Below please find the notes that Nandu took at the meeting earlier this week. These topics are important because: 1. The long form in Census 2000 may be the LAST one. Instead of the long form, the Census Bureau is testing a continuous survey (some surveys each month, on an on-going basis), called the American Community Survey (ACS). However, there are still many Congressional concerns about the ACS and the content, sample size, and cost. 2. Although you very likely understand the value of the CTPP tabulations, most of you have never used the Census Public Use Microdata Sample (PUMS). Some of the work on TRANSIMS (the microsimulation system) that is aiming to replace (?) the 4-step travel demand modeling system uses PUMS, so it is important to keep this file as useable as possible for transportation applications. We have asked Jim Ryan of FTA to write an article for the next issue of the CTPP Status Report about this process in TRANSIMS. ―--------------------------------- Notes from Nandu Srinivasan on APDU 2000 Annual Meeting Adjusted Counts: The Census Bureau is conducting a coverage evaluation program in 11,000 blocks across the country. In February 2001, the Census Bureau will decide whether an adjustment of counts is needed for the PL-94-171 file. If they decide to adjust, only the population and not the housing units will be adjusted in the 100 percent data releases. This is because the coverage evaluation program only counts people, not housing units. However, by the time SF3 comes along, the Census Bureau will adjust the household counts too. PUMS for 2000 Louisa Miller of the Census Bureau, Population Division gave a talk on PUMS 2000. Louisa said that the scheme she presented is not final, but a final proposal will soon be made. We should respond to her soon if we need any changes to this scheme. I asked for a copy of a research document the Census Bureau will be soon putting out on the disclosure issue. After the May 22, 2000 PUMS Meeting, and the Letter-Writing campaign, Louisa reported that there was an overwhelming preference for the following option on PUMS. 1. The five percent State-Level File; and a 2. One percent National Characteristics File. PUMAs will be delineated through input gathered from the State Data Centers (SDCs). State PUMAs will contain at least 100,000 people. PUMAs would be constituted completely within State boundaries. Super PUMAs of 400,000 people would be defined for the National Characteristics file, and these will fall strictly within State Boundaries. Records in the National file will include a State PUMA variable. Content: Census Bureau (CB) is planning a minimum population threshold of 10,000 people (nationwide) for identification of groups within most categorical variables. The CB will use weighted counts to determine the number of people nationwide, and not number of individual long form responses. For example, if Asian Indians were determined to be more than 10,000 nationwide, then "Asian Indian" will be identified as a separate ancestry group in every PUMA. CB is considering post-processing for state files: i.e. they will first select all the records to report in PUMS, examine the characteristics in the data, and then decide: a. What variables will be collapsed/categorized? b. How will the variables be collapsed/categorized? This comment evoked response from Joe Salvo who said that such a process can result in delay of the final product. Current Census Bureau Proposal for Specific Variables: The dollar amounts for income types, utility costs, mortgage costs, rent, condominium fees, hazard insurance costs, and mobile home fees will be rounded as follows: Money $1-$7 =$5 $8-$999= nearest $10 $1000-$49,999= nearest $100 $50,000 or more= nearest $1000 Age Single year categories for 0-89 with topcoding of 90 and above. Year of entry Year of entry for foreign born (in both national and state files): Bottom coded, corresponding to topcoding for age detail. They did not say what the bottom code would be, but I assume they will use the 10,000 limit. Country of Origin Louisa did not mention this variable specifically, but she said that Census Bureau will use the 10,000 people nationwide rule for including specific countries (she illustrated by saying that if there were 10,000 Lithuanians nationwide, then "Lithuanian" would be used as a category). Race and Hispanic Origin: CB will go by a pre-defined list of 63 categories for the national file. For the state file, the additional criteria of 10,000 people nationwide will be used. Household Size: The CB proposal is to omit geographic detail of State PUMA from records of households with 10 or more persons. Only the State name will be on the record. This statement evoked the most concern from almost everyone. To prevent individual disclosure, attendees preferred that the geographic detail of State PUMA be retained, but the age of the respondent be rounded. Joe Salvo said such households are usually concentrated in New York, and the state PUMA identifier is very important. One difference between 2000 and 1990 Census is that in 1990 every household that had 10 or more people was flagged and revisited to ascertain if they were group quarters or not. In 2000, the Census Bureau used its Master Address File (MAF) to designate which households were group quarters and which were not. Since the CB did not perform field checks to verify households with 10 or more people, the assignment could be wrong. Some of these are probably group quarters. Travel Time: Travel time will be treated as a continuous variable with standard topcoding, but no other collapsing or rounding. Louisa did not know the details, but I said USDOT asked for 1 minute increments up to the top code which was 90 minutes using 1990 data. Departure Time: They showed the following scheme: 2400-0259: 30 minute intervals. 0300-0459: 10 minute intervals. 0500-1059: 5 minute intervals. 1100-2359: 10 minute intervals. Several attendees felt (Patty Becker talked at length on this) that the real disclosure issue was the PL-94-171 file since there will be 63 categories at the block level. Some felt that the Census Bureau has to exercise more control on the "aggregate cells" in tabulations at a lower level of geography rather than cutting down on PUMS. American Community Survey (ACS) Census Bureau Views Cost Issues Charles (Chip) Alexander presented ACS in the Wednesday morning session. He said the Census Bureau is working out the cost of the ACS. Chip asserted that CB cost projections show that ACS cost will range somewhere near decennial costs based on what it cost them for the 31 sites in 1990. However, he did not provide any supporting materials to substantiate this projection. Local Update of Census Addresses (LUCA) and Viability of ACS Chip Alexander also mentioned that there will be a continuous LUCA program implemented by the geography division, and that item response in ACS was just as good as the long form for 1990. Incidentally, Ken Prewitt repeatedly stressed this point at both the SDC and the APDU meetings, claiming that item response effectively proves the viability of ACS. However, Chip Walker (staff from Rep. Dan Miller's, (R.-F.L.) Census sub-committee) said that ACS is in its "test" stage. Chip Alexander also briefly touched on the residence issues of ACS. He said that the ACS uses the "current residence" rule implying people are counted where they are found. View from the Hill Chip Walker said in the "View from the Hill" session that Rep. Dan Miller is opposed to having a long form in 2010. The issues Congress wants to thrash out with ACS are primarily those of privacy, and the "mandatory to answer" issue. Both the Congressional personnel that addressed the "View from the Hill" (Tuesday afternoon) session repeatedly pointed out that the transportation questions elicit the maximum concern from the public. Republican Side: Chip said that all the questions will need to be revisited to determine legal requirements for including specific questions in the ACS. He also said they will work with Kathy Wallman's OMB interagency committee to review the wording of the questions, so that they do not "appear to violate privacy". Chip, however, said in the end (to me) that the transportation questions (especially the "Time of Departure" question) elicited some negative response from only from the rural and non-metro areas. People in metro areas did not complain as much. Democratic Side: David McMillen (staff of Rep. Henry Waxman, (D.-C.A.)) said the Congress wants to know the history and genesis of each question in the Census. He said the Census is primarily a political tool, and that it is therefore obvious that it is a politically contentious issue. He said, any question other than the constitutional requirements need to be justified as to: a. Why are they asked? b. Why should they be asked in ACS and not elsewhere? Eg: Why are they asking transportation questions? David continued to talk on how the executive and legislative wings fight over "who has the real power" in the US, and talked at length on history. Census Advisory Board View Joe Salvo talked about the importance of evaluating the ACS data. He stressed that since ACS is adjusted to "current population estimates" of the Census Bureau, the numbers in ACS are only as good as the population numbers. Secondly, the ACS sample is taken from the Master Address File. So, several people can be left out of the sample. Even in the decennial LUCA process, several jurisdictions did not participate. This can adversely affect the uniformity and quality of the data. The other big issue with the ACS (per Joe Salvo) is that only 1 in three of the non-respondents will be followed up. This issue coupled with undercount issues makes the estimates a bit uneven. He said that he was only going to talk about ACS data at a tract level since going below that is "scary". Charles Alexander said that the "confidence intervals" for ACS were larger than the long form. If the long form's estimates are in the range of 10 + 2 percent, the ACS ranges are somewhere at 10 + 2.6 percent. Joe said he had examined the data for Rockville county in NY and the response rates varied between an abysmal 25% to 80% in just the 40 tracts where the survey was conducted. In the end Joe asked the question, "Can the ACS produce good small area estimates?" Notwithstanding all the problems, he felt that five year aggregates produce good tract level data. However, ACS is still in the experimental stage. Joe emphasized that local demographers in the 31 sites need to examine ACS data for their areas using their local knowledge. Local User's Views: Thabet Zakaria of DVRPC pointed out that he cannot use ACS estimates for three reasons. 1. ACS sample size was not enough every year. 2. Local politicians consider anything but the decennial census to be a unauthentic, and will not use those numbers to base "any decision" relating to money and spending. 3. From his 20 year experience in using Census Data, he feels that the decennial Census data are relevant and useable data for at least 10 years. He said all his work depended on the long form, and the ACS is not going to make things easy for him. Charles Alexander replied that confidence intervals in ACS are not that bad, and that things do change rapidly in some areas. For example, in places like Clark County, Nevada, significant population and demographic changes have occurred in the past decade both at lower levels of geography, and at a county level. The ACS is designed to measure such changes, along with measuring everything else that the decennial long form does.