Krishnan and Ed are correct and on target. The 2012-2016 and forthcoming 2017-2021 CTPP do NOT include the “extended workplace allocation”. Previous packages (1990?, 2000, and 2006/2010) DID have extended workplace allocation.

Allocation is the Census Bureau term for missing value imputation.

“Standard Workplace Allocation” is Census-speak for “primary imputation”. This process imputes workplace location down to the county, place and MCD level. (Imputation down to MCD-level is evident in “strong MCD” states such as Vermont. Vermont has 255 MCDs, and 184 Census Tracts!)

The standard allocation is needed for the Census Bureau to provide county-to-county commute patterns to OMB for purposes of defining metropolitan areas.

“Extended Workplace Allocation” is “secondary imputation” down to the TAZ, TAD and Tract levels.

Factor up the 2012-2016 CTPP commuter flows (TAZ-to-TAZ, TAD-to-TAD, tract-to-tract) to match county-to-county-by-mode control totals. Compare commute length frequency distributions before and after this “grossing up” process. It’s an in-elegant imputation procedure.

Also, use household travel survey data on home-based-work trips per worker (or home-work-home tours per worker, in activity models) to factor up workers to “observed home-based work trips” or “observed home-work-home tours”.

Normalize the 2012-2016 data to a single year estimate, say, 2013 or 2014 or 2015. Use published single-year ACS data on workers by county-of-residence, county-of-work, and intra-county workers. Use the single-year ACS PUMS data for inter-county commute controls.

An alternative to normalizing to a particular year, say, 2014, would be to use the single-year ACS PUMS at the PUMA of residence to PUMA of work (POWPUMA). Since POWPUMAs are always counties, the result would be a somewhat rectangular commute matrix of PUMAs-to-counties (by means of transportation). Unless you have really small counties (<100,000 population) which muddies up the analysis.

In my ideal world, you’d somehow have your population microsimulation software create a base year (say, 2014) commuter matrix at a very detailed geographic level, that is somehow “trained” by data from the CTPP, published ACS data, and PUMS.

And somewhere along the line it would be useful to integrate the Census Bureau’s LEHD data into the jumble.

I would recommend the statistical software package “R” for census data analyses, including the packages tidycensus, lehdr, and ctppr (hopefully we’ll see a ctppr package that will handle the new 2017-21 CTPP).

I would recommend the iPUMS (eh-PUMS) site to retrieve PUMS data. And the iPUMS project “NHGIS” for retrieving older, pre-2000 census data. Actually, I’d recommend trying out the tidycensus for PUMS data, as well.

That’s about all I have for now.

The 2012-2016 CTPP

On May 16, 2024, at 9:36 AM, Ed Christopher <edc@BERWYNED.COM> wrote:
Or put another way--The difference between the small areas and county totals is because the Census Bureau cannot (easily, my words) geocode all of the work locations to the Tract, Block Group or Block (including TAZ and TADs) but will code every work location to a county and state. The difference between the TADs and TAZs is more likely due to the CTPP special rounding rules and aggregating rounded values. The fix can be easy, just ID missing workers and allocate them based on best information available. How one chooses to do the allocation is the fun part since it comes down to how hard do you need to justify what you did. Being DC, I would suspect some geographic response concerns. The last work I have seen where someone looked at the distribution of the losses was the folks at the NY City Planning Department when the data first came out.

On 5/15/2024 2:37 PM, Krishnan Viswanathan wrote:

This is due to the workplace allocation issue in Census data. Standard workplace allocation procedures at the Census Bureau imputes only at the county and place level geographies & not below that. In previous iterations of the ACS based CTPP, the Census Bureau did an extended workplace allocation to overcome this difference but did not do so in the 2012-2016 CTPP. Details of how it was done for the 2006-2010 CTPP are here: https://transportation.org/ctpp/datasets/how-to-use-the-data-known-issues-tips-tricks. A more technical writeup is here: https://www.census.gov/content/dam/Census/library/working-papers/2013/demo/wp2013-26.pdf

On Wed, May 15, 2024 at 1:10 PM James Bunch <jabunch.work@gmail.com> wrote:

We are looking at the CTPP flows in the Washington DC region and notice that they are not the same depending on the level of aggregation (County to County, TAZ to TAZ, TAD to TAD). I presume this is due to the fact that some places can be assigned accurately at the county level but not the TAZ level. Has anyone looked at whether the losses are distributed randomly or if there are patterns, and how to account for the differences? Can we just factor up?

Here is an example:
CTPP 2012 - 2016
Table A302103

For Charles County internal trips:
County to County total trips: 30,065
TAD to TAD total trips: 23,745
TAZ to TAZ total trips: 23,752

I presume that the small difference at the TAD level vs the TAZ level is due to summation rounding, but maybe there is a slight difference in coverage.

Jim Bunch

P.S. I also posted this to the CTPP ListServ :-)

--

James (Jim) Allday Bunch, JABunch Transportation Consulting

411 Penwood Road, Silver Spring Maryland, 20901

240-271-3534 jabunch.work@gmail.com

To unsubscribe from the TMIP list, click the following link:
https://TMIP.ZEPHYRTRANSPORT.ORG/scripts/wa-ZEPHYRTRANS.exe?SUBED1=TMIP&A=1

--

Krishnan Viswanathan

To unsubscribe from the TMIP list, click the following link:
https://TMIP.ZEPHYRTRANSPORT.ORG/scripts/wa-ZEPHYRTRANS.exe?SUBED1=TMIP&A=1
-- 
Ed Christopher
708-269-5237
To unsubscribe from the TMIP list, click the following link:
https://TMIP.ZEPHYRTRANSPORT.ORG/scripts/wa-ZEPHYRTRANS.exe?SUBED1=TMIP&A=1