ACS Data Users Group

The Census Bureau is planning to discontinue the production of all 3-year ACS data, according to their 2015 budget justification document:

"The Census Bureau proposes to terminate permanently the “3-Year Data” Product. The Census Bureau intended to produce this data product for a few years when the ACS was a new survey. Now that the ACS has collected data for nearly a decade, this product can be discontinued without serious
impacts on the availability of the estimates for these communities."

www.osec.doc.gov/.../Census_2016_CJ.pdf

There have also been some emails circulating that mention the 2015 ACS Data Users Conference has been canceled but I want to assure everyone that this is not the case!

Hi Mark,
I'm sure I'm missing something here -- but won't this be a real loss for users who rely on the most up to date information available for smaller geographic areas?
Thanks for sharing.
All best,
Susan

Susan,
The loss of 3-year ACS data would have major implications for those who need data for small geographic areas. We have often used them here at PRB because they provide a nice compromise between reliability and timeliness.

Thanks for clarifying! Now I understand that you were just quoting the Census Bureau, not agreeing that the project can be discontinued without serious negative impact. I'd be happy to be involved in efforts to have this product restored...

I don't often use the 3-year data, but I just found myself looking forward to a 3-year PUMS data set with all 3 years aligned with the 2010 PUMAs. Our PUMA boundaries changed significantly, making it very hard to work with multiple year data that straddle the 2000 and 2010 PUMAs. Would this change affect the 2014 data releases?

I generally find it difficult to use any ACS products 1,3, or -yr because of the MOEs. That said, over the last five or so years I have gravitated to focusing on the 3-yr products for following trends because they, for my needs, give the very best tradeoff between recency and MOEs.

So for general background work, for me, it would be a total loss to lose the 3-yr as the 1-yrs generally have too much error and the 5-yrs are too out of date.

Obviously the population of the geography you are interested in determines what you can reliably use.

stan drezek

My understand of what I've read is that the 2014, or 2012-14 3 year, release this fall will go ahead as scheduled. The 3 year product stops for vintage 2015. I have used 3 year data a lot, especially for communities between 65K and 200K where the 1 year data have unacceptably high MOEs for most data, and I strongly object to discontinuing them. I am unaware of any efforts to consult with users in this decision, and the reference to 3-year as a "temporary product" goes back more than 10 years to the very beginning of ACS. I would argue that it's now an institutionalized product of the same importance as 1 year and 5 year. In my opinion, it would make more sense to stop issuing 5 year products on an annual basis.

My understanding is that this is at root a budget issue. Nonetheless, I think this is a huge loss and a real problem for medium size communities, in that 20K to 250K range, where the one year data is too unstable year to year to use for policy formulation but a five year timeframe for data is very problematic in different ways for determining public policy. In some cases five year values are not really meaningful (What is a five year average or median rent? Is that really that useful a value?). In most cases 5 year data cover too long a period of social change to give a good sense of what is happening on the ground. (Think about 2005-9 or 2008-12 where the economy underwent a major shift and resulted in a lot of tumult and dislocation.)

I have been using 3 year data since it became available as the best compromise between timeliness and accuracy. I'll have a hard time explaining this to the people I work with and for.

[Updated on 2/3/2015 4:52 PM]

The 5 year ACS file is very useful for those of us who look at the characteristics of U.S. immigrants by national origin in different types of geographic areas. Given the small population size of all immigrant groups except Mexicans, the 5-year file is the only ACS file that has sufficient foreign-born cases for many types of analyses. It would be a big mistake to offer up the 5-year ACS in order to keep the 3-year ACS. If cost constraints are the issue, issuing the 5-year on an alternate year basis would not be a major loss but cutting it completely would force immigration researchers to focus on the total foreign born and assume homogeneity within that population. Our research based on the 5-year ACS shows that there is considerable national origin heterogeneity in settlement and integration processes and that can only becomes clear if one has a sample with a sufficient foreign-born cases to permit national origin disaggregation. I might add that the 5-year ACS (2007-11) has a smaller foreign-born sample than the 2000 PUMS 5% decennial file even though the immigrant population continues to increase annually.

It seems clear to me that since the Census Bureau has been collecting ACS data every year now for 2005 and because 5-year data is the most detailed we don't need 3-year data anymore.

Under no circumstances do I advocate cutting out the 5 year data set. The question is: how often should it be issued? I could make an argument for every 5 years, twice a decade, when each edition would have completely new data. One of them could be years 8, 9, 0, 1, 2; this would match the census. Perhaps others can make an argument for more frequent production.

The 3-year does not provide data for all the counties in my study-area or the small block-group size Traffic Analysis Zones (TAZ) that we use for transportation modeling. The yearly offering of 5-year moving averages provides some understanding of trends and helps when I need data for an "off" year. So I would argue for a contiuation of yearly 5-year data releases.

All that said, the arguments for the timeliness of a 3-year set that are made above are persuasive and I would support the continuation of the product based on the need of those who are using it now. A part of me want to get enough samples collected each year to support a 3-year moving average and the spatial resolution we need here! :)

Though I hate to see datasets discontinued, this is the best candidate for elimination. The 1-year offers the most timely results and the 5-year offers aggregation at the smallest level (Block Group) and ZIP Code Tabulation Area. The 3-year offers none of these benefits.

As a demographer for a mid-sized city (134,000 pop) I use the 3-year estimates since they provide the best compromise between timeliness and accuracy. Large areas may be fine with the 1-year estimates, but for smaller places the margins of error are often too high and the three year product is preferred over the one year.

I find all of them useful, the 1 year, 3 year and 5 year, and so don't want to see any of them discontinued. But what's the process? At this point, it's just something the Census Bureau is proposing, right? Will there be a comment period, are population or statistical groups going to lobby to keep them, can we write to our congress people?

My agency uses 3-year ACS data for low-income designation statistically valid surveys. We determine the median family income for specific areas to use in determining representativeness of the sample size. While we could use 5-year data, our Office of Chief Economist suggests the 3-year statistics are the most reliable for this purpose.

This is a pretty big loss to the ACS program, but the dissemination piece of the ACS program struck me as pretty daunting. They turnaround the data really quickly and for lots of geographies.

I think if the Census Bureau follows through with this "cost-reduction" plan they won't start it back up. But, they aren't losing the underlying data unlike other cuts federal program make where they cut the sample, variables, etc. This could be resurrected.

So, someone could propose to do this for the Census Bureau via an RDC, but it is a big job. It isn't just a 1-person job and the Census Bureau would want to make sure of quality control, which would use personnel and $$.

We in transportation planning definitly need the 5 year data to get at smaller geography. Every year with this helps. The 3 year misses too many smaller areas, I only minimally used the 3 year. However, the data is still there, its just that the Census does not want to spend the money to summarize the 3 year data. We have all these private companies collecting data from cell phones all the time, but we can't pay to have the federal government summarize its own data. THis is part of that whole cutting of discretionary fudning, time to ask that extra money be found for this, and how much is the savings supposed to be? IF its under a Billion, I say time to fund it as part of the cost of data needed by many including the private sector.

The 3-yeaer ACS file is essential for my research. Besides its necessity for small areas, it is essential for small subpopulations. The 5-year files, although also very useful, have a timeliness implication, by extending the time-lag for estimates. I think that efforts should be made to keep the 3-year files and perhaps savings can be made by eliminating some questions.

I use 3-year estimates for almost everything. Vermont has small counties-in a small state. And a couple more releases would have really helped with being able to look at any kind of trend; 1 year just doesn't work for that here, and I'll be waiting for ever for enough non-overlapping 5-year sets to do that. Thanks Mark for the heads-up!

When I suggested that the 3-year series could be eliminated as long as the 5-year series was retained, I assumed that the 1-year updates would continue. Wouldn't that make it possible to have an annual (rolling) 5-year series update and isn't that what is intended if the 3-year series is eliminated?

It is worth noting that the concerns about margin-of-error on the 5-year products are more a problem of population and sample size for small jurisdictions, and less a problem of the 3-year versus 5-year aggregations.

Larger sample size might allow the Census Bureau to provide finer geographic detail in the 1-year products and would surely improve accuracy in the 5-year.

And (in reply to Oleh) - the Census Bureau just completed an exercise to investigate eliminating questions. That has a much broader impact and is generally not well received in the data user community. Every question has an important purpose.