{smcl} {* *! version 1.0.0 17june2014}{...} {cmd:help mqgamma} {hline} {title:Title} {p2colset 5 16 18 2}{...} {p2col :{hi:mqgamma} {hline 2}}Marginal quantile estimation{p_end} {p2colreset}{...} {marker syntax}{...} {title:Syntax} {p 8 15 2} {opt mqgamma} {it:depvar} [{it:indeps}] {ifin} {cmd:,} [{cmd:quantile(}{it:numlist}{cmd:)} {cmd:lns(}{it:varlist}{cmd:)} {cmd:fail(}{it:varname}{cmd:)} {cmdab:aeq:uations} {cmd:from(}{it:matrix}{cmd:)} {it:display options} ] {marker description}{...} {title:Description} {pstd}{cmd:mqgamma} estimates marginal quantiles for the potential outcome corresponding to each treatment level from censored observational data in which the dependent variable is inherently positive, such as time-to-event data and health-expenditure data. Differences in these marginal quantiles are quantile treatment effects. {it:mqgamma} fits a two-parameter gamma distribution the conditional distributions from which the marginal distributions are subsequently estimated by regression adjustment. {pstd}Drukker (2014) derives the implemented estimator and provides an introduction to this command. {marker options}{...} {title:Options} {phang}{cmd:treat(}{it:varname}{cmd)} is a required option and it specifies the binary treatment variable. The treatment variable must be coded 0 for control cases and 1 for treated cases. {phang}{cmd:quantile(}{it:numlist}{cmd:)} specifies the marginal quantiles to estimated. Each specified quantile must be in {cmd:(0,1)}. {phang}{cmd:lns(}{it:varlist}{cmd:)} specifies the variables used to model the natural log of the scale. See the Methods and Formulas section of Drukker (2014) for details. {phang}{cmd:fail(}{it:varname}{cmd:)} specifies the binary failure indicator which must be coded {cmd:1} for an observed value and {cmd:0} for a censored observation. {phang}{cmd:aequations} specifies that the auxiliary-equation parameters should be displayed. {phang}{cmd:from(}{it:matrix}{cmd:)} specifies a row vector of initial values for the optimization routine. Each element in the specified matrix specifies the initial value for the corresponding parameter. {phang}{it:display options} are the standard display options. See {help estimation options##display_options:estimation options} {marker examples}{...} {title:Examples} {pstd}Setup{p_end} {phang2}{cmd:. use exercise}{p_end} {pstd}Medians of potential outcomes estimation{p_end} {phang2}{cmd:. mqgamma t active, treat(exercise) fail(fail) lns(health)}{p_end} {pstd}QTE(.25) and QTE(.75) estimation{p_end} {phang2}{cmd:. mqgamma t active, treat(exercise) fail(fail) lns(health) quantile(.25 .75)}{p_end} {phang2}{cmd:. nlcom (_b[q25_1:_cons] - _b[q25_0:_cons]) (_b[q75_1:_cons] - _b[q75_0:_cons])}{p_end} {title:References} {phang}Drukker, D. M. 2014. Quantile treatment effect estimation from censored data by regression adjustment. Working paper, submitted to the Stata Journal. {browse "http://www.stata.com/ddrukker/mqgamma.pdf":pdf paper}