Adjusting survey data for age and sex

[Laura Andrews]

I had adjusted my survey data by age groups according to the 2010 US Census, but now I want to simultaneously adjust by sex. How would I go about creating the new standard age and sex adjusted weights?

For an example of how I calculated the age-adjusted weight variable, I will include my code below. I used an age variable with 7 age groups. I also included the survey command I used to calculate prevalence estimates of my outcome variable applying the standardized weights.

gen std_wgt10 = 0.220724 if age7cat==0
replace std_wgt10 = 0.171133 if age7cat==1
replace std_wgt10 = 0.185875 if age7cat==2
replace std_wgt10 = 0.178898 if age7cat==3
replace std_wgt10 = 0.124713 if age7cat==4
replace std_wgt10 = 0.070752 if age7cat==5
replace std_wgt10 = 0.047905 if age7cat==6

svy: mean cms ,stdize(age7cat) stdweight(std_wgt10)

[Clyde Schechter]

The -stdize()- option only accepts a single variable. So you have to create a combined age-sex variable and corresponding weights for it. This entails that you must already have a weight variable for sex adjustment. I'll call it std_wgt_sex

egen age_sex = group(age7cat sex)
gen combined_weight = std_wgt10*std_wgt_sex
svy: mean cms, stdize(age_sex) stdweight(combined_weight)

[Maarten Buis]

Since Laura has census data, and she can create weights directly:

//===================== "create" example data
// you obviously don't have to do this step, as you already have the data
// "create" a (silly) census
clear all
set seed 583275935
set obs 1000
gen female = runiform() < .5
gen agecat = ceil(10*runiform())*10
tempfile census sample
save `census'

// draw a sample
bsample 100
save `sample'

//========================== create weights
use `census', clear

// number of nonmissing observations in census
count if !missing(female, agecat)
tempname ncensus nsample
scalar `ncensus' = r(N)

// create a table of number of observation per age and gender
contract female agecat, freq(census_freq)
tempfile table
save `table'

use `sample'
// count the number of observations per age and gender in sample
bys female agecat : gen obs_freq = _N if !missing(female,agecat)

// count number of nonmissing observations in sample
count if !missing(female, agecat)
scalar `nsample' = r(N)

// merge in the counts from the census
merge m:1 female agecat using `table'

// create weights
gen w = census_freq/obs_freq * `nsample'/`ncensus'

[Laura Andrews]

Thank you Clyde and Maarten! I wasn't sure how to correctly combine the age and sex variable. Now I am able to complete the analysis.

[John Lure]

Hi everyone,

the command <svy: mean cms ,stdize(age7cat) stdweight(std_wgt10)> reports the adjusted mean in a dataset adjusted by age groups, but how can I obtain the percentiles and the SD in the same example?

Thanks, best, John

[Steve Samuels]

Welcome to Statalist, John. To improve your chances of a good answer, read FAQ 12 and do everything it says. To have readable questions, paste code and results from your log or screen between code delimiters [CODE] and [/CODE]. into the forum editor. See the answers above.

Here you refer to an unidentified example and ask for for percentiles and SDs, but of what? Use dataex, described in the FAQ, and provide an extract of data with just enough variables to illustrate your problem or use one of the datasets available from within Stata.

 help datasets

I woujld also suggest that you not attach your questions to old threads, unless they are followups on that thread's topic. Here you are apparently asking about a completely different. topic.

[Steve Samuels]

Ah.. in the light of morning, I see that the statistics you want are for the variable "cms". To get the estimated population standard deviation and percentiles, you'll need to create a post-stratification weight to feed into svyset. The code below uses Nick Winter's survwgt poststratify package (SSC) to do this. After svysetting the data with the new weight, svy: mean is run, followed by estat sd. Selected percentiles are found with the _pctile command (part of percentile) and with summarize, detail, with aweights. I have Stata 14; Some of this may be unnecessary in Stata 15.

.  set seed 400912

.  use http://www.stata-press.com/data/r14/hbp, clear

.  egen group = group(age race sex) if inlist(year, 1990, 1992)
(675 missing values generated)

.  gen selwt = 100+ runiform()*1000

.  gen bp = 100+rnormal(0,5)

.  by group, sort: gen stdw = _N

.  svyset _n [pw = selwt]

      pweight: selwt
          VCE: linearized
  Single unit: missing
     Strata 1: <one>
         SU 1: <observations>
        FPC 1: <zero>

.. /* Standardized Mean */
.  svy: mean bp,  stdize(group) stdweight(stdw)
(running mean on estimation sample)

Survey: Mean estimation

Number of strata =       1        Number of obs   =        455
Number of PSUs   =     455        Population size =  273,072.4
N. of std strata =      24        Design df       =        454

--------------------------------------------------------------
             |             Linearized
             |       Mean   Std. Err.     [95% Conf. Interval]
-------------+------------------------------------------------
          bp |   99.90081   .2600828       99.3897    100.4119
--------------------------------------------------------------

.
. /* Use survgt package to compute  post-stratified weights */
.
. cap ssc install survwgt, replace

. survwgt poststratify  selwt, by(group) totvar(stdw) gen(adjwt)

.  svyset _n [pw = adjwt]

      pweight: adjwt
          VCE: linearized
  Single unit: missing
     Strata 1: <one>
         SU 1: <observations>
        FPC 1: <zero>

.  svy: mean bp
(running mean on estimation sample)

Survey: Mean estimation

Number of strata =       1        Number of obs   =      1,130
Number of PSUs   =   1,130        Population size =      1,130
                                  Design df       =      1,129

--------------------------------------------------------------
             |             Linearized
             |       Mean   Std. Err.     [95% Conf. Interval]
-------------+------------------------------------------------
          bp |   99.79771   .1727626      99.45874    100.1367
--------------------------------------------------------------

.  estat sd

-------------------------------------
             |       Mean   Std. Dev.
-------------+-----------------------
          bp |   99.79771    5.191779
-------------------------------------

.  _pctile  bp [pw = adjwt], percentiles(5 10 25 50 75 90)

.  return list

scalars:
                 r(r1) =  91.30558013916016
                 r(r2) =  93.28947448730469
                 r(r3) =  96.23242950439453
                 r(r4) =  99.72535705566406
                 r(r5) =  103.0582885742188
                 r(r6) =  106.5712661743164

.  summarize bp [aw =adjwt], det

                             bp
-------------------------------------------------------------
      Percentiles      Smallest
 1%     87.58885       83.87115
 5%     91.30558         85.007
10%     93.28947       85.21439       Obs               1,130
25%     96.23243       85.26249       Sum of Wgt.       1,130

50%     99.72536                      Mean           99.79771
                        Largest       Std. Dev.      5.191779
75%     103.0583       113.7785
90%     106.5713       113.7878       Variance       26.95456
95%     108.4101       114.5298       Skewness       .0106384
99%      112.118       117.9779       Kurtosis       3.039936

[John Lure]

Hi Steve, thank you very much for your advices; I'll follow them when using this Forum. And a big thank you for your answer. It totally solved my problem, best regards, John.