Finding marginal effects for sampleSelection model

Question

I am estimating a gravity trade model using the sampleSelection package and having trouble calculating marginal effects. I would like to use the marginaleffects package. The model converges with results below.

The data is available here: (I would make it smaller, but I cannot figure out how to reduce the size and keep the problem).

The model was having trouble converging so I followed the instructions to this two-step process.

selec_lm_1206_nums <- 
bin_1206 ~ pct_lag_1_yr_of_ukr_sunflowers + log(total_world_exports_1206) + member_wto_d + agree_pta_goods + covid_year + deaths + events + deaths * agree_pta_goods + events * agree_pta_goods

outcom_lm_1206_nums <- 
log(netweight_kg_1206) ~  pct_lag_1_yr_of_ukr_sunflowers + log(total_world_exports_1206) + member_wto_d + agree_pta_goods + covid_year + deaths + events + deaths * agree_pta_goods + events * agree_pta_goods + month_february + month_march + month_april + month_may + month_june + month_july + month_august + month_september + month_october + month_november + month_december + partner_armenia + partner_australia + partner_austria + partner_azerbaijan + partner_bangladesh + partner_belarus + partner_belgium + partner_bulgaria + partner_china + partner_czech_rep +  partner_denmark + partner_egypt + partner_estonia + partner_finland + partner_france + partner_georgia + partner_germany + partner_hungary + partner_india + partner_iraq + partner_italy + partner_lebanon + partner_morocco + partner_netherlands + partner_pakistan + partner_poland + partner_portugal + partner_rep_of_moldova + partner_romania + partner_russian_federation + partner_spain +    partner_sweden + partner_switzerland + partner_turkey + partner_united_kingdom

selec_1206_nums_2S <- selection(selection = selec_lm_1206_nums, outcome = outcom_lm_1206_nums, data = ukr_exports_analysis_sunflowers, method = "ml",maxMethod="SANN", parscale = 0.001)

selec_1206_nums <- selection(selection = selec_lm_1206_nums, outcome = outcom_lm_1206_nums, data = ukr_exports_analysis_sunflowers, method = "ml", start = coef(selec_1206_nums_2S))

summary(selec_1206_nums)

> summary(selec_1206_nums)
--------------------------------------------
Tobit 2 model (sample selection model)
Maximum Likelihood estimation
Newton-Raphson maximisation, 8 iterations
Return code 2: successive function values within tolerance limit (tol)
Log-Likelihood: -6585.88 
4194 observations (2297 censored and 1897 observed)
68 free parameters (df = 4126)
Probit selection equation:
                                 Estimate Std. Error t value Pr(>|t|)    
(Intercept)                     1.879e+01  4.317e+00   4.352 1.38e-05 ***
pct_lag_1_yr_of_ukr_sunflowers  2.864e-02  2.618e-03  10.938  < 2e-16 ***
log(total_world_exports_1206)  -8.678e-01  1.955e-01  -4.438 9.31e-06 ***
member_wto_d1                  -1.166e-01  5.973e-02  -1.952   0.0510 .  
agree_pta_goods1                5.668e-01  6.893e-02   8.223 2.63e-16 ***
covid_year1                     1.320e-01  7.484e-02   1.764   0.0779 .  
deaths                          4.391e-05  2.100e-04   0.209   0.8344    
events                         -3.939e-03  3.351e-03  -1.175   0.2399    
agree_pta_goods1:deaths         2.469e-04  3.932e-04   0.628   0.5301    
agree_pta_goods1:events         1.666e-03  5.520e-03   0.302   0.7627    
Outcome equation:
                                 Estimate Std. Error t value Pr(>|t|)    
(Intercept)                    33.9286921 10.5381573   3.220 0.001294 ** 
pct_lag_1_yr_of_ukr_sunflowers  0.0289553  0.0081910   3.535 0.000412 ***
log(total_world_exports_1206)  -1.0689663  0.4786482  -2.233 0.025582 *  
member_wto_d1                  -1.5041634  0.4674271  -3.218 0.001301 ** 
agree_pta_goods1                0.5015109  0.2095119   2.394 0.016723 *  
covid_year1                     0.2456088  0.1737226   1.414 0.157496    
deaths                          0.0013708  0.0005869   2.336 0.019560 *  
events                         -0.0376453  0.0088565  -4.251 2.18e-05 ***
month_february1                -0.0758067  0.1989450  -0.381 0.703191    
month_march1                   -0.1663013  0.1959581  -0.849 0.396121    
month_april1                    0.1921004  0.2014316   0.954 0.340304    
month_may1                     -0.0005563  0.2059650  -0.003 0.997845    
month_june1                    -0.1007850  0.2146457  -0.470 0.638708    
month_july1                    -0.1424848  0.2174332  -0.655 0.512309    
month_august1                  -0.6256302  0.2279437  -2.745 0.006083 ** 
month_september1                0.0350345  0.2076408   0.169 0.866020    
month_october1                  0.4296219  0.1991467   2.157 0.031039 *  
month_november1                 0.4897676  0.2045072   2.395 0.016671 *  
month_december1                -0.0395080  0.1948822  -0.203 0.839358    
partner_armenia1                0.6667811  0.4272304   1.561 0.118670    
partner_australia1              2.0320576  1.0742590   1.892 0.058616 .  
partner_austria1               -4.4739967  0.3911026 -11.439  < 2e-16 ***
partner_azerbaijan1             0.0777747  0.4583665   0.170 0.865272    
partner_bangladesh1             3.3733063  1.2954631   2.604 0.009249 ** 
partner_belarus1               -2.0996295  0.4959205  -4.234 2.35e-05 ***
partner_belgium1                2.2120262  0.4485863   4.931 8.50e-07 ***
partner_bulgaria1               1.7822446  0.3461572   5.149 2.75e-07 ***
partner_china1                  1.1775708  0.4552847   2.586 0.009731 ** 
partner_czech_rep1              0.4901475  0.4240523   1.156 0.247803    
partner_denmark1                0.3459561  0.3971738   0.871 0.383781    
partner_egypt1                  1.2197630  0.3712942   3.285 0.001028 ** 
partner_estonia1                0.1315394  0.3839395   0.343 0.731913    
partner_finland1                1.7183312  0.3989766   4.307 1.69e-05 ***
partner_france1                -5.1335809  0.3603393 -14.247  < 2e-16 ***
partner_georgia1                1.2181676  0.3618550   3.366 0.000768 ***
partner_germany1                2.1981464  0.3376699   6.510 8.43e-11 ***
partner_hungary1               -0.7684804  0.4972521  -1.545 0.122313    
partner_india1                  1.4954587  0.3779445   3.957 7.72e-05 ***
partner_iraq1                   0.0907309  0.4449478   0.204 0.838431    
partner_italy1                  0.4845864  0.3986407   1.216 0.224208    
partner_lebanon1               -0.2231801  0.4473566  -0.499 0.617886    
partner_morocco1                2.9440561  0.8725557   3.374 0.000748 ***
partner_netherlands1            1.6506210  0.3821066   4.320 1.60e-05 ***
partner_pakistan1               2.5107821  0.7899278   3.178 0.001491 ** 
partner_poland1                 1.6749368  0.3356626   4.990 6.29e-07 ***
partner_portugal1               1.1662223  0.5492737   2.123 0.033796 *  
partner_rep_of_moldova1         0.8918645  0.3736330   2.387 0.017031 *  
partner_romania1                1.4345824  0.3438745   4.172 3.08e-05 ***
partner_russian_federation1     2.2659206  0.4644291   4.879 1.11e-06 ***
partner_spain1                  0.2770208  0.3671673   0.754 0.450603    
partner_sweden1                 0.8975827  0.4214778   2.130 0.033263 *  
partner_switzerland1            1.9644726  0.7822877   2.511 0.012070 *  
partner_turkey1                 2.8141320  0.4212573   6.680 2.70e-11 ***
partner_united_kingdom1        -0.0482882  0.4040543  -0.120 0.904878    
agree_pta_goods1:deaths        -0.0001057  0.0009334  -0.113 0.909866    
agree_pta_goods1:events         0.0251029  0.0133068   1.886 0.059301 .  
   Error terms:
      Estimate Std. Error t value Pr(>|t|)    
sigma  2.17716    0.08559   25.44   <2e-16 ***
rho    0.65668    0.05969   11.00   <2e-16 ***
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
--------------------------------------------

When I attempt to find the marginal effects:

> mfx_s <- marginaleffects(selec_1206_trends)
Error: Unable to extract the data from model of class `selection`. This can happen in a variety of cases, such as when a `marginaleffects` package function is called from inside a user-defined function. Please supply a data frame explicitly via the `newdata` argument.
> mfx_s <- marginaleffects(selec_1206_trends, newdata = datagrid())
Error: object of type 'symbol' is not subsettable
In addition: Warning message:
Could not get model data. 
> mfx_s <- marginaleffects(selec_1206_trends, newdata = ukr_exports_analysis_sunflowers)
Error: object of type 'symbol' is not subsettable
> marginaleffects(selec_1206_nums, newdata = datagrid(newdata = ukr_exports_analysis_sunflowers))
Error: object of type 'symbol' is not subsettable

I am not certain what data to give it. I have seen means and medians and using the original data. I have tried giving it the original data in a couple of ways shown above. Really interestingly, if I give the marginal effects function the initial guess the function says the selection models are not supported, but the documentation says they are.

> marginaleffects(selec_1206_nums_2S, newdata = datagrid(newdata = ukr_exports_analysis_sunflowers))
Error: Models of class "selection" are not supported.

I would like to generate a conditional marginal effects plot. Any ideas?

Answer 1

The original poster shared data and code with me, and it helped me identify a bug which is now fixed. In principle, installing the development versions of the insight and marginaleffects packages should solve this problem:

library(remotes)
install_github("easystats/insight")
install_github("vincentarelbundock/marginaleffects")

Make sure you restart R completely. Then,

library("sampleSelection")
library("marginaleffects")

data("Mroz87", package = "sampleSelection")
Mroz87$kids  <- (Mroz87$kids5 + Mroz87$kids618 > 0)
dat <- Mroz87

mod <- selection(lfp ~ age + I( age^2 ) + faminc + kids + educ,
                 wage ~ exper + I( exper^2 ) + educ + city, 
                 data = dat)
mfx <- marginaleffects(mod)

summary(mfx)
#> Average marginal effects 
#>     Term     Contrast     Effect Std. Error z value  Pr(>|z|)      2.5 %
#> 1    age        dY/dX -7.759e-03  2.901e-03  -2.675 0.0074752 -1.344e-02
#> 2 faminc        dY/dX  2.120e-06  7.637e-07   2.775 0.0055131  6.228e-07
#> 3   kids TRUE - FALSE -1.622e-01  9.040e-02  -1.794 0.0728393 -3.393e-01
#> 4   educ        dY/dX  3.556e-02  2.135e-02   1.666 0.0958117 -6.287e-03
#>       97.5 %
#> 1 -2.074e-03
#> 2  3.617e-06
#> 3  1.502e-02
#> 4  7.740e-02
#> 
#> Model type:  selection 
#> Prediction type:  response