Hi Ken: I read your excellent analysis of Census 2000 Part III tract-level, county-level and state-level data sets. Your findings in the detailed spreadsheet for 12 counties in Texas are not unique to Texas. According to my evaluation of CTPP 2000 data for the Delaware Valley Region, the errors and completeness of census data stem mainly from rounding and disclosure threshold rules. The 2000 rounding and disclosure threshold rules resulted in 3 and 62 percent loss in the worker flow at the TAZ level, respectively. As you know, Table 3-06, which shows TAZ-to-TAZ worker flow by means of transportation to work became totally useless due to suppressed flow of workers. It is simply not possible to have synthesized Part III dataset at the TAZ level. We tried to develop a rational and convenient method to do this but we could not and thus were unable to use TAZ-to-TAZ data by means of transportation to work. Relaxation of the disclosure rules would not solve the problem. If the 2000 disclosure threshold were decreased from 3 unweighted workers to 1, the damage to the TAZ-to-TAZ worker flow would be significant still and the data would be useless. Based on this evaluation, I recommended last year not to use any disclosure threshold in the future. There was no disclosure threshold in the 1990 census and no one complained because of disclosure. Imputation, swapping and data rounding were used in the 1990 census to protect confidentiality and were sufficient to avoid disclosure. The "super TAZ" concept will not work in the northeast states, as well as Michigan, Wisconsin and Minnesota because these states have many small minor civil divisions, which consist of only one TAZ. Such small municipalities are interested in their own data. Also, data from "super TAZs" will not be useful for planning small transportation projects such as subway stations, highway interchanges, park and ride lots and local bus routes. Finally, I should state that the margin of error in the ACS tabulation will be larger than that found in Census 2000. The CB is saying now that the 2005 ACS will provide data for areas of 65,000 population and larger because the sample size was increased from that of 2004. Yet the 2005 ACS sample was not large enough to produce a table for all means of transportation to work and a table for workers by place of work and by industrial sector. This information was suppressed completely for Gloucester County, NJ (277,000 population), not 65,000 as promised by the CB. I know that Ed Christopher, Elaine Murakami, Jonette Kreideweis, Dave Clawson, Alan Pisarski, Nanda Srinivansan and others are aware of these CTPP issues. Hopefully, they will be able to resolve these problems and produce accurate CTPP data in the future. Thabet Zakaria Deputy Director, Technical Services Delaware Valley Regional Planning Commission Philadelphia, PA 19106 Phone: 215-238-2885 Email: tzakaria(a)dvrpc.org Fax: 215-592-9125 -----Original Message----- From: ctpp-news-bounces(a)chrispy.net [mailto:ctpp-news-bounces@chrispy.net] On Behalf Of Ken Cervenka Sent: Tuesday, October 03, 2006 2:17 PM To: ctpp-news maillist Subject: RE: [CTPP] CTPP Discussion Issues Ed, Thanks for the additional info. The issues raised in the draft email I was working on relate to what you just said: 1) the idea of larger TAZs to limit disclosure problems; and 2) the idea of data synthesis to still get very small-area data. So my apologies for redundancy in addressing the same thing below, but in different words. ********************************************************************* Ed said at the end of his email to post questions/comments to the listserv, so... I do indeed hope AASHTO will approve the new CTPP Pooled Fund initiative. In spite of the difficulties "us MPOs" and other organizations have with uncertainties on the accuracy of Census products (and the ACS and CTPP 2000 comparisons people have been making are pointing out some important issues), and the impacts of disclosure rules applied to journey-to-work tables, the "long form" questions are nevertheless still important to the transportation planning process. As others before me have said (I recall Alan Pisarski giving a speech or two or three on this topic), the word "data" is not particularly glamorous to people outside the profession (and oftentimes rather boring to those within the profession), but yet represents the foundation to an informed (and democratic) decision-making process. So with those sincere accolades on the underlying importance of CTPP data--and hopes for AASHTO pooled funding--stated, now on to the technical issues. The current disclosure rules have been a major hindrance to "good" small-area transportation planning in which we want more than simply place-of-residence and place-of-work information: now I understand why the Census Bureau must always retain procedures that maintain confidentiality of individuals, but I wonder if there isn't more to be checked and confirmed, just to be sure that the "rules" in place are not being too restrictive. The impacts of the restrictions have been discussed many times before, but can be easily seen by working with the Census 2000 Part III tract-level, county-level, and state-level datasets available from: http://www.transtats.bts.gov (note: when you are at this location, just type in CTPP 2000 Part III in the "search this site" box to get to the appropriate downloading page). The attached spreadsheet shows some statistics for 10 different states, in which I summarized the Tract-to-Tract data in the State, for selected variables. Since not everyone's email is set up to receive attachments, here is the gist of what's shown on the spreadsheet: -- Question T301C1 represents total workers, which for these 10 states is 56,157,132 workers. -- Question T302C1_1 represents all workers residing in households that are definable by their "vehicles available in household" and seven "Means of Transportation" aggregated categories; the 10-state total is 55,643,556 workers, which is equal to 99.1% of the total workers (I presume the very minor loss in workers is due to the "residing in households" definition). -- Question T305C1 represents all workers residing in households that are definable by a household income category; the 10-state total is 65.7% of the total workers. -- Question T306C1 represents all workers residing in households that are definable by one of the 17 "Means of Transportation" detailed categories; the 10-state total is 66.4% of the total workers. Of equal significance are the variations of "completeness" for each of these 10 selected states, e.g., for Question T306C1 there is a range of 56.0% for New York to 79.8% for Oregon. When confronted with these "missing data" issues, I suspect some planners simply factor up the available trip table so it appears to represent the full universe--but even a cursory examination of county-level Part III trip tables versus "Part III tract-level aggregations to county" trip tables will show that the missing data is not equally represented across all categories. For example, consider the 17 "Means of Transportation" data for Dallas County: the tract-level aggregations represent 60% of the reported county-level total for drive alone trips; 80% of the total for walk trips; and 100% of the total for work-at-home trips. So if someone used a simple factoring process across all modes in the tract-level datasets to reach the full universe, the end result is that the actual number of drive alone trips would be significantly under-represented and the number of walk trips and work-at-home trips would be significantly over-represented in the factored tract-to-tract data. [Note: if anyone is interested, I have a detailed spreadsheet that shows these issues for 12 counties in Texas--but it is a rather messy spreadsheet that has not been thoroughly checked, so I don't want to forward to people unless they ask me via private email, and promise not to forward to a listserv]. Since most TAZs (at least as defined by the MPOs) are generally a LOT smaller than Census Tracts, the TAZ-to-TAZ trip tables from the CTPP 2000 effort wind up having even greater amounts of missing data for the income-related and "detailed means of transportation" categories. Which brings up two possibilities for future improvement (assuming there will be Part III datasets in the future at anything other than a County or PUMA level): 1. I wonder if there might be opportunities to someday have two sets of TAZ-to-TAZ files: one dataset would have the same "missing records" limitations we currently see (and agonize about), and the other dataset would be based on a process that uses some of the held-back information to "synthesize" the missing data to develop a fully-complete TAZ-to-TAZ table. Now independent researchers can already use various "Iterative Proportional Fitting" or similar imputation procedures to deal with missing data issues, but I am concerned that their techniques will not be as good as what the Census Bureau staff could do, since the CB staff would have access to more of the underlying raw data that would (or at least could/should) result in a "better" synthesis process. Now I can understand the cost implications and potential confusions of two sets of "official" TAZ-to-TAZ data, but this would give us end-user planners the greatest number of choices for making decisions about future land use and capacity changes. Plus, by making the "missing records" tables available, this would still give agencies an opportunity to conduct (at their own expense) their own imputation process that could then be compared against the CB's synthesized tables. 2. If it is simply not possible to have a synthesized Part III dataset at the TAZ level--and there is not much hope for any significant relaxation of the disclosure rules--then maybe we need to pay great attention to the "Super TAZ" concept, in which each TAZ is carefully defined so that it is "just large enough" to reduce the more serious confidentiality issues, while still being useful for planning. An example would be a typical CBD: this is MUCH smaller than a County or PUMA, but is often composed of many existing TAZs and Census Tracts. So instead of having a CBD represented by 25 TAZs, maybe it could be divided into just four or five TAZs--or even just one TAZ, if that's what it takes to keep the understandable confidentiality issues from over-taking our desires for a more democratic decision-making process that is based on understanding the world we live in. Ken Cervenka North Central Texas COG (Dallas-Fort Worth MPO) Kcervenka(a)nctcog.org ************************************************* -----Original Message----- From: ctpp-news-bounces(a)chrispy.net [mailto:ctpp-news-bounces@chrispy.net] On Behalf Of ed christopher Sent: Tuesday, October 03, 2006 12:35 PM Cc: ctpp-news maillist Subject: Re: [CTPP] CTPP Discussion Issues Ken--the short answer is yes; Part 1, part 2 and Part 3 (flow data). With that said there will be some caveats and of course of lots of issues to be discussed, debated, researched and decided upon. In hearing from users and those active on the AASHTO SCOP Data Working Group people have been very clear that they want small area TAZ to TAZ flow data. Under the proposed pooled-fund, that would happen but certain concessions would have to be made to accommodate the Census Bureau's Disclosure Review Board concerns. One very practical notion on the table is to have a larger geography (call it a super TAZ for now) where we would have all our data with no suppression or other disclosure rules. From that we would extend a more traditional data package for the smallest zones (traditional TAZs) that would likely be synthesized. Some spot research is going on in this vein and the proposed pooled-fund calls for more as well as an NCHRP 8-36 proposal that we are hopeful will get funded. For now before we get too far a field I think it is critical to see that the pooled-fund comes to fruition. Ken Cervenka wrote: TAZ-to-TAZ update. data Census been pooled-fund the Questions the geography. TAZ. early then unique geographical was CB file Not focus result, tables for before subscribers the