{smcl}
{* 20 August 2009}{...}
help for {hi:margeff} {right:Version 2.2.0 (20 August 2009)}
{hline}
{title:Obtain partial effects after estimation}
{p 4 12 2}
{cmd:margeff} [{cmdab:c:ompute}]
[{cmd:if} {it:exp}]
[{cmd:in} {it:range}]
[{cmd:,}
{p_end}
{p 16 16 2}
{cmd:at(}{it:atlist}{cmd:)}
{cmdab:cons:tant}
{cmdab:d:ummies(}{it:varlist_1} [{cmd:\} {it:varlist_2} ..]{cmd:)}
{p_end}
{p 16 16 2}
{cmdab:l:ink(}{it:log}|{it:loglog}|{it:logc}|{it:nbinomial}|{it:power #}|{it:opower #}{cmd:)}
{cmdab:m:odel(}{it:stata_cmd}{cmd:)}
{cmdab:nooff:set}
{cmdab:now:ght}
{cmdab:out:come(}{it:#}{cmd:)}
{cmdab:p:ercent}
{cmdab:r:eplace} ]
{p_end}
{p 4 12 2}{cmd:margeff} {cmdab:r:eplay} [{cmd:,} {cmdab:l:evel(}{it:#}{cmd:)} ]
{p 4 4 2} where {it:stata_cmd} is one of
{col 9}[group 1:] {col 20}{help probit}, {help logit}, {help logistic}, {help cloglog}, {help heckprob}
{col 9}[group 2:] {col 20}{help oprobit}, {help ologit}, {help mlogit}, {help biprobit}
{col 9}[group 3:] {col 20}{help poisson}, {help nbreg}, {help zip}, {help zinb}
{col 9}[group 4:] {col 20}{help truncreg}, {help cnreg}, {help tobit}, {help heckman}
{p 4 4 2} and {it:atlist} is {p_end}
{p 8 8 2}{c -(} {cmd:mean} | {cmd:median} | {cmd:zero} [ {it:varname}
{cmd:=} {it:#} [{cmd:,} {it:varname} {cmd:=} {it:#}] [{it:...}]] {c )-}{p_end}
{p 8 8 2}{c -(} [{cmd:mean}] |[{cmd:median}] | [{cmd:zero}] {it:varname}
{cmd:=} {it:#} [{cmd:,} {it:varname} {cmd:=} {it:#}] [{it:...}] {c )-} {p_end}
{p 4 4 2} and {it:varlist_1} [{cmd:\} {it:varlist_2} ... ] are lists of dummy variables, where all dummies of a list indicate
different categories of the same underlying categorical variable.
{title:Description}
{p 4 4 2}
{cmd:margeff compute} analytically estimates partial effects after estimation.
Standard errors of partial effects are also estimated using the delta method.
Partial effects can be obtained not only after the above listed models, but also after
{p 8 8 2}
1. generalized linear models (see help for {help glm}),
{break}
2. the panel-data ({cmd:xt}) version of the supported models inlcuding {cmd:xtgee},
{break}
3. generalized ordered models (see help for {help gologit2} if installed).
{p_end}
{p 4 4 2}
The default behavior of {cmd:margeff} and main alternatives to the defaults can be summarized as follows:
{p_end}
{p 4 8 2}{cmd:1.}
By default, {cmd:margeff} calculates average partial effects.
Estimation of partial effects evaluated at sample means or at any other evaluation points is also possible with the help of the {cmd:at(}{it:atlist}{cmd:)} option.
In either case, calculations are restricted to the estimation sample.
{p_end}
{p 4 8 2}{cmd:2. }
Partial effects are discrete partial changes in the quantities of interest as the variable under study increases
by the unit of measurement.
For dummy variables (variables coded 0/1) and count variables, unit of measurement is 1.
For other variables, units of measurement are detected using an algorithm implemented in the official ado file {cmd:codebook}.
This partial change approach eases computations, and approximates very well the partial changes calculated using the classic
marginal effects approach.
{p_end}
{p 4 8 2}{cmd:3.}
Quantities of interest are defined as follows:
{p 8 19 4}[group 1:] probability of positive outcome {p_end}
{p 8 19 4}[group 2:] probabilities of all possible outcomes defined by the dependent variable(s){p_end}
{p 8 19 4}[group 3:] expected number of counts or the incidence rate {p_end}
{p 8 19 4}[group 4:] expected value of dependent variable conditional on the dependent variable being observed{p_end}
{p 4 8 2}{cmd:4.}
The calculations automatically adjust for {cmd:fweight}s, {cmd:iweight}s, or {cmd:pweight}s used during estimation.
If you wish to apply (different) weights, however, you can specify your own {cmd:fweight}s, {cmd:iweight}s, or {cmd:pweight}s; see help {help weights}.
{p 4 8 2}{cmd:5.}
{cmd:margeff} behaves as a post-estimation command (see help {help postest}).
However, option {cmd:replace} forces {cmd:margeff} to behave as an estimation command
(see help {help est}).
This enables the use of post-estimation commands like {cmd:lincom} or {cmd:test} after {cmd:margeff}.
{p 4 4 2}
Typed without arguments, {cmd:margeff} replays the results of the previous {cmd:margeff}
computations, provided the last use of {cmd:margeff} was combined with the {cmd:replace}
option.{p_end}
{p 4 4 2}
{cmd:margeff replay} replays the results of the previous {cmd:margeff} computation.
{title:Options}
{p 4 8 2}
{cmd:at(}{it:atlist}{cmd:)} forces {cmd:margeff} to estimate partial effects at points specified in
{it:atlist}, instead of estimating average partial effects.
{p_end}
{p 8 8 2}
{cmd:at(} {cmd:mean} | {cmd:median} | {cmd:zero} [ {it:varname} {cmd:=} {it:#} [{cmd:,} {it:varname} {cmd:=} {it:#} [{it:...}]] {cmd:)}
specifies that the partial effects be evaluated at means, at medians of the independent variables, or at zeros.
It also allows users to specify particular values for one or more independent
variables, assuming that the rest are means, medians, or zeros.
{p_end}
{p 8 8 2}
{cmd:at(} [{cmd:mean}] |[{cmd:median}] | [{cmd:zero}] {it:varname} {cmd:=} {it:#} [{cmd:,} {it:varname} {cmd:=} {it:#}] [{it:...}] {cmd:)}
specifies that the partial effects be evaluated at particular values for one or more independent
variables, assuming that the rest are means.
{p_end}
{p 4 8 2}
{cmd:constant} has the effect that quantities of interests computed at zero values of the independent variables are also displayed.
Thus the constant term shows the baseline prediction.
{p_end}
{p 4 8 2}
{cmd:dummies(}{it:varlist_1} [{cmd:\} {it:varlist_2} ... ]{cmd:)}
modifies the calculation of partial effects if sets of indicator or dummy variables were included in the model.
The modification is the following: if the dummy variable {it:Dvar1} appears only in {it:varlist_1} then all variables appearing
in {it:varlist_1} are set to zero, but variables appearing in {it:varlist_2} [{cmd:\} {it:varlist_3} ... ] are not affected.
This option should be specified so that each {it:varlist_k} ({it:k) = 1,2,...) consists of indicator or dummy variables referring to different categories of a single underlying variable.
{p_end}
{p 4 8 2} {cmd:level(}{it:#}{cmd:)} specifies the confidence level in percent for the confidence intervals of the coefficients; see help
{help level}.
{p 4 8 2}
{cmd:link(}{it:log}|{it:loglog}|{it:logc}|{it:nbinomial}|{it:power #}|{it:opower #}{cmd:)} forces {cmd:margeff} to define the
quantity of interest as {it:F(xb)}, where {it:xb} is the linear prediction and {it:F} is the inverse of the link function.
For example, the option {cmd:link(}{it:log}{cmd:)} defines the quantity of interest as {it:exp(xb)}.
{p_end}
{p 4 8 2}
{cmd:model(}{it:stata_cmd}{cmd:)} forces {cmd:margeff} to estimate partial effects as if the preceeding estimation command were {it:stata_cmd}.
{it:stata_cmd} must be one of the supported commands that are listed above.
This option is likely to be helpful if you wish to obtain partial effects
after an estimation command having the same link function as {it:stata_cmd}.
It is the user's responsibility to ensure that the link functions are the same.
{p_end}
{p 4 8 2}
{cmd:nooffset} causes {cmd:margeff} to ignore the offset variable during the calculations.
{p_end}
{p 4 8 2}
{cmd:nowght} causes {cmd:margeff} to ignore weights used during previous estimation.
{p_end}
{p 4 8 2}
{cmd:outcome(}{it:#}{cmd:)} causes {cmd:margeff} to display only one outcome.
This option is useful after estimation commands listed as [group 2] models.
The number specified is interpreted as follows:
{p_end}
{p 8 12 2}
{cmd:1.} After {cmd:biprobit}, numbers 1 2 3 and 4 refer to outcomes p00 p01 p10 p11.
{p_end}
{p 8 12 2}
{cmd:2.} After {cmd:ologit}, {cmd:oprobit}, {cmd:gologit2} and {cmd:mlogit}, number # refers to the #th category of the dependent variable.
Thus, number 1 always indicates the lowest category. And if the dependent variable has, say, 5 categories, then
number 5 indicates the highest category.
{p_end}
{p 4 8 2}
{cmd:percent} causes {cmd:margeff} to display the results in a percentage form.
{p_end}
{p 4 8 2}
{cmd:replace} causes {cmd:margeff} to overwrite the estimation results left behind.
This option is useful if
{p_end}
{p 8 12 2} {cmd:1.} you wish to include partial effects in publication-quality tables using either the official {help estimates table} command or
the user-written commands {help outreg} or {help estout}; or
{p_end}
{p 8 12 2} {cmd:2.} your model contains a variable (such as age-squared) which is a mathematical transformation of another independent variable,
and you wish to obtain the total effect of that variable (age) using the {help lincom} or the {help nlcom} command.
{p_end}
{title:Remarks}
{p 4 4 2}
{cmd:margeff} is a work-in-progress; comments, suggestions, bug reports are welcome!
Please direct correspondence to the adress described at the end of the help file. {p_end}
{p 4 4 2}
To keep {cmd:margeff} up-to-date, visit the website{p_end}
{p 8 4 2}{browse "http://www.uni-corvinus.hu/bartus"} {p_end}
{p 4 4 2}or type{p_end}
{p 8 4 2} {cmd: net from "http://web.uni-corvinus.hu/bartus/stata"} {p_end}
{p 8 4 2} {cmd: net install margeff , replace }{p_end}
{title:Examples}
{p 4 4 2}{cmd:Illustrating the importance of the dummies( varlist_1 \ ... ) option }
{p 4 4 2}Type the following commands:{p_end}
{p 4 8 2}{cmd:. [save mydata, replace]}{p_end}
{p 4 8 2}{cmd:. tabi 60 30 10 \ 20 60 20 \ 10 10 80 , replace }{p_end}
{p 4 8 2}{cmd:. xi: mlogit col i.row [fw=pop] }{p_end}
{p 4 8 2}{cmd:. margeff }{p_end}
{p 4 8 2}{cmd:. margeff , at(mean) }{p_end}
{p 4 8 2}{cmd:. margeff , at(mean) dummies(_I*) }{p_end}
{p 4 4 2}You can see that the first and the last {margeff} commands produced the expected results.
{p_end}
{p 4 8 2}{cmd:Easy calculation of the total effect of age }
{p 4 4 2} Suppose you wish to model employment status as a function of the usual human capital variables: gender, years of education, experience,
and the square of experience. Assume further that the sample average of experience is 20. The total effect of experience on employment probabilities can be estimated as follows:
{p 4 8 2}{cmd:. logit employed gender edu exp exp2 }{p_end}
{p 4 8 2}{cmd:. margeff, at(mean) replace } {p_end}
{p 4 8 2}{cmd:. lincom exp+40*exp2 }{p_end}
{title:Acknowledgements}
{p 4 4 2}
Some parts of the code are taken from the official ado-files codebook.
Earlier versions of margeff relied on margfx (version 30 Jul 1999 for Stata 5)
written by Jonah B. Gelbach, Dept of Economics, Univ of MD at College Park.
Helpful suggestions were received from Richard Gates at StataCorp.
{title:Also see}
{p 4 13 2}
Online: help for {help est}, {help postest}; {help mfx}
{title:Author}
{p 4 8 2} {browse "http://www.uni-corvinus.hu/bartus":Tamas Bartus} {p_end}
{p 4 8 2} Institute of Sociology and Social Policy,{p_end}
{p 4 8 2} Corvinus University, Budapest, Hungary{p_end}
{p 4 8 2} URL: {browse "http://www.uni-corvinus.hu/bartus":http://www.uni-corvinus.hu/bartus}{p_end}
{p 4 8 2} Email: {browse "mailto:tamas.bartus@uni-corvinus.hu":tamas.bartus@uni-corvinus.hu}{p_end}