Variances in Census Block data for two files

Hi Angie,
These missing block groups may be related to the Census Bureau's new rules for suppressing median income and earnings estimates in cases where the margin of error is larger than the estimate. "Beginning in data year 2015, a new methodology was applied to the 5-year dollar-based medians and quintile limits that suppresses the table for the geographic area if the margin of error is larger than the estimate itself. "

Table B19013 is one of the tables affected by their new methods:
www.census.gov/.../2016-01.html

Thanks Mark! It's curious that there would be such a high margin of error. Do you have any insight into that?

Thanks Mark! It's curious that there would be such a high margin of error. Do you have any insight into that?

MOEs are almost entirely driven by the number of cases - interviews - in the group. Census blocks, depending on how many housing units they have, may have 1 or NO interviews in a 5 year period. Where there are few cases, MOEs are sky-high. The ACS really is simply not reliable at the census block level for anything; the only point of releasing block data is to make possible custom geographic aggregations. Even block groups, or in some cases tracts, have very high MOEs.

Thanks Patty, just to clarify--while decennial census data are available down to the block level, block groups are the smallest geography for which ACS data are released.

Thanks to Angie for pointing out the missing data.  

And thanks to Mark for suggesting that it might be due to a change in suppressing rules for median incomes in 2015 - I was not aware of this change and it is important information!

I was curious to see if I could replicate Angie's results and, if so, what the median income and MOEs were in 2014 for the missing block groups.

I found similar results - table B19013 for Essex County MA has 13 fewer block groups (534 versus 547) than B19001 for the same year or than B19013 for 2014.  Angie reported 26 missing block groups but I think this may come from AFF row count comparisons that have two rows for each block group.

Of the 13 missing block groups, 5 had MOEs larger than the estimate in the 2014 version of table B19013.  

		2010-2014	2010-2014
Missing block groups:		Median Income	MOE
250092011002		40556	46651
250092052002		36719	79314
250092063003		40781	46185
250092102001		79464	56366
250092173001		39397	41674
250092181002		65982	26585
250092221001		93750	33670
250092502004		47115	39097
250092508005		34590	19621
250092511002		27917	36972
250092609002		78194	67735
250092610001		50667	29415
250092682001		70726	37365

So Mark's hypothesis that it might be due to the new suppression rule seems plausible.  Additional evidence: the 2014 table had 16 block groups where the MOE was larger than the estimate and the 2015 table does not have any.

However, if the values were suppressed because of large MOEs, why isn't there any annotation?  The rows are just missing - no annotation is present for these block groups.  Census really should include annotation if these data values are being suppressed. 

Hi David,

We are aware of the issue with the missing annotation and we are hoping to have a solution for next year's release.

Thanks,
KaNin Reese
Census Bureau