ACS Data Users Group

I've found a very suspect value in table B01002D, column B01002D_002, which is MEDIAN AGE BY SEX (ASIAN ALONE), for Males. In Placer County, CA, census tract 210.38, block group 4 (060610210384), the value for the median age is 115+. You can check it out here: factfinder.census.gov/.../0500000US06061.15000

This is the only median age value for any geography in the country greater than 99. I'll e-mail Census, and see if they have any further information.

From a strictly technical perspective, the MOE note states: "An '***' entry in the margin of error column indicates that the median falls in the lowest interval or upper interval of an open-ended distribution. A statistical test is not appropriate."

The 2013 value for male median age for that tract is: 70.0 median (+/-94.7). Such a wide range implies an extraordinarily small sample size.

But from a real-world perspective, it's hard to believe that tract has such a high median age. Satellite images show that most of the tract is your standard, suburban housing. There's a major apartment complex in the tract that advertises "after school programs."

115+ would be a suspiciously high median even in a community entirely comprised of retirement residences or assisted-living facilities. In a run-of-the-mill suburban block group, I'm going to venture to suggest that it's a virtual impossibility.

This case speaks clearly to the problems with ACS survey results at the smallest areas of geography (or for small population groups).

Keep us posted on what you learn!

Table B000001 has unweighted sample sizes for each geographic level. This may help you assess your results,

Interestingly, the 2010 SF1 median age for Asian Alone males in 06061021384 is 39.3. The oldest people the Census counted were 2 males and 1 female in the age group 75-79.

Here's the deep link to see for yourself:
factfinder.census.gov/.../0500000US06061.15000

Here's the Census's response (basically, what you've all been saying):

Hello:

The "+" in this instance does not mean what you think it does. If you look at the table footnotes you will notice that "An '+' following a median estimate means the median falls in the upper interval of an open-ended distribution."

Also, for the MOE there is the following note "An '**' entry in the margin of error column indicates that either no sample observations or too few sample observations were available to compute a standard error and thus the margin of error. A statistical test is not appropriate."

These messages are basically saying that there were too few estimates for that cell and thus it is not a statistically significant statistic and should not be used.

I'm not sure that's a terribly satisfying answer, given that values in those tables are supposed to have been suppressed for small sample size. If the issue in this case is too-small sample size, then why were the data reported in that table?

It still strikes me as surprising that this statistic was reported at all. Does anyone know of other instances where the ACS reports results with similar caveats? It would seem that such instances should be suppressed.

"Data quality filtering rules are applied to ACS 1-year and 3-year estimates, but are not applied to the 5-year estimates. "

www2.census.gov/.../ACSO_Data_Suppression.pdf

You know... that's what I understood (prior to this example), too, but that same table has hundreds of cells with "-" for data value, with footnote that:

"An '-' entry in the estimate column indicates that either no sample observations or too few sample observations were available to compute an estimate, or a ratio of medians cannot be calculated because one or both of the median estimates falls in the lowest interval or upper interval of an open-ended distribution."

... which *implies* some suppression is in use even in the 5 year tables.

So for this case to have passed the filtering test, there had to have been 3 observations (or more?), 2 of which were top-coded 115+.

If anyone from the ACS branch at the Census Bureau is reading this thread I'd be interested to hear their thoughts. Perhaps this traces back to a coding error or a facetious answer, though I image there are a number of processes in place to catch such problems.

It wouldn't be so bad if the ACS text files listed the value as "115+" as this coerces the value to text rather than a number which would prevent inclusion from calculations in databases, statistical packages, and GIS applications. But I just downloaded the text files and this value appears as 116 which means it will show up in calculations (unless specifically controlled). I ran a MIN calculation on the same field and found a census tract with a median age of 1.5 with an MOE of +/-16.5. Strangeness abounds.

I couldn't get the link to work so the 1.5 value is in table B01002D, Male. Census Tract 2421, Los Angeles County, California

[Updated on 12/10/2015 4:15 PM]