Options
Every mecompare option, with an example of each
This page runs each option of mecompare on Stata’s nlsw88 data, in the order of the help file. The Getting started page explains how to read the table.
sysuse nlsw88, clear(NLSW, 1988 extract)drop if missing(union, married, age, race, hours, collgrad, ttl_exp, grade, wage)(371 observations deleted)
. quietly logit union i.married age i.race hours i.collgrad, vce(robust)
estimates store m1. quietly logit union i.married age i.race hours i.collgrad wage grade, vce(robust)
estimates store m2Syntax
mecompare [varlist] [if] [in] [weight] [, models(name1 [name2]) options]With no varlist, marginal effects are computed for every variable in the model(s). Factor-variable notation (i., c.) is required on the stored models.
models() – one model, several, or none
models() names the stored estimates. One name will calculate that model’s marginal effects. Two names will calculate each model’s marginal effects and automatically calculate the cross-model difference. With three or more names the table lists every model’s marginal effects, one set of rows per model, and reports no differences: choose the comparisons you want and test them with metest (e.g. metest 1 - 2), as in tests across three or more models.
If names are omitted, mecompare then uses the estimates in memory, which are either the single model just fit (labeled m1 in the mecompare output) or, after suest2, the models of that system.
mecompare i.married age, models(m1 m2)Predicting: Pr(union)
Marginal effects and cross-model differences (N_m1=1875) (N_m2=1875)
| ME # Estimate Robust SE P>|z|
---------------------------------+---------------------------------------
married |
Married - Single |
m1 | 1 -0.026 0.022 0.222
m2 | 2 -0.019 0.021 0.370
Difference | 3 -0.007 0.003 0.017
---------------------------------+---------------------------------------
age + 1 (centered) |
m1 | 4 0.003 0.003 0.427
m2 | 5 0.003 0.003 0.410
Difference | 6 -0.000 0.000 0.847mecompare i.married age, models(m2)Predicting: Pr(union)
Marginal effects (N_m2=1875)
| ME # Estimate Robust SE P>|z|
---------------------------------+---------------------------------------
married |
Married - Single |
m2 | 1 -0.019 0.021 0.370
---------------------------------+---------------------------------------
age + 1 (centered) |
m2 | 2 0.003 0.003 0.410With no varlist, marginal effects are calculated for every predictor:
mecompare, models(m1 m2)Predicting: Pr(union)
Marginal effects and cross-model differences (N_m1=1875) (N_m2=1875)
| ME # Estimate Robust SE P>|z|
---------------------------------+---------------------------------------
married |
Married - Single |
m1 | 1 -0.026 0.022 0.222
m2 | 2 -0.019 0.021 0.370
Difference | 3 -0.007 0.003 0.017
---------------------------------+---------------------------------------
age + 1 (centered) |
m1 | 4 0.003 0.003 0.427
m2 | 5 0.003 0.003 0.410
Difference | 6 -0.000 0.000 0.847
---------------------------------+---------------------------------------
race |
Black - White |
m1 | 7 0.079 0.024 0.001
m2 | 8 0.093 0.024 0.000
Difference | 9 -0.014 0.004 0.002
Other - White |
m1 | 10 0.101 0.097 0.300
m2 | 11 0.089 0.101 0.377
Difference | 12 0.012 0.015 0.444
---------------------------------+---------------------------------------
hours + 1 (centered) |
m1 | 13 0.002 0.001 0.067
m2 | 14 0.001 0.001 0.185
Difference | 15 0.000 0.000 0.009
---------------------------------+---------------------------------------
collgrad |
College gra - Not colleg |
m1 | 16 0.102 0.024 0.000
m2 | 17 0.046 0.039 0.234
Difference | 18 0.056 0.032 0.085
---------------------------------+---------------------------------------
wage + 1 (centered) |
m1 |
m2 | 19 0.013 0.002 0.000
Difference |
---------------------------------+---------------------------------------
grade + 1 (centered) |
m1 |
m2 | 20 0.001 0.007 0.830
Difference | Supported models
mecompare accepts one or more models from the following families (see also the help file).
Ordinary single-level models
regress; logit and logistic; probit; poisson; nbreg; ologit; oprobit; and mlogit.
glm; cloglog; tobit; intreg; maximum-likelihood heckman; and parametric streg, all parametric distributions.
gologit2, all forms.
ivregress 2sls. fracreg with estimators logit and probit.
betareg, all four links; truncreg; hetprobit; zip and zinb, both inflation links; and biprobit.
ivprobit and ivtobit.
Panel models
xtreg with estimators mle, fe, be, re, and pa. xtreg, cre is not supported; the correlated random effects (Mundlak) specification is fit with the re estimator instead, as described in the help file.
xtlogit with estimators re, fe, and pa. xtprobit with estimators re and pa. xtcloglog with estimators re and pa; the re estimator requires intpoints(24) or more.
xtologit and xtoprobit.
xtmlogit with estimators re and fe.
xtpoisson with estimators re (normal or gamma random effects), fe, and pa. xtnbreg with estimators re and pa; xtnbreg, fe is not supported.
Marginal effects after the fixed-effects (fe) estimators have known problems, especially for categorical outcomes; the help file’s section on correlated random effects gives the alternative, and the within-person effects example shows it.
Multilevel models
mixed, mle; melogit; meprobit; mecloglog; mepoisson; menbreg; meologit; meoprobit; and mestreg, all parametric distributions.
meglm with the family-link pairs Gaussian-identity and Gamma-log.
After streg and mestreg the marginal effects are changes in a predicted survival time – the median after streg, the mean after mestreg – and a note under the table says which; see Survival models in the help file.
Which models can be compared
Models can be compared when they all return the same number of predictions: e.g., two logits, a logit and a regress, a logit and a melogit, or an ologit and an mlogit with the same outcome categories. Each model uses its own default prediction, or the one given in predict(), so a logit vs regress comparison contrasts predicted probabilities with linear predictions.
Survey, multiply imputed, and weighted models
Models may be fit with the svy:, mi estimate, post:, or mi estimate: svy: prefixes, specified on the stored models rather than on mecompare; all models in a comparison must carry the same prefixes, and mi estimate requires the user-written mimrgns package. Weights can instead be applied on the stored models (logit y x [pw=w], the same weight on every model), and multilevel models need a stage weight as well. The help file has the details.
statistics() – which columns to show
The default columns are the estimate, its standard error, and the p-value. statistics() takes any of estimate, se, pvalue, z, ll, ul, or all.
mecompare i.married age, models(m1 m2) statistics(est se ll ul)Predicting: Pr(union)
Marginal effects and cross-model differences (N_m1=1875) (N_m2=1875)
| ME # Estimate Robust SE CI LL CI UL
---------------------------------+--------------------------------------------------
married |
Married - Single |
m1 | 1 -0.026 0.022 -0.069 0.016
m2 | 2 -0.019 0.021 -0.061 0.023
Difference | 3 -0.007 0.003 -0.013 -0.001
---------------------------------+--------------------------------------------------
age + 1 (centered) |
m1 | 4 0.003 0.003 -0.004 0.009
m2 | 5 0.003 0.003 -0.004 0.009
Difference | 6 -0.000 0.000 -0.001 0.001amount(), centered, uncentered – the size of a continuous change
For a continuous variable the default is a one-unit change, centered on the observed value (half a unit below to half a unit above). 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, dydx in margins). If a single value is listed in amount(), it is applied to all continuous variables. When multiple values are listed in amount(), they are applied in the order of the continuous variables in the varlist.
mecompare age hours, models(m1 m2) amount(sd)Predicting: Pr(union)
Marginal effects and cross-model differences (N_m1=1875) (N_m2=1875)
| ME # Estimate Robust SE P>|z|
---------------------------------+---------------------------------------
age + SD (centered) |
m1 | 1 0.008 0.010 0.427
m2 | 2 0.008 0.010 0.410
Difference | 3 -0.000 0.001 0.847
---------------------------------+---------------------------------------
hours + SD (centered) |
m1 | 4 0.018 0.010 0.067
m2 | 5 0.014 0.010 0.185
Difference | 6 0.005 0.002 0.009mecompare age hours, models(m1 m2) amount(10 sd)Predicting: Pr(union)
Marginal effects and cross-model differences (N_m1=1875) (N_m2=1875)
| ME # Estimate Robust SE P>|z|
---------------------------------+---------------------------------------
age + 10 (centered) |
m1 | 1 0.026 0.032 0.427
m2 | 2 0.027 0.032 0.410
Difference | 3 -0.001 0.005 0.847
---------------------------------+---------------------------------------
hours + SD (centered) |
m1 | 4 0.018 0.010 0.067
m2 | 5 0.014 0.010 0.185
Difference | 6 0.005 0.002 0.009mecompare age, models(m1 m2) amount(rate)Predicting: Pr(union)
Marginal effects and cross-model differences (N_m1=1875) (N_m2=1875)
| ME # Estimate Robust SE P>|z|
---------------------------------+---------------------------------------
age(rate) |
m1 | 1 0.003 0.003 0.427
m2 | 2 0.003 0.003 0.410
Difference | 3 -0.000 0.000 0.847mecompare age, models(m1 m2) amount(p10-p90)Predicting: Pr(union)
Marginal effects and cross-model differences (N_m1=1875) (N_m2=1875)
| ME # Estimate Robust SE P>|z|
---------------------------------+---------------------------------------
age(10-90%) |
m1 | 1 0.023 0.029 0.428
m2 | 2 0.024 0.029 0.411
Difference | 3 -0.001 0.004 0.847Multiple amounts for one variable can be specified in parentheses, with each effect reported in its own rows. E.g., amount((one sd rate)) would apply all three to every continuous variable. Whereas amount((one sd) 5) gives the first continuous variable in varlist two amounts (one and sd) and the second continuous variable one amount (5). A list in parentheses cannot be combined with a value list in start() or with end().
mecompare age, models(m1) amount((one sd rate))Predicting: Pr(union)
Marginal effects (N_m1=1875)
| ME # Estimate Robust SE P>|z|
---------------------------------+---------------------------------------
age + 1 (centered) |
m1 | 1 0.003 0.003 0.427
---------------------------------+---------------------------------------
age + SD (centered) |
m1 | 2 0.008 0.010 0.427
---------------------------------+---------------------------------------
age(rate) |
m1 | 3 0.003 0.003 0.427mecompare age hours, models(m1) amount((one sd) 5)Predicting: Pr(union)
Marginal effects (N_m1=1875)
| ME # Estimate Robust SE P>|z|
---------------------------------+---------------------------------------
age + 1 (centered) |
m1 | 1 0.003 0.003 0.427
---------------------------------+---------------------------------------
age + SD (centered) |
m1 | 2 0.008 0.010 0.427
---------------------------------+---------------------------------------
hours + 5 (centered) |
m1 | 3 0.009 0.005 0.067uncentered makes the change start at the observed value and end at that value plus the amount() (from x to x + amount) rather than being centered on it. The two conventions give different numbers whenever the effect is nonlinear. Especially for larger changes like sd, we recommend using the default of centered changes.
mecompare age, models(m1 m2) amount(sd) uncenteredPredicting: Pr(union)
Marginal effects and cross-model differences (N_m1=1875) (N_m2=1875)
| ME # Estimate Robust SE P>|z|
---------------------------------+---------------------------------------
age + SD (uncentered) |
m1 | 1 0.008 0.010 0.432
m2 | 2 0.008 0.010 0.415
Difference | 3 -0.000 0.001 0.847start() and end() – where the change begins and ends
By default each observation’s own value is the starting point. start() sets it: a value, atmeans, or a list of values for one variable, which gives one numbered row set per value so that metest can compare them.
mecompare age, models(m1 m2) start(age=40) amount(5)Predicting: Pr(union)
Marginal effects and cross-model differences (N_m1=1875) (N_m2=1875)
| ME # Estimate Robust SE P>|z|
---------------------------------+---------------------------------------
age + 5 (centered) |
m1 | 1 0.013 0.016 0.430
m2 | 2 0.013 0.016 0.412
Difference | 3 -0.000 0.002 0.847mecompare age, models(m1) start(age=(35 40 45)) amount(5)Predicting: Pr(union)
Marginal effects (N_m1=1875)
| ME # Estimate Robust SE P>|z|
---------------------------------+---------------------------------------
age + 5 (centered) |
at 35 | 1 0.012 0.015 0.413
at 40 | 2 0.013 0.016 0.430
at 45 | 3 0.013 0.017 0.444end() sets where the change ends, with start(): start(age=35) end(age=40) is the change from 35 to 40 (amount() and centering do not apply to that variable). With a list in both, the values pair up: start(age=(35 40)) end(age=(40 45)) gives one row for 35 to 40 and one for 40 to 45.
mecompare age, models(m1) start(age=35) end(age=40)Predicting: Pr(union)
Marginal effects (N_m1=1875)
| ME # Estimate Robust SE P>|z|
---------------------------------+---------------------------------------
age 35 to 40 |
m1 | 1 0.013 0.016 0.422mecompare age, models(m1) start(age=(35 40)) end(age=(40 45))Predicting: Pr(union)
Marginal effects (N_m1=1875)
| ME # Estimate Robust SE P>|z|
---------------------------------+---------------------------------------
age |
35 to 40 | 1 0.013 0.016 0.422
40 to 45 | 2 0.013 0.017 0.437cochange() – variables that change along with the focal variable
cochange(varlist) specifies variables that change along with the focal variable instead of being held constant, e.g. mediators. The table then reports two marginal effects of the focal variable: the usual one, and one labeled with co-change in which the co-change variables change too. metest can be used to test the difference between the two. Only one focal variable may be given in the varlist.
start() and end() set the values the co-change variables change from and to. Here, the effect of a college degree, and the effect of a college degree when the hourly wage also changes from 6 to 9 dollars:
quietly logit union age hours wage ttl_exp i.race i.collgrad i.married, vce(robust)
estimates store m3. mecompare i.collgrad, models(m3) cochange(wage) start(wage=6) end(wage=9)Predicting: Pr(union)
Marginal effects (N_m3=1875)
| ME # Estimate Robust SE P>|z|
---------------------------------+---------------------------------------
collgrad |
College gra - Not colleg |
m3 | 1 0.052 0.025 0.040
---------------------------------+---------------------------------------
collgrad with co-change |
m3 | 2 0.093 0.024 0.000
NOTE: collgrad with co-change: collgrad 0 to 1; wage 6 to 9.metest 2 - 1 | estimate se pvalue
---------------------------------+--------------------------------
cochange - collgrad | 0.041 0.008 0.000 Without start() and end(), co-change variables change as focal variables do (see amount()). Here, age and total work experience both increase by five years:
mecompare age, models(m3) cochange(ttl_exp) amount(5)Predicting: Pr(union)
Marginal effects (N_m3=1875)
| ME # Estimate Robust SE P>|z|
---------------------------------+---------------------------------------
age + 5 (centered) |
m3 | 1 0.014 0.016 0.376
---------------------------------+---------------------------------------
age with co-change |
m3 | 2 0.008 0.019 0.678
NOTE: age with co-change: age + 5 (centered); ttl_exp + 5 (centered).A nominal focal or co-change variable with three or more categories needs the two levels it changes between in start() and end(), e.g. start(race=1) end(race=2). cochange() cannot be combined with a numlist of values in start() or end(), multiple amounts for one variable in amount(), amount(rate), groupsd, pwcompare, or mcompare(). The generalized marginal effects page has a longer example.
covariates() and atmeans – values of the other variables
Covariates are the model’s other independent variables, not included in the varlist. By default they stay at their observed values and the effect is averaged over the sample. covariates() fixes them: covariates(atmeans) (or just atmeans) at their means, covariates(married=1) at a value, and a list for one covariate – covariates(hours=(20 40 60)) – gives one row set per value, each a counterfactual with the whole sample set to that value.
mecompare i.married, models(m1 m2) atmeansPredicting: Pr(union)
Marginal effects and cross-model differences (N_m1=1875) (N_m2=1875)
| ME # Estimate Robust SE P>|z|
---------------------------------+---------------------------------------
married |
Married - Single |
m1 | 1 -0.027 0.022 0.222
m2 | 2 -0.019 0.022 0.370
Difference | 3 -0.007 0.003 0.019mecompare i.married, models(m2) covariates(hours=(20 40 60))Predicting: Pr(union)
Marginal effects (N_m2=1875)
| ME # Estimate Robust SE P>|z|
---------------------------------+---------------------------------------
married |
Married - Single |
hours=20 | 1 -0.018 0.020 0.375
hours=40 | 2 -0.019 0.021 0.369
hours=60 | 3 -0.021 0.023 0.366metest 1 = 2 = 3Tests of equality
| chi2 df pvalue
---------------------------------+--------------------------------
marri~20 = marri~40 = marri~60 | 0.896 2.000 0.639 by() and over() – effects within levels of another variable
by(varlist) reports the marginal effect at each level of a binary or nominal moderator, computed as a counterfactual (everyone set to that level, then averaged); over(varlist) computes it within each observed subgroup instead. Both can take more than one variable. For a continuous moderator, covariates(z=(numlist)) plays the same role as by(). The interactions pages show how to use these to simplify tests of interaction.
mecompare age, models(m1 m2) by(collgrad)Predicting: Pr(union)
Marginal effects and cross-model differences (N_m1=1875) (N_m2=1875)
| ME # Estimate Robust SE P>|z|
---------------------------------+---------------------------------------
age + 1 (centered) |
m1 Not college grad | 1 0.002 0.003 0.427
m1 College grad | 2 0.003 0.004 0.428
m2 Not college grad | 3 0.003 0.003 0.410
m2 College grad | 4 0.003 0.004 0.411
Difference Not college grad | 5 -0.000 0.000 0.705
Difference College grad | 6 0.000 0.001 0.792mecompare age, models(m1 m2) over(collgrad)Predicting: Pr(union)
Marginal effects and cross-model differences (N_m1=1875) (N_m2=1875)
| ME # Estimate Robust SE P>|z|
---------------------------------+---------------------------------------
age + 1 (centered) |
m1 Not college grad | 1 0.002 0.003 0.427
m1 College grad | 2 0.003 0.004 0.428
m2 Not college grad | 3 0.003 0.003 0.410
m2 College grad | 4 0.003 0.004 0.410
Difference Not college grad | 5 -0.000 0.000 0.827
Difference College grad | 6 -0.000 0.001 0.896pwcompare – all pairwise contrasts of a nominal variable
For an i. variable the default rows contrast each level with the base level. pwcompare reports every pair.
mecompare i.race, models(m1 m2) pwcomparePredicting: Pr(union)
Marginal effects and cross-model differences (N_m1=1875) (N_m2=1875)
| ME # Estimate Robust SE P>|z|
---------------------------------+---------------------------------------
race |
Black - White |
m1 | 1 0.079 0.024 0.001
m2 | 2 0.093 0.024 0.000
Difference | 3 -0.014 0.004 0.002
Other - White |
m1 | 4 0.101 0.097 0.300
m2 | 5 0.089 0.101 0.377
Difference | 6 0.012 0.015 0.444
Other - Black |
m1 | 7 0.021 0.099 0.829
m2 | 8 -0.004 0.103 0.970
Difference | 9 0.025 0.015 0.098mcompare() – adjusting for multiple comparisons
mcompare(method) adjusts the p-values (and confidence intervals, when shown) of the contrasts of nominal focal variables for multiple comparisons; method is bonferroni or sidak. A variable’s contrasts within one model form a set, and its Difference rows form a set of their own. The estimates, e(b) and e(V) do not change.
mecompare i.race, models(m1 m2) pwcompare mcompare(bonferroni)Predicting: Pr(union)
Marginal effects and cross-model differences (N_m1=1875) (N_m2=1875)
| ME # Estimate Robust SE P>|z|
---------------------------------+---------------------------------------
race |
Black - White |
m1 | 1 0.079 0.024 0.003
m2 | 2 0.093 0.024 0.000
Difference | 3 -0.014 0.004 0.007
Other - White |
m1 | 4 0.101 0.097 0.900
m2 | 5 0.089 0.101 1.000
Difference | 6 0.012 0.015 1.000
Other - Black |
m1 | 7 0.021 0.099 1.000
m2 | 8 -0.004 0.103 1.000
Difference | 9 0.025 0.015 0.293
NOTE: p-values (and CIs) of the contrasts are Bonferroni-adjusted within each variable and model; th
> e Difference rows are adjusted as a set of their own.meinequality and totalme – summary measures
meinequality adds the marginal-effect inequality summary of meinequality above a nominal variable’s contrasts (weighted, the default, unweighted, or all). totalme adds the Total Marginal Effect of totalme above every focal variable, most often used to summarize effects with multi-category outcome models like an ordinal or nominal model.
mecompare i.race, models(m1 m2) meinequality(all)Predicting: Pr(union)
Marginal effects and cross-model differences (N_m1=1875) (N_m2=1875)
| ME # Estimate Robust SE P>|z|
---------------------------------+---------------------------------------
race |
ME Inequality |
m1 | 1 0.079 0.050 0.116
m2 | 2 0.079 0.028 0.005
Difference | 3 -0.000 0.027 0.999
Unwgt ME Ineq. |
m1 | 4 0.067 0.065 0.300
m2 | 5 0.062 0.016 0.000
Difference | 6 0.005 0.066 0.938
Black - White |
m1 | 7 0.079 0.024 0.001
m2 | 8 0.093 0.024 0.000
Difference | 9 -0.014 0.004 0.002
Other - White |
m1 | 10 0.101 0.097 0.300
m2 | 11 0.089 0.101 0.377
Difference | 12 0.012 0.015 0.444egen hours_ord = cut(hours), at(0 30 40 50 81)
quietly ologit hours_ord c.age i.married i.race, vce(robust)
mecompare age married race, totalmePredicting: Pr(hours_ord)
Marginal effects (N_m1=1875)
| ME # Estimate Robust SE P>|z|
---------------------------------+---------------------------------------
age + 1 (centered) |
Total ME | 1 0.006 0.003 0.053
Outcome 0 - m1 | 2 0.003 0.002 0.055
Outcome 30 - m1 | 3 0.003 0.001 0.053
Outcome 40 - m1 | 4 -0.004 0.002 0.055
Outcome 50 - m1 | 5 -0.002 0.001 0.054
---------------------------------+---------------------------------------
married |
Married - Single |
Total ME | 6 0.121 0.019 0.000
Outcome 0 - m1 | 7 0.066 0.011 0.000
Outcome 30 - m1 | 8 0.055 0.009 0.000
Outcome 40 - m1 | 9 -0.073 0.012 0.000
Outcome 50 - m1 | 10 -0.048 0.009 0.000
---------------------------------+---------------------------------------
race |
Total ME Ineq. | 11 0.049 0.045 0.276
Black - White |
Outcome 0 - m1 | 12 -0.012 0.011 0.263
Outcome 30 - m1 | 13 -0.010 0.009 0.267
Outcome 40 - m1 | 14 0.014 0.012 0.264
Outcome 50 - m1 | 15 0.008 0.007 0.268
Other - White |
Outcome 0 - m1 | 16 -0.044 0.043 0.315
Outcome 30 - m1 | 17 -0.038 0.043 0.381
Outcome 40 - m1 | 18 0.045 0.038 0.238
Outcome 50 - m1 | 19 0.036 0.048 0.454groups, groupnames(), groupme, groupsd – models fit to separate samples
The groups option tells mecompare that the models were fit to non-overlapping samples, so each model’s marginal effects are averaged over its own sample and the Difference row compares the effects across the groups (with three or more groups each group’s effects are listed and metest can be used to test the differences). groupnames() labels the rows; groupme (two groups only) adds the average difference in the outcome between the groups; groupsd uses each group’s own SD for amount(sd); we generally do not recommend groupsd. 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 logit union age hours i.collgrad if south == 1, vce(robust)
estimates store msouth
quietly logit union age hours i.collgrad if south == 0, vce(robust)
estimates store mnonsouth. mecompare age hours collgrad, models(msouth mnonsouth) groups groupnames(South NonSouth) groupmePredicting: Pr(union)
Marginal effects and cross-model differences (N_South=797) (N_NonSouth=1078)
| ME # Estimate Robust SE P>|z|
---------------------------------+---------------------------------------
age + 1 (centered) |
South | 1 -0.005 0.004 0.273
NonSouth | 2 0.007 0.005 0.137
Difference | 3 -0.011 0.006 0.066
---------------------------------+---------------------------------------
hours + 1 (centered) |
South | 4 0.002 0.001 0.092
NonSouth | 5 0.003 0.001 0.019
Difference | 6 -0.001 0.002 0.744
---------------------------------+---------------------------------------
collgrad |
College gra - Not colleg |
South | 7 0.026 0.032 0.407
NonSouth | 8 0.135 0.033 0.000
Difference | 9 -0.109 0.046 0.018
Average conditional difference in the outcome (NonSouth - South),
each model averaged over its own sample:
Estimate = 0.130 SE = 0.019 p = 0.000predict() – which prediction
With one model any prediction that margins accepts after that command can be requested; with two or more models, any prediction returning one quantity per model. Multi-category models report every outcome by default, or one with predict(pr outcome(#)).
quietly nbreg hours c.age i.collgrad i.race, vce(robust)
estimates store cnt. mecompare age collgrad, models(cnt)Predicting: Predicted number of events
Marginal effects (N_cnt=1875)
| ME # Estimate Robust SE P>|z|
---------------------------------+---------------------------------------
age + 1 (centered) |
cnt | 1 -0.046 0.077 0.545
---------------------------------+---------------------------------------
collgrad |
College gra - Not colleg |
cnt | 2 2.495 0.569 0.000mecompare age collgrad, models(cnt) predict(pr(0))Predicting: Pr(hours=0)
Marginal effects (N_cnt=1875)
| ME # Estimate Robust SE P>|z|
---------------------------------+---------------------------------------
age + 1 (centered) |
cnt | 1 0.000 0.000 0.569
---------------------------------+---------------------------------------
collgrad |
College gra - Not colleg |
cnt | 2 -0.000 0.000 0.232marginsopt() – anything else margins should be told
marginsopt(string) is passed to margins verbatim. We recommend using it only in combination with the commands and details options to double check that margins produced what you want. Its main use is expression() with one model, for example effects in percentage points:
mecompare i.married age, models(m2) marginsopt(expression(100*predict(pr)))Predicting: 100*predict(pr)
Marginal effects (N_m2=1875)
| ME # Estimate Robust SE P>|z|
---------------------------------+---------------------------------------
married |
Married - Single |
m2 | 1 -1.906 2.126 0.370
---------------------------------+---------------------------------------
age + 1 (centered) |
m2 | 2 0.266 0.322 0.410Display options
decimals(#) sets the decimal places (default 3), mod1name() and mod2name() label the model rows, labwidth(#) and statwidth(#) set the column widths, and norownum drops the # column.
mecompare i.married age, models(m1 m2) decimals(4) mod1name(Base) mod2name(Full) norownumPredicting: Pr(union)
Marginal effects and cross-model differences (N_Base=1875) (N_Full=1875)
| Estimate Robust SE P>|z|
---------------------------------+--------------------------------
married |
Married - Single |
Base | -0.0263 0.0215 0.2215
Full | -0.0191 0.0213 0.3700
Difference | -0.0073 0.0031 0.0175
---------------------------------+--------------------------------
age + 1 (centered) |
Base | 0.0026 0.0032 0.4274
Full | 0.0027 0.0032 0.4097
Difference | -0.0001 0.0005 0.8467commands and details – seeing what was run
commands prints the command of each model and the suest2 and margins commands mecompare used; details prints their output too.
mecompare i.married, models(m1 m2) commandsModel 1 (m1) is:
logit union i.married age i.race hours i.collgrad
Model 2 (m2) is:
logit union i.married age i.race hours i.collgrad wage grade
suest2 model is:
suest2 m1 m2, nowarn
margins command is:
margins , at( married=(0 1)) post
Predicting: Pr(union)
Marginal effects and cross-model differences (N_m1=1875) (N_m2=1875)
| ME # Estimate Robust SE P>|z|
---------------------------------+---------------------------------------
married |
Married - Single |
m1 | 1 -0.026 0.022 0.222
m2 | 2 -0.019 0.021 0.370
Difference | 3 -0.007 0.003 0.017store() – results for coefplot and esttab
store(stub) saves the marginal effects as stored estimates (stub_model1, stub_model2, stub_diff), ready for coefplot and esttab. The coefficients are named with the table’s labels, e.g. age + SD (centered) or Black - White. The plotting page shows how.
After mecompare: metest, e(b), e(V)
Every row of the table is posted to e(b) and e(V) in # order, so metest, lincom, nlcom, and test all work on them. mecompare, coeflegend lists the coefficient names.
mecompare i.married age, models(m1 m2)Predicting: Pr(union)
Marginal effects and cross-model differences (N_m1=1875) (N_m2=1875)
| ME # Estimate Robust SE P>|z|
---------------------------------+---------------------------------------
married |
Married - Single |
m1 | 1 -0.026 0.022 0.222
m2 | 2 -0.019 0.021 0.370
Difference | 3 -0.007 0.003 0.017
---------------------------------+---------------------------------------
age + 1 (centered) |
m1 | 4 0.003 0.003 0.427
m2 | 5 0.003 0.003 0.410
Difference | 6 -0.000 0.000 0.847metest 1 - 4 | estimate se pvalue
---------------------------------+--------------------------------
m1 |
married - age | -0.029 0.022 0.182 mecompare, coeflegend------------------------------------------------------------------------------
| Coefficient Legend
-------------+----------------------------------------------------------------
married |
m1 | -0.026 _b[married:m1]
m2 | -0.019 _b[married:m2]
Difference | -0.007 _b[married:Difference]
-------------+----------------------------------------------------------------
age |
m1 | 0.003 _b[age:m1]
m2 | 0.003 _b[age:m2]
Difference | -0.000 _b[age:Difference]
------------------------------------------------------------------------------