07-24-2012 06:44 PM
There's an option on SURVEYMEANS called STDERR, and it outputs a "Std Dev" of the sum. I've been told this is a standard error, except I don't know why it isn't named as such. To add to the confusion, on SURVEYFREQ there is also a "Std Dev" that is output from the WTFREQ option. However, there's also a standard error of the percent that outputs by default.
My question is: are the "Std Dev" actually standard errors? If not, is (standard error of percent)/(percent)*(weighted frequency) a valid version of the standard error?
A quick test I did to try to show that "Std Dev" should be standard error is that I worked backwards from the 95% confidence intervals, and got similar numbers to "Std Dev". But the sum standard error is poorly documented on the Web, so I could be doing it wrong.
07-24-2012 07:32 PM
Proc surveymeans outputs the mean, st.error, and confidence limits for the mean by default. If you are requesting other statistcs, you need to provide the specific syntax and options you are using. The interpretation of the output may depend on the position of the option in the procedure syntax. Proc statement options, Ratio statement options, etc.
Standard error is the estimate of standard deviation of the estimator of the statistic in question, based on the given sample size. So yes, the confidence limits for the statistic (e.g., the mean) will be based on this standard error.
Standard deviation is usually referring to the variation of the original variable raw values in the population, not the variation of a statistic from the sample with a fixed sample size.
07-25-2012 04:58 AM
Okay, I think I got it: even though the output of surveymeans and surveyfreq says "Std Dev" of sum or weighted frequency, it really means the standard error. I guess it's just imprecisely worded and documented. However, it seems odd to also have a standard error of percent value, as this seems to mean that the percent is of a different sample than the "Std Dev of Wgt Freq", since it's a different value than the "Std Dev".