Saving mean and CI after "svy: mean" or a more efficient way to plot survey trends with CI's?

[Karl Yesler]

Hello Statalist,

I'm new to using the -svy- commands and I am trying to construct trend graphs with confidence intervals that take into account the complex survey design. Whereas typically, I would have used twoway (rcap) and twoway (line) and repeat for ~ 4 groups, since I need to account for the survey design, I am using the following instead.

svyset [pweight=poolwt],  strata(varstr) psu(varpsu)  

svy: mean qty, over(year)

This produces the following output.

--------------------------------------------------------------
             |             Linearized
        Over |       Mean   Std. Err.     [95% Conf. Interval]
-------------+------------------------------------------------
qty          |
        2007 |   139.2986   2.012716      135.3455    143.2517
        2008 |   176.8884   5.057833      166.9546    186.8223
        2009 |   195.9715   5.236264      185.6872    206.2558
        2010 |   217.3776   4.675099      208.1955    226.5597
        2011 |   265.5767   10.75066      244.4619    286.6915
--------------------------------------------------------------

when I run this with -return list- I find that the matrix is part of the rclass information. How do I extract the mean and confidence interval so that I can use the twoway (rcap) and twoway (line) to plot the results?

Alternately, would there be a more efficient way? I am also new to marginsplot but it seems to be a command used after a regression and furthermore, is there a way to specify the confidence intervals and plot multiple lines and confidence intervals within 1 graph like I can with twoway?

Essentially, I would run the following and save those means and CI's as well, then plot all the different lines on 1 graph

svy: mean qty_2, over(year)

Thank you in advance.

[Nicholas Winter]

You can use Stata's -marginsplot- if you use first use svy:regress to calculate the means. For example:

sysuse auto
gen sweight=1
svyset [pw=sweight ]
svy: reg mpg i.rep78
margins rep78
marginsplot

* another option not to connect the individual estimates 
marginsplot, recast(scatter) name(version2)

* another option to turn the CI into a shaded area with transparency
marginsplot, recastci(rarea) ciopt(color(%25)) name(version3)

[Karl Yesler]

Thanks Nicholas, I looked into marginsplot, but I think the issue is that I am trying to plot multiple outcomes in the same graph, for example,

twoway (line qty year) ///
       (line qty_2 year)

and I don't see a way for marginsplot to do that. So I wonder if there is a way to generate and save the mean and CI from svy: mean to create the graphs myself?

[Nicholas Winter]

If you have multiple DVs that you want to include on the same graph, my -combomarginsplot- (available from SSC) is probably exactly what you need.