LODES Data Questions...

Mike-Gmail

unread,

Jun 2, 2020, 6:22:58 PM6/2/20

to tm...@mg.tmip.org, TRBA...@googlegroups.com, hka...@dot.ga.gov

Dear TMIP and TRB ADD30 Members,

Has anyone used/evaluated the LEHD Origin/Destination Employment Statistics (LODES; https://lehd.ces.census.gov/data/#qwi) dataset for the purposes of doing any type of transportation planning?

In particular, I'd like to hear from anyone who has experience or information on systematic biases/under reporting of employment counts. Technical paper references would also be extremely helpful.

Thanks in advance for any help anyone can provide.

Yours truly,

Mike Greenwald

Planning Director/MPO Technical Coordinator

Macon-Bibb County Planning & Zoning Commission/MATS MPO

200 Cherry St., Suite 300

Macon, GA 31201

Tel: 478-338-9472

E-mail: mgree...@mbpz.org

Ed Christopher

unread,

Jun 2, 2020, 8:04:20 PM6/2/20

to tm...@mg.tmip.org, TRBA...@googlegroups.com, hka...@dot.ga.gov

You might want to look at this http://onlinepubs.trb.org/onlinepubs/nchrp/docs/NCHRP08-36(98)_FR.pdf
NCHRP 08-36/Task 98 [Final]
Improving Home-to-Work and Employment Data for Transportation Planning

Systematic bias would be linked to the source data and any warts that might exist in the state's QCEW data. Stuff like how employers with multiple locations are handled and whats missing in the QCEW are obvious places to look.

--
You received this message because you are subscribed to the Google Groups "Transportation and Land Development Committee" group.
To unsubscribe from this group and stop receiving emails from it, send an email to TRBADD30+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/TRBADD30/001101d6392c%245d0cc2f0%24172648d0%24%40gmail.com.

-- 
Ed Christopher
Transportation Planning Consultant
708-269-5237

Graham, Todd

unread,

Jun 9, 2020, 1:47:56 PM6/9/20

to Mike-Gmail, tm...@mg.tmip.org, TRBA...@googlegroups.com

Hi Mike—

This is a perennial question in the 16 years that LODES has been around. I did something like an informal assessment of it several years ago. Since the question keeps coming up, I’ll find time this summer to update it and formalize it.

For now, informally, five observations:

The LODES Workplace Area file is a lightly-scrambled-eggs version of QCEW. QCEW and “wage detail” payroll files are the frame/universe for it. The vast majority of jobs are represented. But the employment that is outside of QCEW frame is also not in LODES (self-employed workers, gig workers, family farms, certain industries not covered by UI, etc.).
The most-complained-about aspect of QCEW (and thus LODES WAC) is the accuracy of the worksite location. The state workforce agencies that collect QCEW and “wage detail” payroll files ask for the physical location of work. But what they sometimes get is the address of the payroll/HR office of a company – and this may or may not be the workplace. This is a particular problem of multi-worksite establishments. And also a problem of industries where workers do not have a traditional, fixed worksite (NAICS 23 and 56).
Location perturbation: To protect establishment reporters’ “privacy” and to obscure the actual location of establishments, Census has an algorithm to push or pull jobs away from their actual location. Jobs can end up being reported in adjacent or nearby block groups or zones. However, I do not find that this perturbation compromises the big-picture view of spatial distributions; really the spatial distribution in LODES (or at least Minnesota’s LODES files) is mostly satisfactory.
Location imputation for worksites that could not be geolocated: I think this is a more serious problem than #3 above. In geocoding business locations, there’s always a share that need some human-analyst-intervention, some knowledge of locally weird addressing. Census Bureau does not have the local knowledge. Ultimately the worksites that fail initial geolocation will be given imputed locations. And the imputed location could be substantially distant for the actual location. I suspect (but I don’t have proof) that the types of addresses that fail geolocation (Rural routes? Or business-campus-internal addresses confused as street addresses?) have a different spatial pattern than those successfully geolocated. And so that will introduce some systematic undercounting some places (overcounting in other places).
Final observation concerns the LODES O-D file. It’s been our practice to interpret the O-D file as describing a commute. But that’s an assumption. Even before the present pandemic, there were large numbers of workers who were attached to employers in a payroll relationship, but who work remotely – maybe in another state entirely – and may only very infrequently ever visit the employer’s “physical address”. LODES is blind to this; the state workforce agencies don’t capture that information. You’re only going to get that thru a TBI, or ACS, or something similar.