Options
Every totalme option, with an example of each
totalme varlist [if] [in] [, options]The examples use Stata’s nlsw88 data with occupation collapsed to three groups and modeled with mlogit.
sysuse nlsw88, clear(NLSW, 1988 extract)drop if missing(occupation, wage, ttl_exp)(9 observations deleted)recode occupation (1/2 = 1 "Professional/managerial") (3/5 = 2 "Sales/clerical/craft") ///
(6/13 = 3 "Other"), gen(occ3)(1,920 differences between occupation and occ3)
. quietly mlogit occ3 i.collgrad i.race c.age i.south, vce(robust)
estimates store basemod
quietly mlogit occ3 i.collgrad i.race c.age i.south c.ttl_exp c.wage, vce(robust)
estimates store medmodmodels() – one model or two
models() is optional for one model (the estimates in memory are used) and required for two. With two models the Total ME is reported for each and the difference is tested; the two models can be the same or different estimation commands. vce(robust) is strongly recommended on both.
totalme collgrad age, models(medmod)Total ME Estimates (N = 2237)
| Estimate Std. err. z P>|z|
-------------------------+-----------------------------------------------
collgrad |
College gr vs Not colleg | 0.264 0.023 11.527 0.000
-------------------------+-----------------------------------------------
age |
+ 1 (centered) | 0.007 0.003 2.267 0.023 totalme collgrad age, models(basemod medmod)Total ME Estimates (N_basemod = 2237 , N_medmod = 2237)
| Estimate Std. err. z P>|z|
-------------------------+-----------------------------------------------
collgrad |
Model 1 (basemod) | 0.248 0.021 11.572 0.000
Model 2 (medmod) | 0.264 0.023 11.527 0.000
Cross-Model Diff. | -0.016 0.008 -2.049 0.040
-------------------------+-----------------------------------------------
age + 1 (centered) |
Model 1 (basemod) | 0.005 0.003 1.831 0.067
Model 2 (medmod) | 0.007 0.003 2.267 0.023
Cross-Model Diff. | -0.001 0.001 -1.152 0.249 amount(), centered, uncentered – the size of a continuous change
For a continuous variable the default is a one-unit change, centered on the observed value. amount() takes a number, sd, 2sd, trimrange (5th to 95th percentile), range (minimum to maximum), p#-p# (from one percentile to another, e.g. p10-p90), or rate (the instantaneous rate of change); one value applies to all continuous variables, several apply in order. uncentered makes the change run upward from the observed value.
totalme age, models(medmod) amount(sd)Total ME Estimates (N = 2237)
| Estimate Std. err. z P>|z|
-------------------------+-----------------------------------------------
age |
+ SD (centered) | 0.020 0.009 2.268 0.023 totalme age, models(medmod) amount(trimrange)Total ME Estimates (N = 2237)
| Estimate Std. err. z P>|z|
-------------------------+-----------------------------------------------
age |
p5 to p95 | 0.058 0.026 2.280 0.023 totalme age, models(medmod) amount(p10-p90)Total ME Estimates (N = 2237)
| Estimate Std. err. z P>|z|
-------------------------+-----------------------------------------------
age |
p10 to p90 | 0.058 0.026 2.280 0.023 totalme age, models(medmod) amount(10) uncenteredTotal ME Estimates (N = 2237)
| Estimate Std. err. z P>|z|
-------------------------+-----------------------------------------------
age |
+ 10 | 0.060 0.024 2.494 0.013 start() – where the change begins
totalme age, models(medmod) start(age=40) amount(5)Total ME Estimates (N = 2237)
| Estimate Std. err. z P>|z|
-------------------------+-----------------------------------------------
age |
start (40) + 5 (centere) | 0.032 0.014 2.297 0.022 atmeans – covariates at their means
totalme collgrad, models(medmod) atmeansTotal ME Estimates (N = 2237)
| Estimate Std. err. z P>|z|
-------------------------+-----------------------------------------------
collgrad |
College gr vs Not colleg | 0.281 0.025 11.105 0.000 weighted, unweighted, all – nominal variables
For a nominal variable the per-outcome quantity is the ME inequality of meinequality, so the Total ME is a Total ME inequality, weighted by category shares by default. unweighted gives every pairwise comparison the same weight; all reports both. For a continuous or binary variable these options are ignored and the variable keeps its one row.
totalme race, models(medmod)Total ME Estimates (N = 2237)
| Estimate Std. err. z P>|z|
-------------------------+-----------------------------------------------
race |
total ME Ineq. | 0.147 0.043 3.427 0.001 totalme race, models(medmod) unweightedTotal ME Estimates (N = 2237)
| Estimate Std. err. z P>|z|
-------------------------+-----------------------------------------------
race |
Unwgt total ME Ineq. | 0.149 0.051 2.922 0.003 totalme race, models(medmod) allTotal ME Estimates (N = 2237)
| Estimate Std. err. z P>|z|
-------------------------+-----------------------------------------------
race |
total ME Ineq. | 0.147 0.043 3.427 0.001
Unwgt total ME Ineq. | 0.149 0.051 2.922 0.003 groups and groupnames() – models fit on distinct samples
groups specifies that the two models used for comparison are fit on distinct samples; each model’s marginal effects are then averaged over its own sample. group(varname), the syntax of earlier versions, is also accepted; varname must take one value in each model’s sample and a different value in each model.
quietly mlogit occ3 i.collgrad c.age if south == 0, vce(robust)
estimates store nonsouth
quietly mlogit occ3 i.collgrad c.age if south == 1, vce(robust)
estimates store south. totalme collgrad, models(nonsouth south) groups groupnames(NonSouth South)Total ME Estimates (N_nonsouth = 1300 , N_south = 937)
| Estimate Std. err. z P>|z|
-------------------------+-----------------------------------------------
collgrad |
Model 1 (NonSouth) | 0.277 0.028 9.886 0.000
Model 2 (South) | 0.228 0.036 6.292 0.000
Cross-Model Diff. | 0.049 0.046 1.079 0.281 by() and over() – within levels of another variable
by(varname) computes the Total ME at each level of a binary or nominal covariate as a counterfactual (the whole sample set to that level); over(varname) computes it within each observed subgroup. The variable must be in the model with the i. prefix.
With either option the table also holds a Diff. row for each pair of levels – the Total ME at the first level minus that at the second, with its standard error and test, which is the test of whether the effect differs across the groups (a test of interaction). Neither option may name one of the focal variables; mecompare handles that case.
totalme collgrad, models(medmod) by(south)Total ME Estimates (N = 2237)
| Estimate Std. err. z P>|z|
-------------------------+-----------------------------------------------
collgrad(Not south) |
College gr vs Not colleg | 0.272 0.024 11.451 0.000
-------------------------+-----------------------------------------------
collgrad(South) |
College gr vs Not colleg | 0.254 0.023 11.130 0.000
-------------------------+-----------------------------------------------
collgrad Diff. |
College gr vs Not colleg | 0.018 0.009 2.002 0.045
Diff. = Not south - Southtotalme collgrad, models(medmod) over(south)Total ME Estimates (N = 2237)
| Estimate Std. err. z P>|z|
-------------------------+-----------------------------------------------
collgrad(Not south) |
College gr vs Not colleg | 0.275 0.024 11.261 0.000
-------------------------+-----------------------------------------------
collgrad(South) |
College gr vs Not colleg | 0.248 0.022 11.398 0.000
-------------------------+-----------------------------------------------
collgrad Diff. |
College gr vs Not colleg | 0.027 0.009 2.892 0.004
Diff. = Not south - SouthDisplay options: ci, level(), decimals(), labwidth(), title()
totalme collgrad age, models(basemod medmod) ci decimals(4) title("Total MEs on occupation group")Total MEs on occupation group (N_basemod = 2237 , N_medmod = 2237)
| Estimate Std. err. z P>|z| 95% LL 95% UL
-------------------------+-----------------------------------------------------------------------
collgrad |
Model 1 (basemod) | 0.2478 0.0214 11.5722 0.0000 0.2058 0.2897
Model 2 (medmod) | 0.2638 0.0229 11.5272 0.0000 0.2189 0.3087
Cross-Model Diff. | -0.0160 0.0078 -2.0490 0.0405 -0.0314 -0.0007
-------------------------+-----------------------------------------------------------------------
age + 1 (centered) |
Model 1 (basemod) | 0.0054 0.0029 1.8307 0.0671 -0.0004 0.0111
Model 2 (medmod) | 0.0065 0.0029 2.2673 0.0234 0.0009 0.0121
Cross-Model Diff. | -0.0011 0.0010 -1.1524 0.2492 -0.0031 0.0008 commands and details
commands prints the command of each model and the margins command (and with two models the suest2 command) that produced the estimates; details prints their output.
totalme collgrad, models(basemod medmod) commandsModel 1 (basemod) is:
mlogit occ3 i.collgrad i.race c.age i.south
Model 2 (medmod) is:
mlogit occ3 i.collgrad i.race c.age i.south c.ttl_exp c.wage
suest2 model is:
suest2 basemod medmod
margins specification is:
margins , at(collgrad=(0 1)) post
Total ME Estimates (N_basemod = 2237 , N_medmod = 2237)
| Estimate Std. err. z P>|z|
-------------------------+-----------------------------------------------
collgrad |
Model 1 (basemod) | 0.248 0.021 11.572 0.000
Model 2 (medmod) | 0.264 0.023 11.527 0.000
Cross-Model Diff. | -0.016 0.008 -2.049 0.040 Weights, svy, and mi
Weights, the svy: prefix, and mi estimate go on the stored models, not on totalme; with two models both must carry the same. Multilevel models need a higher-level weight as well or, better, the svy: prefix. See the help file.
Under mi, one model may be fit with plain mi estimate: or with mi estimate, post:, which return the same pooled statistic; to compare two models, fit both with mi estimate, post:. mi estimate requires the user-written mimrgns package.
Saved results
totalme is r-class. For a continuous or binary variable, r(tmcm1#), r(tmcm2#), and r(tmcd#) hold the Total ME for model 1, model 2, and their difference for the #th such variable in varlist; for a nominal variable r(tmwm1#), r(tmwm2#), r(tmwd#) hold the weighted Total ME inequality and r(tmuwm1#), r(tmuwm2#), r(tmuwd#) the unweighted. With by() or over(), _level is appended, and the Diff. rows append _dlevel1_level2 (e.g. r(tmcm11_d0_1)). r(n_mods) and r(n_vars) hold the number of models and the number of variables. r(table) holds the displayed table and r(se_missing) counts quantities whose standard error could not be computed. The margins results are stored as totalme_margins and, with two models, the combined system as totalme_suest2.
totalme collgrad race, models(basemod medmod)Total ME Estimates (N_basemod = 2237 , N_medmod = 2237)
| Estimate Std. err. z P>|z|
-------------------------+-----------------------------------------------
collgrad |
Model 1 (basemod) | 0.248 0.021 11.572 0.000
Model 2 (medmod) | 0.264 0.023 11.527 0.000
Cross-Model Diff. | -0.016 0.008 -2.049 0.040
-------------------------+-----------------------------------------------
race total ME Ineq. |
Model 1 (basemod) | 0.152 0.044 3.492 0.000
Model 2 (medmod) | 0.147 0.043 3.427 0.001
Cross-Model Diff. | 0.005 0.006 0.902 0.367 return listscalars:
r(se_missing) = 0
r(tmwd1) = .0053498747963924
r(tmwm21) = .1469832421949591
r(tmwm11) = .1523331169913515
r(tmcd1) = -.0160303373827734
r(tmcm21) = .2637991970588639
r(tmcm11) = .2477688596760905
r(n_vars) = 2
r(n_mods) = 2
matrices:
r(table) : 6 x 6Bootstrap standard errors
capture program drop boot_tot
program define boot_tot, rclass
mlogit healthR i.race4 c.age i.woman, base(1)
totalme race4
return scalar w_tot = r(tmwm11)
end
bootstrap w_tot=r(w_tot), reps(1000): boot_tot