ACS Data Users Group

Hello everyone,

I am looking for population, basic demographics, and housing costs at the census tract level per year. I know that ACS has yearly estimates at the MSA level and 5 year estimates at the tract level but neither will work for what I am doing (see below). Does anyone know of a data source for any of the above available at the tract level for each year? I am not worried about small sample sizes and would be happy with just population estimates.

Bus ridership is at its lowest point in 20 years. I am using transit ridership data at the bus stop level at 100,000 bus stops in 10 regions over 10 years. I need population, basic demographics, and/or housing costs to explain the changes in ridership over time at a hyper-local level.

Cheers,

Simon

There is no easy answer - especially across 10 regions. Offhand, there's annual data from The Home Mortgage Disclosure Act on mortgage loan amounts (which you can get through the CFPB) back to the late 1990s - this would be less helpful for high rental/multifamily areas. LODES data from Census would give you the number of workers back to 2002-2015 who live in a given area (block level), but that's not all the population and race/education only goes back to 2009.

So you could patch together those for the later years in combination with trend analysis of 2000 to 2010 to 2012/2016 ACS data (using NCDB or the Brown DB s4.ad.brown.edu/.../bridging.htm, which standardizes data for tract boundaries.) For reference, my list of most nationally-available datasets with small area data is a little outdated, but could be helpful for brainstorming and is at www.neighborhoodindicators.org/.../list-national-data-sets-small-area-data .

Hi Kathy,

Thank you so much! These data sources are exactly what I have been looking for! A couple follow-up questions:
- Why is HMDA data less helpful for high rental/ multifamily areas?
- Why is LODES data only available for 2002-2015. Have they stopped collecting data or will 2016 come out soon?

Also you list of data sources is phenomenal. I'll go through it in detail tomorrow

I can answer re LODES. I reached out to them about 2016 in April and received the following response: "No release date yet - we're waiting on some legal agreements to be finalized between the Census bureau and one of the data providers. There are beginning to be rumblings so I'd expect an internal push for 2016 LODES data in the immediate future."

Thank you Bernie, I'll send them an email just in case it's around the corner.

HMDA doesn't track rents in MF buildings or cash transactions by investors, and with large MF, there's usually not enough sales for trends. My HMDA guide is dated, but discusses strengths/limitations of the data set and gives sample indicators and analysis. www.urban.org/.../guide-home-mortgage-disclosure-act-data

Thank you Kathy, your guide to HMDA is an excellent ressource! I think I'll be able to use HMDA data but I am still looking to represent population/demographics per year.

My study period is 2008-2017 so its ok if LODES race/education data only goes back to 2009. The issue for me is missing 2016 and 2017.

I have read in the ACS Design and Methodology guide that the Master Address File, from which addresses are sampled, comes from USPS' Delivery Sequence Files. These files contain every registered address in the country and are updated twice per year. Marketing companies often buy address listings to send promotional material by mail. I haven't been able to find the number of addresses per census tract for each year. Do you know if these data are available?

More generally, do you know of any other data source with only population estimates (or some proxy) at the census tract - year level?

The FCC has population, household and housing unit estimates at the census block level, and you can aggregate that to Census Tract level. They have the data from 2010-2016 (pop2010 pop2011 pop2012 pop2013 pop2014 pop2015 pop2016)

here is the link : www.fcc.gov/.../staff-block-estimates

I am not 100% sure about the accuracy of the estimate but you might want to double check the documentation (methodology) (transition.fcc.gov/.../Annoted Methodology v1.2.pdf)

No matter who does the estimate for this sort of thing, one effectively allocates the data at a larger geography (here it would be PUMA) to a lower level geography, here it would be tracts. The second thing you would need to do would be to move them ahead a couple of years, since you would, in effect, be using the 5 year file, which centers 3 years out of date. We are working on this for Social Explorer. But we do not have them ready to go at this point for what you need.

Thank you so much all for your help finding data for this exciting research!

Bernie - I called and they said 2016 LODES should be available in 2 months but no guarantees!

Girma - Thank you for this! Unfortunately, I cannot use FCC data because they use county-level data to estimate block-level population. I am looking to see if the change in population of adjacent tracts (or blocks or block groups) affects transit ridership in these places.

Andrew - Since I really need single-year estimates, I've been thinking about deriving them from ACS' 5-year estimates. The problem is that it's a system of equations with n equations and n+4 unknown. Even using the 2010 Census, I wouldn't be able to get those 3 center years. Is that what you are referring to? Am I missing something?

Please let me know if anyone has other ideas this is super helpful!

Could you pool ridership across the comparable years and pair that with ACS 5-year data (e.g. ridership from 2012-2016 to match with ACS 2016 5-year)?

Given your study period and scope, there are few things to keep in mind if you do go with ACS 5-year data (whether to derive your own single-year estimates or use the data as-is)...
1 - Census tract boundaries changed in 2010
2 - Data were re-benchmarked to reflect 2010 Census counts
3 - Tracts with a large group quarters population may show variations that have nothing to do with change "on the ground" (the controlling routine for group quarters at the sub-state level can introduce unexpected noise)

Hi Beth,

Thank you for your response. Unfortunately we can't use 5 year estimates because the trends we are measuring are quite subtle.

What do you mean by data were re-benchmarked in 2010?

Can we get rid of the problem altogether by starting our analysis in 2011?

I hate to say it, but if you're looking for very subtle changes at a small geography, it could be tough to use any ACS data. The margins of error are fairly high, and any tricks you might use to estimate higher 1 year estimates to the tract level will likely exacerbate the issue. LEHD might be a your best bet here, as it uses administrative rather than survey data.

The population control totals used for estimates changed dramatically between 2009 and 2010. See: www.census.gov/.../2010_Change_Population_Controls.pdf
In short, variation you see between 2008/2009 and 2010-present may reflect corrections made during the decennial census, not change "on the ground."

And yes, you could avoid this issue (and the change in census tract boundaries) by starting your analysis in 2010.

Bernie - I wouldn't mind a little margins of error because I am using 100,000s of censustract-year combination. Nonetheless, I agree with you that LEHD/LODES will be my best bet for such subtle changes in small geographies. I was hoping to find something that extends beyond 2015, but hopefully 2016 will come out before the end of the analysis.

Beth - Thank you for the tip. I will also use some ACS data for other parts of the analysis and for comparison. I'll just stick to 2010/2011 and beyond.

Everyone - Thank you so much for your help. This forum has been incredibly helpful!