Deriving ACS Estimates without an existing MOE

2005-2009 has overlapping data 4 of the 5 years with 2006-2010 -- meaning 4/5 of the data is identical -- so a test of significance would be invalid.

I think your larger question is how you can manipulate the MOE to match the manipulation of the estimate to be in a uniform geography. The short answer is that you can't. However, you can make a guess at the MOE by using some of the algorithms in the appendix of the Handbook for Researchers (www.census.gov/.../ACSResearch.pdf), depending on how you manipulated the estimate.

Friendly note: Some of the URLs for ACS changed recently. Handbooks may be found here: www.census.gov/.../handbooks.html

What level of geography are you trying to compare?

Diane,

You can't make this comparison because of the overlapping years. But, here's how to produce new margins of error based on concatenated geographies:

Tool: www.psc.isr.umich.edu/.../
Instructions: www.psc.isr.umich.edu/.../ACS_Aggregator.pdf

And, here's an intro into the noisiness of ACS data and why you might want to combine geographies:

www.psc.isr.umich.edu/.../

Note: This is an old document so it refers to some events that have happened (increased ACS sample size).

Thank you for all of the responses!

To elaborate a bit more... I am working with block groups and would like to compare 2005-2009 through 2009-2013 ACS data.

I studied the What Researchers Need to Know ACS document (Appendix 4) and found that I can compare overlapping ACS years by using statistical significance. But I have not come across how I can manipulate 2010 Geography MOE's to fall into 2000 Geography MOE's. www.census.gov/.../researchers.html

Diana, a universal convertor of many types of Census and other semi-official geographies is provided by Missouri Census Data Center, mcdc.missouri.edu/.../geocorr12.html.

As far as correlations due to overlapping years go, if I were desperate, I would use a correlation of 0.8 if 4/5 of the years overlap. But generally I would expect the estimates at the low levels like block groups to be extremely sensitive to variance estimation methods, and obviously bluntly adding up the squared margins of error, or even accounting for 0.8 correlation, isn't going to be quite right. Unfortunately, you won't even be able to get consistent PUMAs for the 2009-2013 ACS, as they switched to the new ones right in the middle.