Hello, I would like to offer two additional "checks" that could be done, but with the caveat that while the following is not perfect by any means (and in some cases could be potentially misleading, if the subsequent analysis is faulty), they could lead to a better understanding of the Big Picture accuracy/usefulness of CTPP flows in specific situations. The first is to prepare dot density maps. Here is an example: -- Identify all tracts (or all TAZs) that compose the CBD for a moderate-to-large city. -- Create a database that identifies the total number of workers living in a tract (or TAZ) who have their place of work recorded as being in the CBD (i.e., in one of the tracts or TAZs that compose the CBD). -- Prepare a dot density map and visually examine the results to see if they make sense. I would think such a visual analysis could help identify Big Picture situations where the ACS geocoding effort may have consistently identified the wrong workplace location for a large number of records. But what if nothing stands out as particularly odd to those with considerable familiarity about the region? It doesn't seem to be done very often, but for those with access to an employer database for a specific work location (i.e., a database with lots of location-specific workers), in which each worker's home location is known (at least their zip code, or maybe even just their place of residence), it could be interesting to prepare and compare two different dot density maps: one based on the employer database, the other based on the CTPP. There will most likely be LOTs and LOTs of small differences in the spatial distribution of home locations (plus for that matter, limitations associated with use of a five-year accumulation of ACS records that are compared against a "current" database of employee's homes), but that is expected from any analysis based on small samples. The comparative visual checks of interest would be to see if there are any big differences that result in a "say what?" kind of reaction, e.g., if the employer database shows that 25% of all workers have residences located in the quadrant northeast of the workplace, but the CTPP flows indicate a distribution of 10% or 40%, that is worth some additional exploration. It would also be interesting to prepare a dot density map based on LEHD flows, to see how that compares. This email is already getting into too many nuances that makes this appear a lot more complicated than it really needs to be, but other sources that might be interesting to explore include the flow data available for purchase from private vendors and use of license plate surveys to identify the "home zip code locations" for cars parked in downtown lots/garages. Another check is perhaps a bit too much on the "exploratory research" side for most people's tastes, but it would be interesting to use a traffic assignment model to "assign" the TAZ-to-TAZ CTPP flows (for auto trips) to a metropolitan area road network, and compare the assigned volumes to AM peak period TOD counts. The reason to compare to AM peak counts, rather than daily counts, is that most home to work trips take place in the AM peak period, plus most trips in the AM peak period trips are commute trips. Certainly not a perfect test, but if the assigned CTPP link volumes are wayyy off from the AM peak counts, that could raise some eyebrows. Best wishes, Ken Cervenka FTA Office of Planning and Environment From: ctpp-news-bounces(a)chrispy.net [mailto:ctpp-news-bounces@chrispy.net] On Behalf Of Murakami, Elaine (FHWA) Sent: Monday, March 31, 2014 4:52 PM To: ctpp-news(a)chrispy.net Subject: Re: [CTPP] tract level data Great timing! Glad to hear that you are looking at the CTPP Part 3 data at the tract level. Recently, Liang Long and I have been looking at some flow pairs at the tract level, but you all (out there in the real world) have local knowledge to best evaluate the results. Wendall Cox of Demographia recently identified a problem in Los Angeles, which is the Port of Los Angeles, next to the Port of Long Beach, showing very large worker flows using transit for a pair that is about 25 miles distance. We have asked Yong Ping Zhang at SCAG to help us figure out what might be going on. Also, we have heard from the Baltimore Metropolitan Council that there appear to be miscoded workplaces between Baltimore City and Baltimore County, for which they have initiated discussions with the Census Bureau on some training for ACS field interviewers. Also, it may be that more responses using the Internet for ACS will ameliorate some of these errors. My advice: As a first step, pull the Part 2 (workplace data) at the tract level and find all your tracts with very high worker counts. For some regions, this might be census tracts with more than 10,000 workers. In other metro areas, this might be tracts with more than 5000 workers. Look at the means of transportation for these tracts and see if they are reasonable using your own professional judgement. You can compare the results to the CTPP 2000, which will be easier in areas where the tract boundaries have been stable. After you have determined that the Part 2 data appear reasonable, THEN look at the Part 3 flows. Yes, the Margins of Error at the tract level will be high as the sample size of the ACS is small, about 50% the sample size of the Census 2000 long form, even after 5 years of accumulated ACS records. You will notice that often the MOE is a value between 110 - 130, no matter what the estimate is. This is probably because the estimate is based on 1,2 or 3 unweighted records. While the CTPP does not include the unweighted number of workers at workplace location, there is a residence-based table A101106 of the unweighted sample count of PERSONS, which will give you a sense of how small the sample of WORKERS might be when workers are distributed to all the workplace locations. You should also look at TAD to TAD flows. We asked the local transportation agencies to define TADs (about the size of 4 or 5 census tracts) because this is reduce the MOEs. If you see a big problem in your Part 2 data, please contact us-either Penelope Weinberger at AASHTO pweinberger@aashto.org<mailto:pweinberger@aashto.org> , Liang Long at Cambridge Systematics llong@camsys.com<mailto:llong@camsys.com> , or me. Currently, Cambridge Systematics is conducting a CTPP Usability study for AASHTO's CTPP program, so everyone's feedback is welcome. Elaine Murakami FHWA Office of Planning 206-220-4460 (in Seattle) From: ctpp-news-bounces@chrispy.net<mailto:ctpp-news-bounces@chrispy.net> [mailto:ctpp-news-bounces@chrispy.net] On Behalf Of Harun Rashid Sent: Monday, March 31, 2014 11:09 AM To: ctpp-news@chrispy.net<mailto:ctpp-news@chrispy.net> Subject: [CTPP] tract level data CTPPers, have any of you done flow analyses at tract level? With each CTPP release, we conduct that for small planning areas. For this, we need tract-tract flows with numbers from A302100 table. In identifying commuter flows to downtown urban core of the region (Charleston MSA, South Carolina), we found very large MOEs for each tract-tract pair. Of a total of 940 tract-tract pairs identified to have flows to the study area, 622 have MOEs larger than estimates! I am checking to see if anyone else experienced the same issue, and how this is being handled. I was thinking of using some sort of filters, e.g. considering only the pairs that have estimates larger than MOEs. Any suggestions will be appreciated. TIA. Harun Harun Rashid, AICP Senior GIS Planner BCD Council of Governments 1362 McMillan Avenue, Suite 100 North Charleston, SC 29405 T: 843.529.0400 F: 843.529.0305 www.bcdcog.org<http://www.bcdcog.org>