ACS Data Users Group

I am comparing several demographic and social-economic variables(pct of population over 65 years, median income, pct of poverty, etc. ) between census 2000 and ACS 2012-2016 data. I mainly used data from census 2000 SF3 which is is from long form questionnaire and based on a sampling type of 1 in 2 or 1 in 4 or 1 in 6/8. In a word, SF3 is a not a count data but sampling data. I know that there is instruction to conduct statistical test between 5 year ACS data, but does it apply to comparison between census 2000 and ACS 5 year estimate? If the statistical test is not applicable, which statistical test should I use ? I also found out calculating error for census 2000 data from the technical documents. I assume the SE of census 2000 can be used to compare it to ACS 5 year data? Any help is apprecciated, thanks!

While I am far from an expert on this particular topic I do believe that you will find instructions in how to calculate standard deviations (and by extension margin of error) for Census 2000 SF3 statistics in statistical publications from that census that include SF3 data. For example, I have a copy of the Summary of Social, Economic and Housing Characteristics for Massachusetts and Appendix G covers this topic*.

You will need to use tables of both unadjusted standard errors and design factors to estimate the standard deviation included in the Appendix. Once you have calculated a standard deviation and margin of error for a statistic then I think you can use the techniques included in the appropriate ACS publication to compare the values. If your Census 2000 variables are from the long form questionnaire (eg % population over 65) then this method does not apply I believe. In such cases compare the Census 2000 value to the ACS margin of error range.

* You can find similar volumes via Google Books it appears.

I welcome corrections!

Hi Cliff, thanks for the reply!

I did find unadjusted standard error and design factor table from census 2000 technical document so that I can compute standard errors and CI. I am concerning with comparing census 2000 SF3 data to ACS 5 year estimate data. It seems you don't think the between ACS 5 year comparison and statistical test is applicable to the comparison between Census 2000 SF3 and ACS 5 year data. Then , would you suggest any other appropriate statistical test for the comparison between them?(Census 2000 SF3 to ACS 5 year )

I DO think that the ACS 5 year comparison and statistical test is likely applicable. However, I suggest that you look over the portions of the Census 2000 appendix to see what is said there about statistical tests. I suspect that the language is a direct parallel with later ACS documentation on the same topic, with which you might want to compare.

Hi Cliff,
Thanks a lot! I find it very helpful you mentioned the Census publication for each state so that I found table of design factor for my research state Maine . Because such table is missing from the census 2000 SF3 technical document. As for the statistical test, I will use the methods used for comparison between ACS 5 year estimate for now as I don't find an instruction on this issue.

Jing
Research Associate, University of Maine