Brfss help!

[Hugo Torres]

Hello all! I am trying to analyze BRFSS data from 2011-2014. I am using the following SVYSET command:

svyset _psu[pweight=_llcpwt], strata(_ststr)

Which, according to BRFSS documentation, is what I should use to analyze data in accordance with the complex survey design of the BRFSS.

However, when I try to do anything, (for example, svy: tab or svy: logistic), I get the following error:

Note: missing test statistics because of stratum with single sampling unit.

What do I do about this problem? Thanks all! Hugo

[Clyde Schechter]

The message means what it says: the formulas used to calculate some of these statistics are not evaluable when there are any strata that contain only a single psu. The solution is to combine such strata with other strata. The combined strata will then be large enough to avoid this problem. In choosing which stratum to merge a singleton stratum with, you should try to pick one that is reasonably similar in terms of other characteristics. So, for example, if the strata are geographic, it would make sense to pick a stratum that is adjacent to the singleton.

There is a user-written program -idonepsu-, that can help identify the offending strata and carry out the merger for you. I have never used it myself, so I can't comment beyond noting its existence.

Added: It is best not to use abbreviations like BRFSS on this forum without explaining. The forum is interdisciplinary and international: I suspect that beyond the small group of American epidemiologists here, few will know what you are talking about. Some will pass over the entire post without reading on to determine that, based on general knowledge of the -svy- commands, they can answer your question without knowing that. In the future, please spell out that kind of abbreviation, and if you think people will need to refer to the actual source, provide a link or a complete reference.

[Joe Malaco]

Hi Clyde,

I have the same issue with the Behavioral Risk Factor Survey System (BRFSS) 2017. Could you please demonstrate how to combine strata?

Thanking you in advance!

[Clyde Schechter]

Run -help svydescribe- and click on the link "Remarks and examples" under the heading "Links to PDF Documentation" found just above the middle of that page. Read the examples and you will see how to do it.

But there is perhaps more to it than that. You will find one or more strata with a singleton PSU. In the example shown in the PDF Documentation, two were present, and they merge them together into a single stratum. But that may not be the best approach. You should take a look at the distributions of the key variables in your analysis by stratum and merge the offending PSUs into strata that resemble them on these distributions (if possible.)