

/opt/homebrew/bin/gfortran

On Tue, Apr 16, 2024 at 7:17 PM Mauricio Caceres Bravo < @.***> wrote:

@jonathandroth https://github.com/jonathandroth Can you install gcc (to get gfortran) and clang, then try again? If they're too cumbersome to install lmk.

— Reply to this email directly, view it on GitHub https://github.com/mcaceresb/stata-pretrends/pull/6#issuecomment-2060057411, or unsubscribe https://github.com/notifications/unsubscribe-auth/AE6EXFAZDXECFH7X6K5JRKLY5WWOTAVCNFSM6AAAAABGFEMCLOVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDANRQGA2TONBRGE . You are receiving this because you were mentioned.Message ID: @.***>

Apr 17 '24 19:04 jonathandroth

@jonathandroth Are you running these from Stata? If so maybe there's an issue with the shell... Can you try make all from the repo root on a terminal and see if you get the same error?

If yes I guess I'll have to figure out another way to debug. If it compiles can you delete L8 of src/install.do then run it to reinstall?

Apr 17 '24 19:04 mcaceresb

I ran make all from the terminal and did not get the same Fortran error. It appears to have run though, although it did give the following warnings:

Warning: Rank mismatch in argument 'abserr' at (1) (scalar and rank-1)

clang -Wall -O3 -bundle -DSYSTEM=APPLEMAC -static-libgfortran -o src/build/pretrends_mvnorm_macosxarm64.plugin -arch arm64 src/plugin/mvtdstpack.o src/plugin/pretrends_mvnorm.c src/plugin/stplugin.c

clang: *warning: *argument unused during compilation: '-static-libgfortran' [-Wunused-command-line-argument]

ld: warning: object file (src/plugin/mvtdstpack.o) was built for newer macOS version (14.0) than being linked (12.3)

ld: warning: could not create compact unwind for mvkrsv: registers 27 and 28 not saved contiguously in frame

ld: warning: could not create compact unwind for mvkbrv: stack size is too large for frameless function

ld: warning: could not create compact unwind for mvbvtl: registers 78 and 79 not saved contiguously in frame

ld: warning: could not create compact unwind for mvbvu: registers 78 and 79 not saved contiguously in frame

ld: warning: could not create compact unwind for mvbvn: registers 21 and 22 not saved contiguously in frame

ld: warning: could not create compact unwind for mvbvt: registers 21 and 22 not saved contiguously in frame

ld: warning: could not create compact unwind for mvspcl: registers 72 and 73 not saved contiguously in frame

ld: warning: could not create compact unwind for mvswap: register 20 saved somewhere other than in frame

ld: warning: could not create compact unwind for mvsort: registers 78 and 79 not saved contiguously in frame

ld: warning: could not create compact unwind for mvlims: registers 74 and 75 not saved contiguously in frame

ld: warning: could not create compact unwind for mvchnc: registers 76 and 77 not saved contiguously in frame

ld: warning: could not create compact unwind for mvphnv: registers 72 and 73 not saved contiguously in frame

ld: warning: could not create compact unwind for mvvlsb: registers 76 and 77 not saved contiguously in frame

ld: warning: could not create compact unwind for mvchnv: registers 74 and 75 not saved contiguously in frame

ld: warning: could not create compact unwind for master.0.mvsubr.isra.0: register 73 saved somewhere other than in frame

ld: warning: could not create compact unwind for mvtdst: registers 21 and 22 not saved contiguously in frame

ld: warning: could not create compact unwind for mvbvtc: registers 72 and 73 not saved contiguously in frame

I then deleted line 8 as you suggested, and reran src/install from Stata. However, I still get:

[image: image.png]

If you'd like to come play with my Macstudio, LMK

On Wed, Apr 17, 2024 at 3:39 PM Mauricio Caceres Bravo < @.***> wrote:

@jonathandroth https://github.com/jonathandroth Are you running these from Stata? If so maybe there's an issue with the shell... Can you try make all from the repo root on a terminal and see if you get the same error?

If yes I guess I'll have to figure out another way to debug. If it compiles can you delete L8 of src/install.do then run it to reinstall?

— Reply to this email directly, view it on GitHub https://github.com/mcaceresb/stata-pretrends/pull/6#issuecomment-2062082996, or unsubscribe https://github.com/notifications/unsubscribe-auth/AE6EXFDVIWA67FVQP7NE2IDY53FWZAVCNFSM6AAAAABGFEMCLOVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDANRSGA4DEOJZGY . You are receiving this because you were mentioned.Message ID: @.***>

Apr 17 '24 19:04 jonathandroth

Sorry, the image didn't come through when responding via email

Apr 17 '24 19:04 jonathandroth

@jonathandroth Can you run test/unit-tests.do and LMK if you get any errors? I'll merge after that.

EDIT: From this branch.

Apr 28 '24 21:04 mcaceresb

I am getting the following error:

May 03 '24 16:05 jonathandroth

Full log below in case helpful:

do test/unit-tests.do

. set seed 1729

. version 15.1

. qui do test/unit-tests-basic.do

. qui do test/unit-tests-mvnorm.do

. * qui do src/ado/pretrends.ado . * qui do src/mata/pretrends.mata . . capture program drop main

. program main

```
unit_test basic_failures , noi
```
```
unit_test basic_checks   , noi
```
```
unit_test mvnorm_checks  , noi
```
end

. . capture program drop unit_test

. program unit_test

syntax namelist(min=1 max=1), [NOIsily tab(int 4) *]

```
cap `noisily' `namelist', `options'
```
```
if ( _rc ) {
```

    di as error _col(`=`tab'+1') `"test(failed): `namelist', `options'"'

```
    exit _rc
```
```
}
```

else di as txt _col(`=`tab'+1') `"test(passed): `namelist', `options'"'

end

scalars: r(LR) = 3.69200221757e-20 r(Bayes) = .2460845368842435 r(Power) = .800065433648153 r(slope) = .1704757216338253

macros: r(PreTrendsResults) : "PreTrendsResults"

matrices: r(results) : 14 x 6 r(delta) : 1 x 14

         |         t    betahat         lb         ub  deltatrue  meanAft~g

-------------+------------------------------------------------------------------ r1 | -6 5.800743 4.838816 6.76267 -.8523786 -.4317186 r2 | -5 3.970423 3.314403 4.626444 -.6819029 -.3412534 r3 | -4 3.436555 2.939535 3.933574 -.5114272 -.2505426 r4 | -3 2.288442 1.897956 2.678929 -.3409514 -.1469665 r5 | -2 1.152475 .8355832 1.469367 -.1704757 -.0435712 r6 | -1 0 0 0 0 0 r7 | 0 .0083689 -.3148921 .3316299 .1704757 .1907623 r8 | 1 -.9156951 -1.300725 -.5306651 .3409514 .3334039 r9 | 2 -2.000477 -2.464547 -1.536406 .5114272 .4706748 r10 | 3 -3.356143 -3.89576 -2.816525 .6819029 .6145395 r11 | 4 -4.075824 -4.691307 -3.460341 .8523786 .7577261 r12 | 5 -5.056917 -5.744972 -4.368862 1.022854 .9030402 r13 | 6 -5.966958 -6.744965 -5.188951 1.19333 1.045248 r14 | 7 -6.920232 -7.892136 -5.948327 1.363806 1.168329 warning: option b() detected without option vcov()

scalars: r(LR) = 8.22305473719e-24 r(Bayes) = .1336177366452207 r(Power) = .8914443562934214 r(slope) = .2

macros: r(PreTrendsResults) : "PreTrendsResults"

matrices: r(results) : 14 x 6 r(delta) : 1 x 14 test(passed): basic_checks, mvnorm (fortran): n = 1 in 0.000s: .867689 (eps: .) mvnormalcv(): 3200 conformability error PreTrends_mvnormalcv(): - function returned error : - function returned error test(failed): mvnorm_checks, r(3200);

May 03 '24 16:05 jonathandroth

@jonathandroth I believe I figured it out; the plugin was getting unloaded for the tests. Can you try again? When you confirm the tests run I'll merge.

May 11 '24 17:05 mcaceresb

The tests run without issue on my Macstudio.

Do you need me to test on my Intel Mac too?

May 20 '24 19:05 jonathandroth

@jonathandroth If you can swing it yeah that would be great!

May 20 '24 19:05 mcaceresb

Hmm I'm getting an error in the unit tests on my Intel Macbook. Here is the output.

. do test/unit-tests.do

. set seed 1729

. version 15.1

. qui do test/unit-tests-basic.do

. qui do test/unit-tests-mvnorm.do

. * qui do src/ado/pretrends.ado . * qui do src/mata/pretrends.mata . . capture program drop main

. program main

```
unit_test basic_failures , noi
```
```
unit_test basic_checks   , noi
```
```
unit_test mvnorm_checks  , noi
```
end

. . capture program drop unit_test

. program unit_test

syntax namelist(min=1 max=1), [NOIsily tab(int 4) *]

```
cap `noisily' `namelist', `options'
```
```
if ( _rc ) {
```

    di as error _col(`=`tab'+1') `"test(failed): `namelist', `options'"'

```
    exit _rc
```
```
}
```

else di as txt _col(`=`tab'+1') `"test(passed): `namelist', `options'"'

end

. . main assertion is false test(failed): basic_failures, r(9);

end of do-file

May 21 '24 13:05 jonathandroth

@jonathandroth Can you try

local c_os_ macosxarm64
cap program drop pretrends_mvnorm_plugin
program pretrends_mvnorm_plugin, plugin using("pretrends_mvnorm_`c_os_'.plugin")

and, regardless of whether there's an error,

local c_os_ macosx86_64
cap program drop pretrends_mvnorm_plugin
program pretrends_mvnorm_plugin, plugin using("pretrends_mvnorm_`c_os_'.plugin")

? If the first one gives an error and the second one succeeds, can you re-run the tests commenting out the basic_failures line? However, if the second one gives an error you might have to help me compile the intel version as well, sorry! LMK

May 21 '24 13:05 mcaceresb

The first code block failed. The second block succeeded.

I then commented out the line

 unit_test basic_failures , noi

in src/unit-tests.do, and re-ran the tests successfully.

May 21 '24 14:05 jonathandroth

@jonathandroth It might be a good idea to figure out which of these fail? But I'm good merging otherwise. Can you LMK which one gives the error when you get a chance?

    ereturn clear
    mata {
        st_matrix("beta",  PreTrendsExampleBeta())
        st_matrix("sigma", PreTrendsExampleSigma())
    }
    cap pretrends power
    assert _rc == 7

    cap pretrends power 0.5
    assert _rc == 198

    cap pretrends power 0.5, numpre(3)
    assert _rc == 198

    cap pretrends power 0.5, numpre(3) b(beta)
    assert _rc == 198

    cap pretrends power 0.5, numpre(3) b(beta) v(ZIGMA)
    assert _rc == 111

    cap pretrends power 0.5, numpre(3) b(beta) v(sigma) zz
    assert _rc == 198

    cap pretrends power 0.5, numpre(10) b(beta) v(sigma)
    assert _rc == 198

    cap pretrends power 0.5, numpre(7) b(beta) v(sigma)
    assert _rc == 198

    cap pretrends power 0.5, pre(1) post(2/10) b(beta) v(sigma)
    assert _rc == 198

    cap pretrends
    assert _rc == 198

    cap pretrends, slope(0.2)
    assert _rc == 198

    cap pretrends, slope(0.2) numpre(3)
    assert _rc == 198

    cap pretrends, slope(0.2) numpre(3) b(beta)
    assert _rc == 198

    cap pretrends, slope(0.2) numpre(3) b(beta) v(ZIGMA)
    assert _rc == 111

    cap pretrends, slope(0.2) numpre(3) b(beta) v(sigma) zz
    assert _rc == 198

    cap pretrends, slope(0.2) numpre(10) b(beta) v(sigma)
    assert _rc == 198

    cap pretrends, slope(0.2) numpre(7) b(beta) v(sigma)
    assert _rc == 198

    cap pretrends, slope(0.2) pre(1) post(2/10) b(beta) v(sigma)
    assert _rc == 198

May 21 '24 14:05 mcaceresb

Those all appear to run now...

On Tue, May 21, 2024 at 10:35 AM Mauricio Caceres Bravo < @.***> wrote:

@jonathandroth https://github.com/jonathandroth It might be a good idea to figure out which of these fail? But I'm good merging otherwise. Can you LMK which one gives the error when you get a chance?
ereturn clear
mata {
    st_matrix("beta",  PreTrendsExampleBeta())
    st_matrix("sigma", PreTrendsExampleSigma())
}
cap pretrends power
assert _rc == 7

cap pretrends power 0.5
assert _rc == 198

cap pretrends power 0.5, numpre(3)
assert _rc == 198

cap pretrends power 0.5, numpre(3) b(beta)
assert _rc == 198

cap pretrends power 0.5, numpre(3) b(beta) v(ZIGMA)
assert _rc == 111

cap pretrends power 0.5, numpre(3) b(beta) v(sigma) zz
assert _rc == 198

cap pretrends power 0.5, numpre(10) b(beta) v(sigma)
assert _rc == 198

cap pretrends power 0.5, numpre(7) b(beta) v(sigma)
assert _rc == 198

cap pretrends power 0.5, pre(1) post(2/10) b(beta) v(sigma)
assert _rc == 198

cap pretrends
assert _rc == 198

cap pretrends, slope(0.2)
assert _rc == 198

cap pretrends, slope(0.2) numpre(3)
assert _rc == 198

cap pretrends, slope(0.2) numpre(3) b(beta)
assert _rc == 198

cap pretrends, slope(0.2) numpre(3) b(beta) v(ZIGMA)
assert _rc == 111

cap pretrends, slope(0.2) numpre(3) b(beta) v(sigma) zz
assert _rc == 198

cap pretrends, slope(0.2) numpre(10) b(beta) v(sigma)
assert _rc == 198

cap pretrends, slope(0.2) numpre(7) b(beta) v(sigma)
assert _rc == 198

cap pretrends, slope(0.2) pre(1) post(2/10) b(beta) v(sigma)
assert _rc == 198
— Reply to this email directly, view it on GitHub https://github.com/mcaceresb/stata-pretrends/pull/6#issuecomment-2122782655, or unsubscribe https://github.com/notifications/unsubscribe-auth/AE6EXFDT7QRZBS7R4RBPOFLZDNLSZAVCNFSM6AAAAABGFEMCLOVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDCMRSG44DENRVGU . You are receiving this because you were mentioned.Message ID: @.***>

May 21 '24 14:05 jonathandroth

stata-pretrends stata-pretrends copied to clipboard

Multivariate Normal Plugin

stata-pretrends
stata-pretrends copied to clipboard