Weighted PROC SURVEYMEANS (applied in NHANES data) generating incorrect standard errors

[mconover]

I am currently taking a very basic look at variable distributions in (publicily accessible) NHANES data - National Health and Nutrition Examination Survey.  Since my data contains survey weights, I'm using PROC SURVEYMEANS to generate estimates of means for continuous variables and the corresponding standard errors. Unfortunately I am getting nonsense values for estimates of the standard error (and standard deviation). In an online tutorial page, NHANES provides code and explanation for using PROC SURVEYMEANS to estimate means and standard errors in their weighted data (see step 2 of http://www.cdc.gov/nchs/tutorials/NHANES/NHANESAnalyses/DescriptiveStatistics/Task3b.htm ). As far as I can tell, I have correctly contructed my weight variable and have appropriately specified the STRATA and CLUSTER statements in PROC SURVEYMEANS.

I can't figure out why but I get really high values for standard deviation and really low values for standard error.  The means however look correct.  I have included my PROC SURVEYMEANS code below.  Since this is all publicly accessible data, I have also included my output below.

Thanks for any help.

CODE:

proc surveymeans data=NHANES_DataMerge NOBS MEAN STDERR STD;

strata sdmvstra;

cluster sdmvpsu;

var LBDLDL RIDAGEYR BMXBMI;

weight weight;

ods output Statistics=printdata;

run;

PROC PRINT DATA=work.printdata;

RUN;

OUTPUT:

Obs    VarName         VarLabel                                                       N            Mean                   StdErr            StdDev

1         LBDLDL          LDL-cholesterol (mg/dL)                            5245       127.594611      0.711109       1912657887

2         RIDAGEYR     Age at Screening Adjudicated - Recode    5245       52.729637        0.189132        779619697

3         BMXBMI         Body Mass Index (kg/m**2)                        5245       28.604494        0.128091        417023923

[ballardw · Accepted solution]

When posting links to sites in this forum make sure that there is no punction at the end as that gets treated as part of the link and hence get "link not found". So place a space before the ")." in the link you posted.

Surveymeans reports Standard Deviation of the SUM when requested using STD.

You might want to look into requesting LCLM UCLM for confidence intervals of the mean to look at dispersion.