The final workshop session that was focused on a particular sector of the American Community Survey (ACS) user base was designed to cover a range of perspectives on business-related uses. This topic includes the direct use of ACS analysis by private-sector firms for planning and implementation, the use of data by economic development authorities to promote business expansion and recruit new business opportunities, and the recoding and repackaging of ACS data by “data aggregators” to make the data easier to use for downstream users. Despite its origin in and widespread use in the public sector, the ACS has become a heavily relied upon resource in the business community. As hinted at in the discussion period for this session, the relationship between the private sector and the ACS is an increasingly important one because of two arguments that have been stirred up on the debate over the future of the ACS: first, whether the ACS is duplicative of information that could be obtained from other (including business-generated) sources and, second, whether—if businesses benefit strongly from the availability of ACS to make decisions—those businesses should pay more for the ACS’s collection and maintenance.
Section 6–A summarizes the use of ACS data in the economic development and workforce community, through the specific example of a development group in Iowa. Section 6–B gives an overview of the experiences of one data aggregator—the Integrated Public Use Microdata Series (IPUMS) project housed at the University of Minnesota—and describes what IPUMS staff know about the users of their versions of ACS files. Three specific business examples
of ACS use are then discussed in turn—Acxiom’s use of ACS data in marketing and business information management services (Section 6–C), the Conference Board’s use of the data to study trends in teleworking and wage inequality (Section 6–D), and AIR Worldwide’s incorporation of the data into their models of risk exposure to natural and man-made catastrophes (Section 6–E). David Crowe, chief economist for the National Association of Home Builders, moderated the discussion block for the session, starting with brief remarks on the ACS’s role in the housing community (Section 6–F).
Described by Andrew Conrad as having “a very academic name for a very practical organization,” the Institute for Decision Making is an economic development organization affiliated with the University of Northern Iowa (UNI). In his remarks, Conrad briefly introduced five projects (or types of projects) conducted by the institute and its partner organization that make particular use of ACS data.
He said that the first of these—regional workforce analysis—had not really been broached in the workshop’s previous sessions, but it relies “tremendously” upon ACS data. He said that he assisted with workforce analyses completed by and for three counties: the Texoma region of southeast Oklahoma and northeast Texas; the Cedar Valley region of northern Iowa; and the Siouxland region spanning parts of western Iowa, Nebraska, and South Dakota. These projects provide businesses and employers in the region in question with a picture of the characteristics of the area’s workforce and estimates supplies and demands for specific worker types and talents, and they make extensive use of ACS data as well as employment data maintained by the Bureau of Labor Statistics (BLS). In fact, he said, he knew a few people in the workshop room had gone through training that he conducted at the Census Bureau and that training called for them to use these three regions. As a result, the training attendees “probably know more about these regions than they want to” because the training makes them work with the BLS and American FactFinder websites (“and see what it is really like for us who are data users”).
Conrad said that these projects faced a major challenge in that a couple of them started before the first ACS 5-year numbers were released. Consistent with what Miller said about the availability of ACS data for rural areas, 64 of Iowa’s 99 counties are under 20,000 population and so have no ACS estimates except for the 5-year numbers; indeed, two-thirds (667 of 950) of Iowa’s incorporated places are under 1,000 population. In these projects, Conrad said that they use ACS data for basic demographic (population by age, sex, and race and Hispanic origin) and economic (household income and poverty status) informa-
tion. However, examinations of the prevailing workforce also depend critically on ACS data on transportation and travel flows (commute time and means of transportation to work), and on the ACS’s ability to profile employment earnings by educational attainment. In rural areas, the “laborshed” from which employers may draw workers can be geographically large; hence, these studies typically work with data at the county (or groups of counties) and state level—with some reference to the nation as a whole.
From the broad regional outlook to tract-level studies, Conrad said that the second type of projects for which his organization uses ACS data are neighborhood economic development technical assistance projects. In particular, he referenced work done with the Waterloo Neighborhood Economic Development Corporation (WNEDC) in his own home community (Waterloo being a neighbor city of Cedar Falls, home of UNI). Waterloo contains some of the most diverse neighborhoods in Iowa, defining “diverse” perhaps less stringently than in very large urban centers as census tracts containing about one-fourth (or greater) minority population. The new diversity of many of these neighborhoods is a fairly recent phenomenon, based on an influx in immigrants since 2000. Accordingly, Conrad said that the arrival of ACS data has been welcomed by groups like the WNEDC.
As with the general regional workforce projects, Conrad said that neighborhood-focused economic development efforts rely on the ACS for basic demographic, income/poverty, and industry/employment data at the census tract and, depending on city size, block group. But ACS items of particular interest for this analysis reflect the interests of the client audience. Neighborhood economic development projects are used to inform prospective entrepreneurs, so one very important metric derived from ACS data is an indicator of how many entrepreneurs are already at work in each small area—that is, the number and share of people who are self-employed. Because self-employment can occur either by choice or by need, Conrad said that additional analysis is needed to better understand the context of self-employment. These data are also used to recruit prospective businesses, and those clients tend to be interested in potential employment pools (daytime population in neighborhood areas) and household characteristics by neighborhood (e.g., what type of income might be tapped from a retail standpoint?).
The third project type Conrad described in which ACS data are used in economic development studies are asset mapping studies, at the state or regional level. He raised as an example a project done principally with two Iowa state agencies—Iowa Workforce Development (the state employment agency) and the Iowa Economic Development Authority—along with affiliated regional and local economic development authorities. This project attempted to take a high-level view of the regional economic development terrain in the state—constructing, for each of 16 state-defined marketing regions, asset maps to help assess the comparative strengths and weaknesses of the regions in terms of
strategic workforce assets. He said that these asset maps made use of a host of ACS data in addition to the standard demographic and income/poverty breakdowns—data on travel/commute time, educational attainment, health insurance and costs, and prevailing housing values and rents. The asset map documents made use of these and other variables at the state and county level, along with a U.S. total benchmark. Singling out the educational attainment variable, Conrad said that ACS data have been useful in documenting and understanding a phenomenon common to many rural areas: high school graduation rates that can actually be very strong, but vastly smaller numbers and shares of individuals with any post-secondary education (including 2-year as well as 4-year programs). Conrad also noted that ACS data involving health insurance information can be useful as a proxy for employment benefits offered in a particular area.
A fourth type of economic development study using ACS data is the derivation and comparison of regional metrics—with a particular eye toward understanding change over time (or the lack thereof) rather than a pure snapshot at a time period. He used as an example an analysis that his institute conducted with the Iowa City Area Development Group. A public/private partnership, this group wanted to better understand the drivers of economic growth in the Iowa City area; Conrad said that the area has grown due to the presence of the University of Iowa and its affiliated hospitals as well as expansion of private-sector firms like ACT, Inc., Proctor and Gamble, and Rockwell Collins (the latter headquartered in nearby Cedar Rapids). In particular, they were interested in using data both to establish a benchmark (current economic development) and to project and estimate future economic activity and the possible impact of new development strategies. ACS data used in developing this set of metrics included earnings by educational attainment, travel/commute time and means of transportation to work, and income/poverty status, as well as basic demographic splits. For use as a benchmark, this work involved use of ACS estimates at the place (city/town), county, state, and national levels; as Conrad said, “Iowa City wants to know what other metro markets are their competitors,” nearby and nationwide, and regional economic development bodies generally have a “hit list” of other localities that they compete against (and against which they need to be compared).
As a final example, Conrad turned to a pure state-level measure—the development of an Iowa Competitiveness Index. Unlike some other states, Iowa does not have a statewide chamber of commerce; instead, the Iowa Business Council brings together representatives from about 20 of the largest employers in Iowa (including firms like John Deere and Company, Pella Corporation, Rockwell Collins, and the major state universities). As Conrad put it, the Business Council’s role is not to lobby but still to “push issues”—as it has for a number of years, for instance, on the status of pre-kindergarten education and services. Like the Iowa City example of regional metrics, the Business Council sought a way to benchmark the state economy and track progress over the years. Accordingly,
in 2011, the Council published the first iteration of the Iowa Competitiveness Index, a set of key indicators that are scored and combined to produce general assessments of the state’s economic strength along five broad dimensions: economic growth, education and workforce readiness, governance and fiscal matters, health and well-being, and workforce demographics and diversity. The index has been updated for 2012, and the Council intends to maintain it as a regular series. Several of the indicators—the levels of which are compared and ranked across all the states and then scored by their “overall competitiveness trend” (improving, no significant progress, or worsening)—are derived from a variety of sources, including data from the Iowa Department of Education (by way of the National Center for Education Statistics) and the Iowa-based ACT, Inc., college testing service. However, several of the indicators are drawn from the ACS, such as per capita income and measures of racial and ethnic diversity.
In closing, Conrad said that he wanted to emphasize that economic development is a process, not an event. Some elected officials mistakenly think of it as a discrete event—a particular ribbon-cutting or ground-breaking—when it is a complicated (and sometimes lengthy) process. He said that he recalled working on a specific project for many months—working with Target Corporation about 10 years ago on the placement of a 1.3 million square foot distribution center. Conrad’s Institute for Decision Making was involved in early work on the project for a long while before they even knew the specific company interested in making the move, and it continued as Target “kicked the tires in the community” before making its final decisions. But Conrad said that the main point is that economic development is not only a process but one that involves data-driven decisions at each step. Like other companies, Target had to approach the decision from a market analysis perspective—where might they be opening or expanding stores, how are they going to serve those markets, and where would the workforce for both the stores and distribution center come from. At even more detail, wanting to enter the community as an “employer of choice,” Target wanted to understand the wages and benefits that it would have to provide to be competitive. Conrad said that data analysis from the ACS (including the use of health insurance as a proxy for benefits, as described above) has proven useful in these kinds of projects.
Conrad also argued that ACS-based analysis is critical for entrepreneurs. He recalled a specific example from Waterloo, of the WNEDC working with a local Hispanic couple interested in opening a tienda in the area. That case involved detailed analysis—possible only with the ACS—of 13 census tracts, using the data to try to pick an ideal location based on the demographics of the area and the access to transportation networks (anticipating that much of their clientele would be foot or public transit traffic). As another example, Conrad said that he had worked with Pella Corporation on numerous site location projects, profiling small-area local workforces and trying to determine which areas have workers with the educational attainment and skills needed for the facility and
whether—if one built a 400-employee facility in a town of 5,000 population—housing would be there for the new workforce. Unfortunately, he said, the ACS data were not available at the time of this project.
During the discussion period for the session, Conrad was asked by his colleague in economic development work—Patrick Jankowski (Greater Houston Partnership and member of the workshop steering committee)—to comment on particular uses of ACS data to profile and promote areas to foreign corporations and foreign investors. Conrad replied that he has worked in such cases before and agreed that the ACS data are very helpful in building cases for foreign investors—with two curious caveats. First, he said that he is struck that foreign firms are “shocked at the level of detail that we are able to provide” through public data resources. It is not only the ACS that yields this response—when economic developers are attempting to weave a data-based story for investors, data resources like the Bureau of Labor Statistics’ Quarterly Census of Employment of Wages and its Occupational Employment Statistics program also play important roles. But the level of demographic detail from the ACS surprises foreign firms—which occasions the second “caveat,” which is that foreign investors will commonly engage other consultant firms to review the data work by economic developers as a check on whether the numbers are correct.
Put briefly, Conrad said that professionals in the economic development community “use the ACS data on a daily basis”—and communicate impressions and findings from those data to their end clients. To be sure, he said, ACS data are not the only important data resource; they rely on other government-produced data series such as those produced by the Bureau of Labor Statistics. But, he said, the ACS is becoming increasingly critical to making the kind of “informed decisions” that help communities and local governments cultivate entrepreneurs and businesses, and continued availability of the data is critical. He noted that other speakers had “wish lists” to make the ACS more useful and relevant; from the economic development perspective, Conrad said that the addition of a question on multiple jobs—how many jobs each person actually holds—would be invaluable in sorting out dynamics that are presently murky. Above all, though, Conrad stressed the importance of the 5-year ACS estimates to a state like Iowa and its roughly 3.1 million population—a state in which two-thirds of counties have less than 20,000 population and only 22 communities have populations sufficient to obtain 3-year average estimates.
Funded by the National Science Foundation and the National Institutes of Health and administered by the Minnesota Population Center (MPC) at the University of Minnesota, the IPUMS project produces (and makes available for
free) microdata series for research use that are recoded from the originals to use consistent labels and concepts across time (past iterations) and space (other countries).1 For the purposes of this workshop—and among U.S. data users—the term “IPUMS” is generally used to refer to products from the IPUMS-USA subproject, including harmonized, consistently coded microdata samples from all U.S. censuses since 18502 and—since 2001—the ACS.3 The other two branches under the IPUMS label are the IPUMS International subproject that enables linkages across multiple iterations of several foreign censuses and IPUMS-CPS, which applies consistent coding to multiple microdata series from the Current Population Survey (CPS) samples. Across all its data sets, MPC currently maintains on the order of 800–850 million individual person records—and it distributes them to a steadily growing user base. Katie Genadek of the IPUMS user support team said that MPC averaged slightly less than 100 gigabytes of data downloaded by users from its servers in 2001, around the time that its web distribution strategy came into place; as of 2010, the data volume moved by MPC had grown 10-fold to about 1 terabyte per week, on average. The IPUMS-USA samples, including the 1-percent microdata samples from the ACS (yearly and multiyear samples), are the most widely used of MPC’s data resources, and currently have about 30,000 registered users on MPC’s website.
Speaking for the IPUMS user support team, Genadek remarked that her workshop presentation would be different from the rest of the workshop because she has used ACS data extensively but is not actively in the business of producing anything from the ACS—“no map, no table, nothing.” But the perspective she said she would bring to the session was that of a data aggregator—repackaging and adding value to the base ACS products and brokering their use by a broader class of data users. Significantly, she noted that she can also speak to the ways in which users manipulate and analyze the IPUMS version of ACS data files.
Genadek said that the “value added” to the IPUMS version of the public ACS files runs along four basic lines—three related to data content. The first is the database’s titular integration. Even in the relatively short lifespan of the ACS, there have been changes to the exact coding of responses to particular questions, to the labels associated with specific responses on the questionnaires, and to un-
2There is one exception to IPUMS’ census coverage; most of the original schedules (the ledger-type “questionnaires” on which personal information were recorded) from the 1890 census, from which microdata would be drawn, were destroyed by fire in 1921.
3Though IPUMS is well known (and a widely used term) among data users, the preferred way that the term should be pronounced is less clear. As an aside to her presentation, Genadek remarked that the preferred in-house pronunciation is ih-pums (soft I)—consistent with the pronunciation of the word (“integrated”) providing the initial—although eye-pums (hard I) is commonly used. As she said, “we are not related to the Apple products”—and try to avoid association with them—making the pronunciation distinction more important in recent years.
derlying geographic and other codes. To the greatest extent possible, IPUMS recodes the public files to be able to facilitate more direct comparisons across years—or to make pooling of different years’ samples possible. The second component of value added is the detail of the online documentation of the data sets. The samples themselves are described with care, as are individual variables; the history of individual variables, including the comparability of data across samples and changes in the underlying questionnaire texts, is recounted in detail. The third content-related value-added component is “new” variables derived by MPC and included on the IPUMS files. These include “crosswalk” occupation and industry variables (for which consistency across year and beween samples is particularly hard to maintain because of changing definitions—and the changing nature of industries). The IPUMS-coded ACS data also include derived variables to reconstruct family and subfamily interrelationships, to permit analyses using families as the basis for analysis rather than household/housing unit. The family relationship variables are derived from various patterns in responses as well as the general census/ACS question on relationship to household head—an item that technically asks about the relationship of each person in a housing unit to “Person 1” (the one who completes the questionnaire or who is interviewed), who is intended to be but is not necessarily the effective head of household—or, for that matter, a family member of some people in the housing unit4).
Genadek said that the fourth value-added component is the online interface and data extraction system, which gives users great flexibility in selecting as many or few variables as they might like from as many samples (census or ACS) as desired. The IPUMS site has also taken care over the years to provide results in many output forms, in whichever way is most convenient for the user to access and analyze. User-specified extracts can be output in formats for common statistical analysis packages (SAS, Stata, SPSS); recently, facility was added to generate results in basic comma-separated value (CSV) format, for easy input into Microsoft Excel or other spreadsheet software. For users without access to statistical software, the MPC platform has built-in access to the Survey Documentation and Analysis (SDA) package developed by the University of California, Berkeley, Computer-Assisted Survey Methods Program. Hence, it is possible for users to directly analyze all of the IPUMS-USA (census and ACS microdata samples) through online queries. Genadek said that she is part of the IPUMS user support team that has proven to be a popular aspect of using the ACS data; she and her small team of colleagues field email and telephone inquiries about the data and using the data, with the goal of responding to all questions within three days (and, more often, within a few hours). At the workshop, she said that
4Similar to earlier versions, the main body of the 2012 ACS questionnaire begins with the parenthetical instruction: “(Person 1 is the person living or staying here in whose name this house or apartment is owned, being bought, or rented. If there is no such person, start with the name of any adult living or staying here.)”
Of all the user-specified extracts generated from IPUMS-USA—samples from the 1850–2000 decennial censuses and the 2001–2010 iterations of the ACS—some version of the ACS was included in 55 percent of the queries. Of those IPUMS requests asking for ACS samples, 137,029 have asked for one or more 1-year PUMS files; smaller but sizable numbers of extracts have asked for the more recently available 3-year (5,726 extracts) and 5-year (2,141) samples. Also a fairly recent development, the SDA-based online analysis component of the IPUMS interface has generated 250,000 user-requested tables using ACS data—13,000 of those in the month preceding the workshop alone.
Turning to the question of who uses the IPUMS-produced files, Genadek noted that academic scholars are the largest segment of IPUMS’ user base. As evidence, she displayed a graph of the number of citations to three MPC databases—including, and with trends driven principally by, IPUMS—each year from 1993 to 2011, as found in Google Scholar searches. Similar to the graph of MPC data downloads, references to IPUMS and MPC resources in the academic literature began to grow precipitously around 2002 and have increased by roughly 50 percent over the last 3 years, hitting a level of just over 1,100 citations in 2011. As Genadek indicated, the graph includes publications from non-IPUMS data maintained by MPC; however, Genadek said that she believed that the IPUMS versions of ACS products account for a large part of this growth and that the ACS “is driving a lot of this research right now.”
Using basic information from the user profiles defined on the MPC site, Genadek briefly displayed a pie chart showing the rough breakdown of IPUMS users’ backgrounds. Indeed, users with academic affiliations make up about 77 percent of the known users; the largest share of these are users affiliated with an economics department (31 percent), followed by demography/sociology (16 percent). Genadek says that MPC and IPUMS staff are aware of IPUMS data being the source for “tons of dissertations”—graduate students may be “our number one user”—and an increasing number of class assignments and projects. But other blocs of IPUMS users are significant—state and local government agencies, private industry, journalists. Genadek said that, at the federal level, IPUMS data had been accessed and used by—among others—the U.S. Government Accountability Office, the Bureau of Labor Statistics, and the Federal Reserve Board (and its member banks).
IPUMS’ online extract system is such that IPUMS staff can determine how frequently users check specific ACS variables for inclusion in their extract data sets. Genadek reported that the data item on race is “by far number one” in terms of included variables. But some ACS-specific variables (or, at least, variables that are consistently reported across small areas by the ACS) are interspersed with the standard demographic variables, suggesting that users are not just using the ACS for general population estimates. To wit, the second-most-
included variable in IPUMS extract is employment status—a major driver in the workforce and employment studies described by Conrad (Section 6–A). Person age ranks third, but place of birth and educational attainment both slip into the rankings ahead of sex and Hispanic origin. Rounding out the list of the most commonly accessed variables in IPUMS ACS, in descending order, are relationship to household head, metropolitan area, state, marital status, income, and language spoken at home.
Genadek said that she also combed through the emails to the IPUMS support team and characterized the most frequent questions in terms of both topic and technical concern. In terms of topics, she said that the IPUMS staff are most often contacted for clarity on poverty estimates and the relationship of poverty levels to the ACS microdata. Next most frequent are questions of a geographic coverage nature—why, in the microdata sample, do people appear from one specific county and not another? Questions about migration and mobility of people within the ACS sampling framework are common, as are questions about the derived variables on family and subfamily composition. Questions about occupation coding, measures of physical or mental disability, and income round out the topic-based questions. On more technical matters, IPUMS staff commonly field questions about how income variables are adjusted based on the rolling, multiyear sample design; they also receive and address questions about weighting, allocation flags, and variance estimation, as well as more general questions on the interpretation and communication of estimates from multiyear samples.
In communication with IPUMS users, Genadek commented that she also has a sense of common wishes—things that IPUMS users would really like to have included in the ACS. In her assessment, the biggest thing that she hears from users is desire for the month of the survey’s administration to be coded in the microdata. In addition to possibly permitting some glimpses at seasonal populations and very precise construction of age variables, the month of interview could also allow advanced users to apply their own income adjustments to the finance and income questions. She conceded that she was not sure exactly how much of a concern the identification of survey month would be from a confidentiality standpoint, but it remains a major part of users’ wish lists for the survey. Another common request is inclusion of a clearer and more detailed relationship question. The IPUMS coding tries to reconstruct family structures as much as possible, but the form of the relationship question and the way in which the question is phrased can obscure some family relationships depending on who answers the question.5 Also high on IPUMS users’ wish lists is the re-
5A previous National Research Council (2006:133) report, following Schwede (2003), cited a simple example of this masking phenomenon: “Consider a case where a man and woman live together, unmarried, along with the woman’s child from a previous relationship. If the man is the census respondent, the woman may be reported as an unmarried partner or an ‘other nonrelative,’ while the child would likely be ‘other nonrelative’; the biological link between woman and child is obscured. If the woman is the respondent, the biological link between her and her child would be
instatement of two variables that were available in previous census long-form samples but were dropped over time. The lifetime fertility question—number of children ever born to a mother—was dropped from the 2000 census long-form sample and has not resurfaced in the ACS. Questions on the place of birth of mother and father have been absent even longer, having been dropped from the 1980 census long form; the Census Bureau announced plans to add parental place of birth questions to the ACS questionnaire in 2013, but withdrew those plans in May 2012.6 But, Genadek said, much though IPUMS ACS users might appreciate such small revisions, her sense is that the users’ main concern and interest is for minimal change in the future. IPUMS users tend to be interested in long, historical time sequences—for that, one wants continuity in content, form of question, and sample size. Hints or rumors of content being changed in major ways or dropped together—periodically, for example, the ACS question on ancestry is thought to be vulnerable—are disturbing prospects for policy-relevant analyses of specific ancestry groups such as Brower’s profile of specific Asian subpopulations in Minnesota (Section 5–A).
Genadek closed by referencing two additional MPC products, including the National Historical Geographic Information System (NHGIS)—essentially, a mapping analogue to the extended time series across censuses and data collections that can be calculated from IPUMS data. Small-area data (down to the finest level of resolution possible in earlier years) can be extracted in mappable form from the 1790–2010 censuses and, now, from ACS 1-year and multiyear products. Like the IPUMS extraction tool, users can pull multiple tables and variables from multiple years at once, and shapefiles for use in geographic information systems software for the different time periods can be downloaded as well. The NHGIS interface also gives users access to a summary file not available from interactive tables on the Census Bureau’s American FactFinder site: a 5-year summary file from the 2006–2010 ACS that can be queried down to the block group level. She said that MPC plans to disseminate all of the ACS files as they become available and are processed—not as a replacement or competitor to FactFinder, but as an alternative for people who find FactFinder hard to navigate.
preserved, but it would be ambiguous to family and household researchers whether the male is the child’s biological father or not.”
6The Bureau posted notice of new ACS questions in the Federal Register (77 FR 18203–18205) in advance of its filing a request for three additional years of clearance from the U.S. Office of Management and Budget (OMB) under the Paperwork Reduction Act. The decision to withdraw the parental place of birth questions was communicated in a memo updating that clearance package, viewable at http://www.reginfo.gov under Information Collection Review number 201202-0607-003.
Formerly a statistician with the Census Bureau, Matthew Christenson (Acxiom Corporation) began his remarks by noting a basic difference between his former and current employers in summarizing the process of getting his slides approved for presentation. At the Census Bureau, the overriding worry was always content; with Acxiom, a marketing services company, the focus is branding. Putting the point succinctly, Christenson began by saying, “I work for a company called Acxiom”—and “after this presentation you may not be able to forget that name,” as it is “probably the most commonly used word in the presentation.” Christenson gave an overview of how ACS products factor into Acxiom’s work, but the broader point he alluded to in his opening—the comparison of a focus on branding with the identity crisis of sorts noted in the media perspectives session (Chapter 4) and other points in the workshop, in which the ACS has struggled for recognition in some circles as something other than general “census data”—would recur during the discussion period.
Headquartered in Little Rock, Arkansas, Acxiom was founded in 1969 and went public (on the NASDAQ exchange) in 1983. Christenson said that Acxiom’s roughly 6,100 employees are distributed across offices on five continents (including several U.S. locations other than the Little Rock headquarters); in fiscal year 2012, Acxiom reported just over $1.1 billion in revenue. He observed that Acxiom’s target clients are Fortune 1000 companies and counts among its clients many of the largest companies in the United States (and the world) in industries ranging from credit card issuers to airlines to telecommunications/ media. Acxiom also engages in work for several federal government agencies.
Christenson described Acxiom’s work as being three distinct but related businesses operating under the same company. The first of their constituent business is what might be called “recognition services”—helping client businesses recognize their individual customers regardless of how they make contact with the company (direct phone, email, or letter contact, or electronically through web browsers). This permits linkages to other data to be drawn when a person contacts the client, to make customer service more efficient; Christenson said that recognition services are also used as part of companies’ fraud prevention plans. The second constituent business—and the branch in which Christenson works—is marketing services, identifying potential customers and establishing contact with them. More colloquially, Christenson said, people “typically accuse” people in marketing services “of being responsible for your junk mail, for the junk emails”—but, done more correctly and precisely, he thinks of marketing services as being responsible “for everything you don’t get.” That is, the goal is a marketing outreach program allowing businesses to target and engage potential customers but without ads for retirement services being
sent to young people, for ads for Depends undergarments to be sent to households with young children, and so on. Finally, Acxiom operates an information management services businesses—which is to say that its computer servers are commonly used to host and manage information in online commerce and banking transactions, a line of work that means that the company hosts some of the largest data sets in the world.
Under Acxiom’s marketing services umbrella, Christenson outlined three key data products that make use of ACS estimates—or to which ACS estimates serve as a complement—in various ways:
- Based on an input name and address, Acxiom’s InfoBase Enhancement product returns a host of individual- and household-level data (demographic, financial, housing, and marketing propensity) that can be appended to the records for further analysis.
- InfoBase List is “our equivalent to the census,” as Christenson put it—it is as complete a roster of “economically active individuals and households” as can be derived from commercial and other records sources, and this is the data resource primarily used for “customer prospecting.”
- The Market Indices product is a distillation of demographic and economic information that can be queried by Acxiom clients by address—providing a detailed small-area look at the immediate area surrounding an address—or directly appended to a client by matching to address.
Of these, Market Indices is the most heavily dependent on ACS estimates—consisting, in present form, of over 500 preprocessed data elements from the ACS, derived for levels of geography ranging from the nation as a whole down to block group. The product is so intertwined with the ACS that (as Christenson mentioned during the closing discussion) Acxiom is now beginning to brand new releases of the product as “Market Indices ACS.” Specifically, Market Indices aggregates and assembles data from the ACS 5-year summary files; Christenson said, in candor, that this work is done with estimates from the 5-year summary files, and that Acxiom has little to no use for the 1-year or 3-year estimates. With equal candor, he commented that “we don’t throw out very much of the data” from the summary file; the data items pulled for use in the product are fairly exhaustive of the questions on the ACS—a limited number of population count variables, but many more variables expressed as percentages, rates, or measures central tendency (mean/median). In terms of the value added to the ACS files, Christenson said that what Acxiom “sells” with the Market Indices product is not so much the data themselves as “the speed [with] which we can append it to huge datasets,” on the order of hundreds of millions of records per month—approaching real-time matching and delivery.
Like the data aggregator role described by Genadek concerning IPUMS, Acxiom’s work with Market Indices depends vitally on meeting end users’ needs—in this case, the needs of Acxiom clients. However, in a pure business
environment, Christenson said that he could not really speak to how Acxiom’s clients use the data—Acxiom sells the data to their clients’ competitors, too, so clients don’t really “clue us into” their uses. But, in broad strokes, companies use Market Indices for online (and offline) targeting of marketing materials, for the selection of prospect lists, and to improve customer care. Through a better understanding of local conditions, companies use the products to devise strategies to reduce customer churn and turnover; they also use it to develop cross-sell or up-sell opportunities in marketing approaches (e.g., a customer buys one product and is told—based on local information—about other products that might be of interest, or bigger/better versions of the product).
The ties between ACS estimates and Acxiom’s InfoBase products are less direct, yet still important. ACS data are used to construct several of the “modeled elements” that go into the individual- and household-level data “appends” that comprise InfoBase Enhancement and that are available as an add-on to InfoBase List. Acxiom itself uses data from ACS 1-year PUMS files to build at least 75 of these modeled elements; still others are constructed (possibly including ACS data as well) by Acxiom’s consulting organization; distributions of the modeled elements matched to InfoBase List are also compared with distributions of related variables in the PUMS files, as a check on whether the models seem to be working properly. These modeled elements range from demographic characteristics (e.g., educational attainment, length of residence) to market relationships and behaviors (e.g., behavior as an investor, media preferences) to online behaviors (e.g., behavioral models to predict Twitter or Facebook usage).
Being matched to person-level attributes rather than just address or geographic location, the modeled elements in the InfoBase products are applied to vastly more records each month or year than the block-group-limited Market Indices file—billions of records each month, Christenson said—but the general uses clients likely make of the data are roughly the same. Though the ultimate fine-grained data (individual and household level) that can be estimated by these products are Acxiom’s stock in trade—ultimately, the items that Acxiom’s clients generally find most valuable—Christenson said that they and their clients both use ACS estimates alongside the more specific analyses; the breadth and completeness of the ACS data are important to their users. Christenson said that he also understands that Acxiom clients are increasingly importing their information into geographic information systems (GIS) packages for mapping and trade-area analysis. More generally, speaking of the product construction process, Christenson said that Acxiom finds the ACS extremely valuable in its model-building exercises; their sense is that the quality of the ACS data increases the precision of their models (and ultimately the modeled elements), and access to the data makes creating and tweaking the individual models easier. The role briefly mentioned above of distributions of PUMS variables being used as a quality check on modeled elements is also important to the company; Christenson said that some results from models for individual- and household-level
data could charitably be described as “funky,” and “the ACS helps smooth out the rough spots.” Hitting a theme raised in Chapter 4 and that would come up again in the discussion period, Christenson also said that Acxiom’s making use of the ACS imparts credibility to the products—it “carries quite a bit of weight with our customers if we say we are using Census [Bureau] data.”
Christenson closed by noting that, from his and Acxiom’s perspective, the ACS is a uniquely useful data resource because of its breadth, depth, and completeness, and it is useful both as a product that Acxiom can develop and make available to users as well as a source for its own modeling and estimation work. From a business perspective, the ready availability of annual releases of small-area data is greatly beneficial; constantly vying with competitors, it is good to feel current and up to date in estimates and projections (and not to feel like competitors have a leg up somehow).
Like other workshop speakers, he said that he would certainly like “more” from the ACS—but in a slightly different way. He said that he regularly hears from the company’s salespeople and clients who are interested in ever more granular data—frustrated that household-level (or, for that matter, individual-level) data are not readily available from the ACS. From the marketing standpoint, their clients are greatly interested in data down to ZIP+4 Codes—a cut that could be as fine as 2–3 households. The reason why these incredibly fine-grained data are not available is fairly straightforward—respondents’ privacy must be respected and personally identifiable information not disclosed—and so Christenson said that the ACS block group level is very valuable to Acxiom. But one concern that he wanted to express concerned the granularity of the categories used to calculate estimates. Until the arrival of the ACS, he said, Acxiom had been using the 2000 census-based Summary File 3—developed its products around the categories used there—and encountered significant problems when the categories used in variables like income or age ranges were made more coarse in the ACS files. Put bluntly, he said, “we don’t care about confidence intervals”—in the sense of being paralyzed by the reported standard errors. Dealing with uncertainty is an accepted part of the bargain; Acxiom has statisticians, and its clients have statisticians, to sort through those issues, and they would rather be in the position of working with finer categories and making their own conclusions about what numbers are fine to use and which are not.
Though-most people may be familiar with the Conference Board through its highly visible Consumer Confidence Index, Gad Levanon (director of macroeconomic research) emphasized that the Conference Board’s analysis and research
work spans a wide range. A membership research organization that counts hundreds of private and public corporations as members, the Conference Board conducts lines of research in economics, labor markets, human capital, and other topic areas important to businesses. Levanon noted that the ACS is a fairly recent discovery for the Conference Board but that it has already factored significantly into several projects; like many other users, awareness of the ACS data—and the availability of those data in sufficiently fine-grained form—really came about with the release of the first 5-year ACS estimates and the PUMS files. Like previous speakers, Levanon said that—prior to the ACS—his analyses depended heavily on the Current Population Survey (CPS); also like previous speakers, he has found the ACS’s sample size relative to the CPS makes a lot of things possible analytically that could not be done before. In his workshop comments, Levanon said that he would review three Conference Board projects relying on ACS data—one still in progress—as case studies.
First, he described analysis that the Conference Board has done on teleworking—working from home. The ACS question on mode of transportation to get to work includes “worked at home” as a response option,7 which permits study of people who usually telework (even though the analysis might omit people who occasionally telework). Other ACS questions and variables allow the analysis to get as close to the teleworking community as possible—including full-time workers who work for some employers and excluding those who are self-employed.8 ACS data (including the not-yet-full-scale collection between 2000 and 2005) suggest that the overall percentage of people who primarily work from home roughly doubled over the past decade, even though the percentage is very small—from roughly 1 percent in 2000 to just over 2 percent in 2010. Levanon explained that this overall percentage presents a distorted view because it includes a lot of people with occupations where the telework percentage is essentially (or necessarily) zero; to wit, elementary school teachers cannot primarily work from home by nature of the job, linked to the workplace (the school). Still, the overall percentage does suggest an escalating trend; growth seems to have been particularly rapid in the last 5 years relative to earlier in the decade.
7Person Question 31 on the 2012 ACS questionnaire asks “How did this person usually get to work LAST WEEK?” and permits the respondent to check multiple responses from the following: car, truck, or van; bus or trolley bus; streetcar or trolley car; subway or elevated; railroad; ferryboat; taxicab; motorcycle; bicycle; walked; worked at home; and other method. Answering “worked at home” routes the respondent to Question 39a, skipping over questions on commute time and unemployment/layoffs. Person Question 30 also asks “At what location did this person work LAST WEEK?,” and that location could presumably be compared with the housing unit address.
8Person Question 41 on the 2012 ACS questionnaire asks about the nature of each person’s current or most recent job activity, including two self-employed categories (depending on whether the person’s business is incorporated or not incorporated) as options; responses for different classes of private- and public-sector employment (e.g., work for a private nonprofit organization or for a state government) are also permitted.
Levanon said that the great advantage of the ACS in this analysis is the level of detail in the coding of occupations. The ACS microdata permit researchers to drill down to very detailed levels of occupations—roughly the equivalent of the 6-digit Standard Occupational Classification (SOC) codes used in other federal statistical data products. From the experience of working with the occupation variables, Levanon raised an issue for future work and clarification by the Census Bureau—working to make the occupational codes more comparable over time, because changes in the occupation coding can make it difficult to construct coherent time series. Indeed, Levanon noted, he and his Conference Board colleagues worked with IPUMS data—and its more aggressive coding to promote comparability over time (see Section 6–B)—on this project, and they ultimately wound up primarily using IPUMS’ OCC1990 variable—a recode to 389 occupational categories based on the occupation codes used in the 1990 census.
Levanon said that it is evident from looking at the work-from-home data by very detailed occupation that there are some work types for which work-from-home is very common, with potential teleworkers making up 10 percent or more of the workforce in some categories. Levanon’s tabulations show one detailed category has a work-from-home percentage that dwarfs all others: medical transcriptionists, with 44.6 percent of workers in that class reporting that they work primarily at home. Other job categories with sizable work-at-home contingents (over 10 percent) include sales engineers/sales representatives and travel agents—jobs in which employers might have incentive to save costs by reducing or eliminating office space for people who may not need to be in the office very often. Other occupational categories with fairly strong telework levels are information technology-related careers, such as web developers, computer network architects, and computer hardware engineers.
Comparing work-from-home estimates from early in the decade (averages from ACS files from 2001–2003) with those late in the decade (2008–2010 average), Levanon also observed that some occupations have experienced particularly strong growth in telework. For some categories, this suggests that networks and remote access tools have developed sufficiently over the decade to make employers more comfortable with the telework option: For instance, telework among records clerks grew from 0.9 percent (2001–2003) to 5.5 percent (2008–2010) and among travel agents from 1.8 percent to 5.9 percent. Insurance underwriters and bill/account collectors were job categories where telework more than doubled over the decade. Still other job categories with fairly strong work-from-home levels did not experience the same amount of growth, which Levanon said was likely due to the fact that improvements in technology could not or would not greatly affect the ability to work from home—these include door-to-door sales vendors (a more modest growth from 5.1 percent to 7.1 percent) and clergy and religious workers (among whom work-from-home decreased slightly, from 6.7 percent to 6.2 percent).
Levanon said that he and his colleagues have also observed interesting dif-
ferences in teleworking propensity by region. In their models of teleworking behavior, they included dummy variables to capture state-level effects (controlling for individual characteristics); classifying and mapping these state-level effects, he said the regional differences are striking. In general, teleworking rates are considerably higher in the West—likely because “hubs of technology” and places with high concentrations of the industries most likely to enable telework are on the West Coast; high levels of telework are also found in the extremes of the Eastern Seaboard (New England, Georgia, and Florida), while the lowest rates are in the Deep South and the Rust Belt.
Of this examination of telework behaviors, Levanon made clear that this is not especially complex work—indeed, he said, this is a quite “simple usage of the microdata.” But he said that he thinks that it speaks to an important base of potential data users—namely, people in the human resources field interested in trends in the workforce. Levanon said that, as best he knows, the Conference Board’s specific analysis of teleworking using the ACS data had not been done before, and he said that the range of interesting kinds of things that can be learned from ACS microdata might spur additional work—and interest in using the microdata—among parties new to the data.
The second ACS-related Conference Board project that Levanon described is one that is still in progress, but one that again makes use of the ACS’s strength in providing consistent measures across a broad range of geography. Specifically, the Conference Board is studying wage inequality, by any number of factors. As an example of the work and the ACS’s utility in it, he displayed the graph shown in Figure 6-1. The graph provides a general sense of geographic differences in wage inequality through a relatively straightforward metric; making use of ACS microdata, he calculated the ratio of the 90th percentile of total wages for fulltime workers in each state to the 10th percentile. Printed at a small size, he noted that it is hard to read, but some of the states at the top—with the largest ratios and hence the greatest spread in wages—include California, Texas, New Jersey, the District of Columbia, Georgia, and Virginia. At the bottom—smaller ratios and less spread across wages—are South Dakota, Maine, Vermont, North Dakota, Wisconsin, and Iowa. The dark vertical bar shows the ratio for the nation as a whole. He said that this fairly simple univariate slice from ACS microdata spurs questions and areas for future probing; based on the states on the high and low ends of the spectrum, “one can speculate that ethnic diversity” might be an important determinant of the inequality, and that is something that can be examined at finer levels of aggregation. The national-level vertical line—and the fact that so many states are below the line—suggests interesting trends as well; in early looks at the change in the ratios over time, Levanon said that shifts in the national-level ratio are more pronounced than are shifts in the distribution of inequality across the states. Again, he observed, these are interesting phenomena that “I don’t think can be done [using] any other data source” with such a level of confidence.
SOURCES: Calculated from American Community Survey data; adapted from workshop presentation by Gad Levanon.
Finally, Levanon described work that he had done on housing characteristics and demand in support of the Demand Institute, a new joint venture sponsored by the Conference Board and Nielsen. Given the current economic climate, one factor that the Demand Institute was interested in learning about is the phenomenon of “doubling up”—multiple families or individuals sharing the same housing unit. Within that “doubling up” population, an important subgroup is young adults continuing to live in the parental home. Levanon said that the ACS data were extremely useful to illuminate some points about this population. He briefly displayed a chart using data from 2006 and 2010 ACS files examining the percentage change in home ownership rates, which varies inversely with age group; home ownership rates dropped by almost 18 percent among 20–24-year-olds and almost 14 percent for 25–29-year-olds, compared to only a 0.5 percent change among 65–69-year-olds. Part of this drop among young age categories seems attributable to “doubling up”; as shown in Figure 6-2, almost half of 20– 24-year-olds and a quarter of 25–29-year-olds postponed independent household formation and continued to live with their parents, as measured in 2010 ACS data. Even among 35–39-year-olds, 8 percent continued to live with parents—a
SOURCES: Calculated from American Community Survey data; adapted from workshop presentation by Gad Levanon.
23 percent increase over 2006 estimates. The contrast with pre-recession numbers (2006) is particularly interesting and suggests the recession’s role in driving housing decisions. Like the other analyses, these general observations raise interesting follow-up questions that can be addressed at the state level (or other aggregations) using the ACS, and this work permits one to see whether “doubling up” is particularly concentrated in states hardest hit by the recession.
Work with the housing data also occasioned Levanon to comment favorably on the variety of questions in the ACS. Even seemingly obscure ones like the number of rooms in the home9 can yield interesting insights from the ACS data. In this case, Levanon observed that home sizes decreased during the recent housing crisis. Detached single family homes10 with 5 or less rooms ticked up slightly from about 23 percent to about 29 percent of the stock of detached single family homes; apartments with 3 rooms or less rose from about 38 percent to just short of 50 percent between 2006–2009, though the percentage dropped (and so apartment sizes grew) between 2009 and 2010. He concluded that these kinds
9Housing Question 7 on the 2012 questionnaire; part a asks about the total number of rooms and part b asks for the number of bedrooms, “count[ing] as bedrooms those rooms you would list if this house, apartment, or mobile home were for sale or rent.”
10Housing Question 1 on the 2012 ACS questionnaire asks about the general nature of the home, asking “Which best describes this building?”; responses include “a one-family house detached from any other house,” “a building with 5 to 9 apartments,” and “boat, RV, van, etc.”
Headquartered in Boston, AIR Worldwide has been modeling the risks of natural catastrophes since 1987, and today does so in more than 90 countries. It is a member of Verisk Insurance Solutions at Verisk Analytics; as that label implies, many of AIR’s clients are insurance and reinsurance companies seeking to understand and manage their risks, but AIR’s client base also includes financial institutions and government entities. Cheryl Hayes, senior research manager at AIR, commented at the workshop about the ACS variables that AIR uses in its research projects and about the general importance of the ACS to the field of catastrophe modeling.
Hayes said that AIR began its work by developing models of hurricanes, tornadoes, and earthquakes and their attendant impacts and risks in 1987, but that catastrophe modeling did not come to the forefront until two extremely costly natural disasters, 1992’s Hurricane Andrew and 1994’s Northridge earthquake. The magnitude of damage and costs in those two disasters sparked awareness of the major impacts that could come from catastrophes, and companies began to realize that catastrophe modeling could help them better manage their financial and personnel risks. The industry, its models, and its methods evolved over the next decade, to be updated and expanded yet again after a major disaster, this time the terrorist attacks of September 11, 2001; models for terrorist behavior and workers’ compensation were first introduced by AIR in 2002. By at least one metric, Hayes said, AIR and its clients have begun to see the benefits of catastrophe modeling; while 11 companies became insolvent as a direct result of Hurricane Andrew in 1992, only three such insolvencies was recorded following the record-breaking 2004 and 2005 hurricane seasons (punctuated by Hurricane Katrina in 2005; see Section 3–C).
The way in which ACS data enter this process is that they are crucial for developing the Industry Exposure Database that is a core component of a catastrophe model. An Industry Exposure Database is essentially a representation of the built environment that may be impacted by a catastrophe; it includes counts of buildings in a particular geographic area and information on characteristics of those buildings, such as their occupancy type, date of construction, and floor area. These variables contribute to another key feature of the database, which is the corresponding replacement values of those buildings. Hayes said that AIR built its first Industry Exposure Database for the United States when the company started in 1987 and has been updating it annually ever since—and, all along, the primary sources of information used to build and update the database are the decennial population censuses, the economic censuses, and, now, the ACS.
Elaborating on that point, Hayes sketched the general processes involved in constructing the database, beginning with the generation of risk counts, which are the numbers of dwellings and establishments. Information that they try to assemble for properties include the manner and type of construction—e.g., whether it is a high- or low-rise structure, and made of wood or concrete—because that information affects the potential vulnerability of the structure. In addition to occupancy type, the floor area of properties is a particularly important variable because it is essential to valuing the property. In addition to the censuses, this information (along with data on construction costs) is derived from housing surveys and from property costing and construction reports. The replacement/rebuild costs are calculated from the data on square footage, using different multipliers based on construction type and height; costs are also adjusted based on local and regional variation in costs of materials and labor. When these elements are combined and then benchmarked against other sources—data from clients and reports from the insurance industry—the result is a robust Industry Exposure Database.
Hayes said that the decennial census provides requisite information on the number of housing units—total housing counts. Displaying a map showing the percentage change in housing counts between the 2000 and 2010 censuses, Hayes noted that many of the states showing the highest rate of change in housing counts are the southeastern states—Florida through North Carolina—that are also prone to the specific disaster of hurricanes, which means that they have an increasing number of properties at risk of damage.11 While the census sheds light on the contours and gross change in housing, the ACS is essential to generate detail on the characteristics of the housing. The specific housing items included in the ACS questionnaire are fairly blunt, but Hayes noted that many of them provide key clues for modeling:
- The type of structure, number of units in the building, and the year the structure was built all hint at the vulnerability of the structure; mobile homes are more likely to be damaged or destroyed by events like tornadoes than large complexes.
- Combined with other information from building permits or industry analyses, the year the structure was built can be used to infer things about the likely type of materials (e.g., whether masonry or wood was more likely to be used in particular states at a certain time).
- The ACS variable on the number of rooms in the house can be used as a rough proxy for the floor area of the housing unit (and structure), which in turn is a key part of the replacement value calculation.
11Texas, also on the Gulf Coast, also ranked in the high growth in housing rate category on Hayes’ map; much of the Mountain West also experienced major growth in the number of housing units, and Nevada stands alone as the most extreme housing growth rate—and in the impact from the housing crisis and recession.
- Other variables like per capita income and employment status have bearing on the valuation (replacement cost) of a property.
Of course, Hayes said, the ACS is not AIR’s sole source of information, but the critical variable of floor area is a good example of the ways in which AIR finds the ACS essential to fill in gaps. She displayed a graph depicting the average residential structure for small areas across the country, based on data that AIR is able to acquire directly through various means. More to the point, the map also showed—shaded in dull grey—the areas in which the direct data are not available. The coverage of the direct data sources is spotty—weak in rural areas in the West but also surprisingly low for some midwestern and eastern states. But, Hayes said, the ACS data provide great assistance—using the ACS variable on the number of rooms in housing units as proxy for floor area, the same map drawn with ACS data has vastly fewer “no data” holes, and the direct measures can be compared with the ACS-based proxies for areas where they overlap in order to judge the quality of the proxy measurement.
In general, Hayes concluded, the ACS is vital to AIR’s annual update and maintenance of a robust Industry Exposure Database and, once updated, that database drives a number of important analyses and modeling efforts. From an ongoing research standpoint, Hayes said that the exposure database is continually used along with historic loss estimates to develop, validate, and recalibrate its core catastrophe models. The models and the exposure database are used for the main thrust of AIR’s work, part of which is to generate real-time estimates of losses—before events occur and as they are unfolding—to enable better planning and management of reserves. The data and methods are also used to validate the losses reported by individual companies—and so assess the quality of their own data—so that they can better manage their financial risk.
Hayes wrapped up her discussion by displaying a table showing AIR’s estimates of the total insured value of properties in each state that borders the Atlantic Ocean or the Gulf of Mexico, along with the percentage of that value corresponding to coastal counties—and so at greater risk of hurricane damage.12 Hayes said that table speaks to the importance of catastrophe modeling to companies writing insurance in areas like Florida—where AIR estimates that almost 80 percent of the insured property value is in coastal areas. Moreover, the development of these models over time has suggested rapid growth in this value—an annual increase in coastal counties’ property valuations of roughly 7 percent. She ended by commending the ACS as an integral part of Industry Exposure Databases, and so argued that the continuance of the ACS is vital to the clients who rely on AIR’s models to manage their risks. She conceded that many of AIR’s clients “might not know it”—might not fully appreciate how big a role that Census Bureau and ACS data play in the modeling and their resulting abil-
12The same table is available at http://www.air-worldwide.com/_public/images/pdf/AIR2008_ Coastline_at_Risk.pdf, which also spells out which counties are defined as “coastal.”
Moderating the discussion session, David Crowe (chief economist, National Association of Home Builders [NAHB]) said that he wanted to take the opportunity to make clear that the clientele and users in the housing industry value the ACS in the same way as other business sectors. He said that housing is one of the most distinctly different and variable commodities from place to place—the old saying stressing the importance of “location, location, location” has never been truer. As suggested in Hayes’ presentation, Nevada experienced a staggering home construction boom—and has been hit accordingly hard by the subsequent bust—while other states (e.g., Texas) were not as hard hit. The way we know such things, Crowe said, is through data sources like the ACS. He observed that NAHB has made extensive use of ACS data—in analyses for its client builder firms to understand the differentiation in housing trends across space.
In addition to documenting the effects of the recent economic trends, Crowe said that the ACS has made possible analyses of important aspects of the broader real estate industry, and he offered two specific examples. The first is home remodeling, which Crowe noted has become a much more productive component of the real estate industry than new construction. ACS estimates were instrumental in NAHB’s construction of county-level estimates of remodeling expenditures, which in turn permit NAHB clients to understand the strengths and weaknesses in this increasingly important segment of the marketplace. As another example, Crowe said that NAHB has used ACS data to derive “affordability indexes”—factors to understand how housing affordability varies across locations and by demographic groups.
Listening to the presentations in this session, Crowe said that the common thread is “the use of data to make appropriate and intelligent decisions”—nothing more elegant or sophisticated than that—but that each presentation is also “a good demonstration that [there] isn’t an immediate transfer from the Census Bureau to the client.” Intermediaries are needed to compile the information, to develop it further, to construct products of the necessary level of detail, and to provide assistance in handling and interpreting the numbers—all of which helps the clients make informed decisions.
Opening the question period, Crowe asked the speakers a form of the same overview question that had been raised in previous sessions: What would each speaker do, and what (if any) data could serve as a backstop, if the ACS were no more? Conrad answered that he and other economic development organizers
would have to turn to private data vendors—and hope that “they are somewhere in the ballpark with their numbers” relative to what is available in the ACS. However, he emphasized that he works with small communities, which are particularly limited in their budgets and so would sharply constrain their ability to acquire the data; the availability of ACS data as a public good is greatly important to users. Genadek offered that IPUMS users would likely turn to the CPS but—in the interest of continuing research across time—would have to hold out hope that the ACS content would return in the form of a decennial census long-form sample. Hayes—from the perspective of a company reliant on statistical modeling—said that AIR would have to do a lot more modeling on its own, pulling together historical data and doing whatever it could to model and project from that base. But, she emphasized, the loss of the ACS “would definitely put us at a big disadvantage.” Levanon concurred with Genadek that, for the Conference Board’s labor market research, the alternative would almost surely be the CPS—but, as noted in other presentations, the sample size limitations on the CPS are such that the CPS is a “far second-best” alternative in many applications. For detailed analysis of occupations like those that are now possible with the ACS, researchers like the Conference Board would face the stark options of either using larger aggregation groups or not doing projects at all, and that would be significantly damaging. The unique nature of Acxiom’s data resources is such that Christenson said that one reaction would be to position the company’s private records-data-based holdings as an alternative for some users’ needs. But, as Christenson emphasized in his presentation, Acxiom depends on the ACS data as well; were the ACS to cease to exist, he said that Acxiom would have to treat the last release of ACS data like a decennial census data file: freeze it and continue to use it in its models going forward, but recognize that it will be less reliable over time. Crowe summarized, adding his and NAHB’s viewpoint that what would result is less definitive decisions—or at least poorer information feeding decisions.
With the floor opened to questions, Lester Tsosie (Division of Economic Development, Navajo Nation) asked about using ACS data to study labor migration—whether it is possible to extract information on between-county migration or, ideally, between tribal reservation areas and surrounding areas, from the PUMS files or other tabulations. Conrad answered that he does not typically use PUMS data much due to the nature of his projects. Reiterating that the basic nature of economic development projects is telling a story with a variety of data sources, he said that he commonly uses the county-to-county migration data developed by the Internal Revenue Service based on annual tax returns and—specific to mobility within labor market areas—data from the Census Bureau’s Longitudinal Employer-Household Dynamics (LEHD) program. Scott Boggess (U.S. Census Bureau) commented that county-to-county migration tabulations from the ACS (based on the question on place of residence 1 year ago)
are available on the Census Bureau website—albeit not directly from the American FactFinder interface, where those tabulations do not naturally fit.13
Crowe asked another question of the set of presenters, asking them whether they think that their clients ultimately have a good sense of where the information comes from—do they know that the products being delivered to them are based on ACS data, or just general “census data” from the Census Bureau? Christenson answered first: “the term ‘American Community Survey’ confuses every single person I talk to, internally and externally.” He sees very little awareness of exactly what the ACS is and that, historically, Acxiom and others wind up using “Census Bureau data” as a descriptor because the ACS label simply does not ring a bell. Hayes agreed, recalling that references to “ACS” were unclear to many, if not most, of the participants of an AIR client conference in April 2012; they, too, turned to general descriptions of “census data,” though they also tried to introduce the concepts and terminology of the ACS. Conrad admitted to using “census data” as a descriptor, but more as a convenience; in the economic development arena, so many data sources and so many acronyms can crop up in products that “census data” clears some of the acronym clutter. However, he argued that—if users like himself and his fellow presenters rely on the data so much—it is incumbent on them to try to inform and educate clients. Another workshop participant added that “ACS is a brand name that nobody knows” on Capitol Hill; in conversations with constituents, congressional staff find that “census data” is less confusing. In short, Crowe summarized, what he concluded from this exchange is that—“to use an Acxiom phrase”—the ACS “needs branding.” Christenson acknowledged this, and added that Acxiom itself is starting to label new releases of its Market Indices product as “Market Indices ACS,” to start to build awareness.
Ken Hodges (Nielsen, and co-chair of the workshop steering committee) said that he wanted to interject a point that perhaps runs to the contrary. He recalled a session about 8 years ago, when his company was still known as Claritas, where he ran a session on the ACS at a well-attended client conference. At that time, awareness of the ACS and its basic features seemed to be quite high. Genadek added that academic researchers in general—such as make up a large share of the IPUMS user base—seem to have high awareness of the ACS nomenclature; in academic circles, at least, the “branding” of the ACS has gone fairly well. Crowe agreed that intensive data users know the ACS very well but that he is not surprised that end-stream clients of business users or data aggregators are not fully aware of the ACS. He said that the challenge, and the point that he wanted to raise in starting the line of questioning, is that if people do not know that “the ACS” is a valuable piece of information, “they will never come to the defense of it” if the survey is vulnerable. Christenson commented that the calls, emails, and questions he has received concerning the ACS have
13These migration data are available at http://www.census.gov/hhes/migration/data/acs.html.
probably increased 10-fold with the legislative moves challenging the ACS (see Section 1–B); a possible “silver lining” of the recent increased scrutiny of the ACS is that awareness of the survey may grow as well.
Constance Citro (Commitee on National Statistics) asked Crowe and the presenters about the housing content of the ACS—what use is made of the questions on the ACS about mortgages and utility bills, because those questions are among the most difficult for respondents to answer. Conrad replied that he regularly makes use of the ACS data on average monthly rents and mortgage payments in characterizing the housing markets in smaller, rural areas. In those rural areas, realtor data—if they exist—-can be particularly sensitive to the role of realtors as sellers; the ACS measures might be less direct but may better reflect prevailing conditions than volatile under- or overstatement of values that might creep into realtor data. Crowe added that these ACS housing variables are important to NAHB’s calculation of affordability indexes because they get at payment burdens and loan-to-value ratios. He added that NAHB has also made use of the ACS’s data item on property taxes paid. The property tax rate in a particular locality is often stated as percent of the value of the home or property, but is often based on something other than the true market value of the property. He said that the ACS variables may provide a better measure of the actual value of the home and, thus, a better basis for comparison across areas. NAHB has published county and even subcounty maps to illustrate particularly high and low property tax rates, based on ACS data. Christenson added that, historically, his clients would most commonly request information on median income and median home value; now, his users are growing much more interested in the more-recently-added ACS questions on first and second mortgages and on health insurance questions.
Roderick Little (U.S. Census Bureau) asked the presenters for their views on the issue of voluntary versus mandatory response to the ACS, and whether they would see a voluntary ACS as having a major impact on their work. Genadek said that she agreed with the comments from presenters in other sessions—that mandatory is better than voluntary if only to stave off declines in the response rate (and effective sample size).14 Conrad echoed Miller’s comment earlier in the workshop (see Section 5–B) about the impacts being potentially worse for rural areas; repeating that Iowa has only 22 counties with populations 20,000 and larger, his bottom-line concern is that the loss of mandatory response could start the slide to exacerbating the divide between “data haves” and “data havenots.” In sum, his big concern is approaching a situation where, for Iowa, there will be a lot of information on Des Moines but nothing about the rest of the state. Citro concluded the session by summarizing the brief, hypothesized re-
14There followed a brief colloquy on exactly what the response rate to the ACS is—the response to the initial mail sample, the weighted response rates including follow-up collection by phone or personal interview, and so forth.