Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
From | Christine Brickman <cbrickman79@yahoo.com> |
To | statalist@hsphsun2.harvard.edu |
Subject | Re: st: holding variables to weighted means with prvalue |
Date | Wed, 11 Aug 2010 10:18:31 -0700 (PDT) |
Thanks, Steve. Moments ago, I just discovered I can accomplish this with margins, which is new to Stata 11, I believe. svyset [pweight=weight] svy: regress i.male [list of remaining variables] margins i.male If you weight the data in the regression command, margins retains the weight. But there’s also an option to turn it off if you choose. ----- Original Message ---- From: Steve Samuels <sjsamuels@gmail.com> To: statalist@hsphsun2.harvard.edu Sent: Wed, August 11, 2010 12:06:24 PM Subject: Re: st: holding variables to weighted means with prvalue Not with -prvalue-. Some of the holding specifications, like -max- and -min- have no survey meaning. So, apparently, none of them do. Below is a do file which uses -adjust- which creates predictions at the observed values for all covariates, holding others at their weighted means. It optionally plots the adjusted predictions. Steve Steven Samuels sjsamuels@gmail.com 18 Cantine's Island Saugerties NY 12477 USA Voice: 845-246-0774 Fax: 206-202-4783 On Wed, Aug 11, 2010 at 10:43 AM, Christine Brickman <cbrickman79@yahoo.com> wrote: > I want to compute predicted values while holding variable at their weighted > means. Even though I'm weighting the regression, STATA is holding the >variables > > at their unweighted means. Here's my syntax: > > svyset [pweight=weight] > svy: regress [VARLIST] > forvalues i = 0(1)1 { > prvalue, x(male=`i') > } > > Is there a way to hold all variables except male at their weighted means? > > Thanks! > Christine /**************************START CODE***********************************/ /*************************************************************************** This do file e creates predicted values from survey regression which vary in one predictor at a time, holding others at their estimated population means. Also computed are upper and lower confidence interval endpoints for the prediction. An optional graphics section plots predictions and CI endpoints for each variable against that variable. ****************************************************************************/ di c(level) // current confidence level . To change, -set level-. sysuse auto,clear svyset _n [pweight=rep78] /*************************************************************************** Define a predictor list for the right-hand-side of the svy: logit equation ****************************************************************************/ local rhs mpg weight trunk svy: mean `rhs' //outputs weighted means in _b[varname] /*************************************************************************** For each varying predictor, build up the -adjust- command. For the adjusted predictor which varies mpg, the adjust command starts: "adjust weight = 2941.106382978724 trunk = 13.73191489361702" ****************************************************************************/ foreach v of varlist `rhs' { local less : list rhs -v //Remove variable v from the rhs local list_`v' " " foreach w of varlist `less' { local lw "`w' = `=_b[`w']'" // e.g. "weight = 2941.106382978724" local list_`v' `list_`v'' `lw' // Augment the list } di "`v'" di "`list_`v''" } /*************************************************************************** SET UP THE -SVY: REG ****************************************************************************/ svy: reg foreign `rhs' /*************************************************************************** For each predictor variable v (e.g. mpg), use -adjust- to compute: pred_v : linear prediction se_v : the se of the linear prediction T llim_v : lower 95% confidence interval endpoint ulim_v : upper 95% confidence interval endpoint ****************************************************************************/ foreach v of varlist `rhs' { adjust "`list_`v''" , g(pred_`v' se_`v') se //The key line gen llim_`v' = pred_`v' - (invttail(e(df_r), (1- c(level)/100)/2))*se_`v' gen ulim_`v' = pred_`v' + (invttail(e(df_r), (1- c(level)/100)/2))*se_`v' label var pred_`v' "Predicted Prob" label var llim_`v' "Lower CI" label var ulim_`v' "Upper CI" } // save (choose a new name) // Optional Graph Module /* foreach v of varlist `rhs' { twoway scatter pred_`v' llim_`v' ulim_`v' `v', msymbol(o o o) title("Plot of Prediction against `:variable label `v''" "Other Predictors at their Mean") note("Program: `pname'. Data: ") saving(graph_`v', replace) } */ /************************END CODE ***********************************/ * * For searches and help try: * http://www.stata.com/help.cgi?search * http://www.stata.com/support/statalist/faq * http://www.ats.ucla.edu/stat/stata/ * * For searches and help try: * http://www.stata.com/help.cgi?search * http://www.stata.com/support/statalist/faq * http://www.ats.ucla.edu/stat/stata/