I have a survey methodology question. I am preparing a report that contains some demographic information. My supervisor instructed that I use the NCHS bridged-race population estimates when reporting the state population. However, he has also instructed that I use the ACS IPUMS to calculate rates on various indicators such as poverty and language isolation. Then, I am to apply that rate to the NCHS population estimate; the product of the ACS IPUMS rate and the NCHS population estimate is the "adjusted true estimate" according to my supervisor. Since then, I have found that what he is instructing me to do is called post-stratification. I am, however, curious as to whether this is sensible practice given the methodology used to calculate NCHS bridged-race population estimates (http://www.census.gov/popest/methodology/2013-natstcopr-meth.pdf). Is my supervisor's proposed post-stratification method correct or poor form of survey analysis methodology?
1 Answer
Post-stratifying ACS sounds very funny to me as a survey statistician. It's like "rectifying Google search". Generally, ACS sample sizes are at least an order of magnitude larger than sample of any other survey (more often, two or three orders of magnitude larger). So ACS should generally be used as the target population to calibrate / post-stratify to. However, if there is a standardized medical (quasi-)population that is only available through NCHS, designed to make trending or comparisons apple-to-apple, then I see the point of this exercise of calibrating towards NCHS.
Please check that you are not confused about terminology, see clarifications in http://www.surveypractice.org/index.php/SurveyPractice/article/view/315.
Also, I think that what you are effectively after are some sort of indirect synthetic ratio estimation, in terms of the traditional small area estimation methodology.