Rationalizing Rural Area Classifications for the Economic Research Service: A Workshop Summary (2016)

2

Official U.S. Rural Area Classification Systems

The purpose of the second session of the workshop, as summarized in this chapter, was to set the context and briefly describe the historical development of current rural classification systems developed by the U.S. Census Bureau, U.S. Office of Management and Budget (OMB), and Economic Research Service (ERS) of the U.S. Department of Agriculture (USDA). James Fitzsimmons (U.S. Census Bureau) described the Census Bureau and OMB classification systems. John Cromartie (ERS) described the four rural classification systems developed by the ERS. Stephan Goetz (Pennsylvania State University) moderated the session.

STATEMENT BY JAMES FITZSIMMONS

Fitzsimmons described two statistical area classifications that provide context for the ERS classifications that are the subject of this workshop: the Census Bureau’s urban and rural classification and OMB’s metropolitan and micropolitan statistical areas classification. These classifications have been part of the federal statistical system landscape for many decades, and both have often been parts of the same conversations as the ERS classifications.

The Census Bureau and OMB classifications yield several kinds of statistical entities. First, the urban and rural program provides two entities: urbanized areas and urban clusters. Currently, there are 486 urbanized areas in the United States and 3,087 urban clusters. Based on the 2010 decennial census, they accounted for 80.7 percent of the U.S. population

and about 3 percent of the land area of the country, with the remaining area population and land area classified as rural.

By comparison, the OMB classification currently features 381 metropolitan statistical areas and 536 micropolitan statistical areas.¹ These areas accounted for a larger share of the 2010 population than did the Census Bureau’s urbanized areas and urban clusters—approaching 94 percent versus about 81 percent. The big difference is in the amount of land area accounted for: more than 47 percent for metropolitan and micropolitan statistical areas in the OMB systems, compared to 3 percent for urbanized areas and urban clusters.

Key Program Similarities and Differences

According to Fitzsimmons, a first point in common between the Census Bureau and OMB classifications is that the period leading up to the 1950 census and the period preceding the 2000 census were formative for both. Also, areas delineated for both classifications are based on Census Bureau data, and the Census Bureau tabulates and publishes data for urbanized areas, urban clusters, and rural areas, as well as for metropolitan and micropolitan statistical areas. In addition, the decennial calendar plays a key role in both programs by establishing their basic rhythm. The underlying criteria are reviewed in the years leading up to the decennial census. The new criteria are then published and used with data from that decennial census to provide new delineations. Other points in common are that the two classifications are maintained solely for statistical purposes, delineate specific statistical areas, and are reflected in the ERS classifications. Further, the Census Bureau’s urban-rural classification provides the cores for the OMB’s core-based metropolitan and micropolitan statistical areas. In other words, the urban-rural classification provides part of the foundation of metropolitan and micropolitan statistical areas.

There are also differences between the two programs, Fitzsimmons pointed out. Administrative responsibility for the programs has been based in two separate agencies: the Census Bureau and OMB. Also, the OMB metropolitan and micropolitan statistical areas program is a federal statistical standard, and the Census Bureau urban and rural program is not. The latter classification is widely used by the Census Bureau, but the metropolitan and micropolitan program also receives use across the federal statistical system. The two programs have fundamentally different conceptual foundations: urbanized areas and urban clusters are

_______________________

¹Additional types of statistical areas are delineated under the OMB classification, including combined statistical areas, metropolitan divisions, and New England city and town areas.

morphological, or footprint, classifications, essentially identifying where the densely settled population is. Metropolitan and micropolitan statistical areas are the products of a functional area classification. They rest on densely settled cores, but how far they extend derives from a functional measure, namely, journey to work or commuting. Finally, although the decennial calendar plays a key role in both classifications, the actual update schedules for the two are different.

Program Histories and Delineation Basics

The urban-rural classification program started at the Census Bureau in the late 19th century, Fitzsimmons explained. The first delineations required a minimum population of 8,000—and later 4,000—within an incorporated place. By the 1910 census, the minimum population needed to be urban was 2,500. From 1910 through 1940, the Census Bureau definition of “urban” was incorporated places of at least 2,500 population. Everywhere else was rural. The first delineation of urbanized areas came in preparation for the 1950 census. The year 2000 was also an important time for this program, as urban clusters made their first appearance in areas delineated with 2000 census data.

Urbanized areas and urban clusters, collectively known as “urban areas,” are delineated using published criteria. Each decade, the criteria associated with the previous census are evaluated, and changes are made based in part on comments received. The Census Bureau published the “Urban Area Criteria for the 2010 Census” in 2011 and then applied those criteria with 2010 census data to produce updated urban area delineations.²

Urban areas, both urbanized areas and urban clusters, are densely settled. Using census tracts and census blocks as geographic components, urban areas extend as far as a minimum population density of 500 people per square mile warrants. If within the entity delineated there is a minimum population of 50,000, that entity qualifies as an urbanized area; an entity with a population of 2,500 to 49,999 qualifies as an urban cluster. These areas are blind to administrative boundaries and extend as far as the minimum density threshold indicates. A pre-2000 change is that urban areas are re-delineated in association with each decennial census. Formerly, delineations started with the previous decade’s urbanized area and evaluated whether territory qualified to be added. Now the areas are delineated in an automated fashion, starting with a blank slate each time.

_______________________

²U.S. Census Bureau. (2011). Urban Area Criteria for the 2010 Census. Federal Register, Vol. 76, No. 164, August 24. Available: http://www.census.gov/geo/reference/ua/urbanrural-2010.html [October 2015].

With the 2010 decennial census, there are many more urban clusters (3,087) than urbanized areas (486). Much more of the U.S. population, however, is in the fewer, bigger areas: 71.2 percent of the population is within urbanized areas, and 9.5 percent is within the smaller urban clusters.

The years leading to 1950 were formative for both programs, according to Fitzsimmons, but especially for the metropolitan and micropolitan classification, which was created in that period. Standard metropolitan areas, as they were called then, were first delineated for the 1950 census. Later the areas became known as standard metropolitan statistical areas, and then metropolitan statistical areas. But the late 1940s was not the first time agencies had the idea for a statistical entity that would capture more than the incorporated place and instead capture something at a broader scale. Before then, in 1905, the Census Bureau delineated industrial districts for use in the Census of Manufactures, and in 1920 the agency delineated metropolitan districts, used for the decennial census through 1940.

The year 2000 was important for both programs. In the case of the metropolitan and micropolitan classification program, the 2000 standards introduced micropolitan statistical areas, combined statistical areas, and metropolitan divisions. These areas were first delineated using 2000 decennial census data. Just as the 2010-based urban areas were delineated using the “Urban Area Criteria for the 2010 Census” referred to earlier, metropolitan and micropolitan statistical areas based on that census were delineated according to the “Standards for Delineating Metropolitan and Micropolitan Statistical Areas.”³ The standards were evaluated in the years leading up to the decennial census. A Federal Register notice in 2009 indicated provisional conclusions from OMB and the interagency committee that advises OMB on this program, then final standards were issued in 2010.⁴ Those were the standards applied with 2010 census data to produce the core-based statistical areas—comprising metropolitan and micropolitan statistical areas—announced in 2013.

Delineation of core-based statistical areas starts with a core provided by the urban and rural program, whether an urbanized area or urban cluster. One or more counties associated with a core urban area (of at least 10,000 population) become the central counties of the core-based statistical area, and counties surrounding the central counties are added (or not)

_______________________

³U.S. Office of Management and Budget. (2010). 2010 Standards for Delineating Metropolitan and Micropolitan Statistical Areas. Federal Register, Vol. 75, No. 123, June 28. Available: http://www.census.gov/population/metro/ [October 2015].

⁴U.S. Office of Management and Budget. (2009). Recommendations from the Metropolitan and Micropolitan Statistical Area Standards Review Committee to the Office of Management and Budget Concerning Changes to the 2000 Standards for Defining Metropolitan and Micropolitan Statistical Areas. Federal Register, Vol. 74, No. 28, February 12.

depending on their commuting ties with the central counties. A minimum of 25 percent commuting with central counties brings a county into the metropolitan or micropolitan statistical area as an outlying county.

The distinction between metros and micros is based on the size of the core population. Metropolitan statistical areas have a core population of at least 50,000, or an urbanized area. Micropolitan statistical areas have cores of 10,000 to 49,999 population, or an urban cluster. (Some urban clusters do not precipitate micro areas, namely, those urban clusters that have populations from 2,500—the floor for urban clusters—up to 9,999.) Metros and micros are updated after decennial censuses, but also throughout the decade. The most common kind of updating between decennial censuses is the creation of new micropolitan statistical areas as areas grow in population, and occasionally a micropolitan statistical area will graduate to a metropolitan statistical area.

The inventory of areas generated by the metropolitan-micropolitan program in 2010 shows many more micros than metros. Paralleling the pattern presented by the urban-rural classification, much more of the population is in the fewer, larger areas: 85 percent of the U.S. population in 2010 resided in the 381 metropolitan statistical areas containing 1,167 counties, whereas 8.8 percent of the population was in 536 metropolitan statistical areas containing 641 counties. The very largest metropolitan statistical areas based on the size of the core—those that have an urbanized area of at least 2.5 million—can be divided into smaller units called metropolitan divisions. In 2013, 11 of current metropolitan statistical areas with a minimum core size of 2.5 million population were divided into 31 metropolitan divisions.

According to Fitzsimmons, another big change that took place in 2000 was that, for the first time, the classification featured a nationally consistent base unit for metro and micro areas, namely the county. Prior to that, New England metropolitan statistical areas were delineated using cities and towns, or minor civil divisions (MCDs). In recognition of that delineation history and the continuing importance and availability of data at the city and town level in New England, the classification provides a second, parallel set of MCD-based areas for that region. There are also currently 38 New England city and town areas.

Area Delineations Comparison⁵

Looking at how metro-micro status crosses with urban-rural status, based on the 2010 decennial census, 88.3 percent of the population in

_______________________

⁵For the purpose of this section’s discussion, metropolitan and micropolitan statistical area delineations used are those that preceded the 2013 update. The 2013 delineations are used in other parts of the presentation.

metropolitan statistical areas is urban according to the Census Bureau’s urban and rural classification, which still means that 11.7 percent of the population in metropolitan statistical areas is rural. Fitzsimmons noted that micropolitan statistical areas are almost evenly split between urban and rural, and in the territory that is in neither a metro or micro area—outside core-based statistical areas—75 percent of the population is rural. From the other direction, even within territory outside of both metro and micro areas, 25 percent of the population is urban.

Focusing on the Census Bureau’s urban-rural classification to see how it intersects with metro-micro status, 24.6 percent of the rural population is not in a core-based statistical area, while another quarter of the rural population is in micropolitan statistical areas, and fully half of the rural population of the United States is in metropolitan statistical areas. Fitzsimmons said this means that programs that use the label “rural” in referring to counties that are not in metropolitan statistical areas are missing half of the nation’s rural population, according to the two classifications.

Purposes and Uses of Areas

Fitzsimmons noted that a recent OMB bulletin⁶ pertaining to the metropolitan and micropolitan statistical areas program offered the following advisory on use of these areas:

Both the urban-rural and metropolitan-micropolitan statistical area classifications are used extensively in Census Bureau tabulations and publications. Metros and micros also are used across the federal statistical system. But, despite the OMB admonition, both classifications are used heavily in other, nonstatistical programs as well, Fitzsimmons said.

Transportation programs are prominent among nonstatistical users of the urban-rural classification. Metropolitan statistical areas are used across a wide spectrum of nonstatistical programs, ranging from those in public health, banking, education, and housing to antiterrorism planning. So it is not a great surprise, according to Fitzsimmons, that a significant share of

_______________________

⁶U.S. Office of Management and Budget. (2013). Revised Delineations of Metropolitan Statistical Areas, Micropolitan Statistical Areas, and Combined Statistical Areas, and Guidance on Uses of the Delineations of These Areas. OMB Bulletin No. 13-01, February 28.

the cost of administering the urban-rural and metropolitan-micropolitan programs is associated with taking inquiries of one kind or another about nonstatistical concerns. These inquiries typically focus on qualification of localities and counties for funding programs.

General Observations

Fitzsimmons noted that a major consideration in making decisions about the design of a classification system is its purpose. Beyond that concern, he said, a first question centers on the geographic components, or building blocks. Immediately following that question will come another concerning the basis for determining the extent of the geographic areas and when it becomes appropriate to view neighboring areas as joined instead of as separate areas.

These questions all arose in reviews of the metropolitan-micropolitan program in which Fitzsimmons participated, reviews that might offer insights for an evaluation of ERS area classifications. For example, he said, counties can be unwieldy area components, especially in some parts of the country. But the metro-micro program is a statistical standard, and many statistical programs do not provide data in more precise geography. In the reviews of the metro-micro program in the 1990s and 2000s, the purpose of a statistical standard and the limitations of some statistical programs meant focusing on counties again, as the classifications had since 1950.

Criteria for determining the geographic extent of areas go to the conceptual basis of a statistical area program. The conceptual underpinnings translate into selection of population density for the urban-rural program, but for the metropolitan or micropolitan statistical areas—a functional areas classification—it means journey to work.

Fitzsimmons said a derivative of the issue of geographic extent that has provoked many of the thorniest issues when reviewing approaches to statistical area delineations comes in determination of when very large areas might be taken apart or shorn of their components, or when neighboring areas might be merged into the larger area. Fitzsimmons stated these two issues have galvanized more expressed opinion and various public comment campaigns than any others in classification reviews with which he is familiar.

Finally, he said, most programs face the issue of consistency with past practice. Classifications that have been in the public realm for an extended time need to balance concerns of coherence and stability with concerns about agility and staying current.

Summary

In summary, Fitzsimmons said the Census Bureau’s urban and rural classification and OMB’s metro and micro classification are prominent statistical area programs that provide part of the context for ERS area classifications. Both the Census Bureau and OMB classifications delineate individual statistical areas. These statistical areas have titles or names, depending on the classification; the areas also have specific geographic boundaries. Both of these programs have served their statistical purposes for many decades. The programs have been stable, and the areas they provide have served many tabulation and publication needs. The intended purposes of a classification system determine measures and procedures used in delineation, but classification systems also find themselves serving additional, nonstatistical roles. These additional roles have brought with them extra scrutiny and pressures.

OPEN DISCUSSION OF FITZSIMMONS’ PRESENTATION

John Logan (Brown University) commented that the dependence of these classification systems on data had not been stressed enough, noting that they are almost entirely decennial census driven. Although there could potentially be other sources of data, there are limited data consistent across the country, resulting in restrictions in the ways the data can be organized and therefore what kinds of areas can be created. He said an associated issue is the quality of the data for a small area, for which sample sizes may be small. The smaller the area that is considered a basis for creating classifications, the larger the uncertainty. While using smaller areas as the basis for a classification system has potential, he said it is also limited by the data quality for those small areas.

Fitzsimmons noted the metropolitan and micropolitan classification does not use only decennial census data. The commuting measures are from the American Community Survey. He said use of Census Bureau data reflects not just a concern for consistency across the nation, although that consideration played a key role, but also concerns regarding the openness, transparency, and accessibility of data. If different, nonfederal data were used, responsible agencies would face the issue of ensuring that those data were available to all users so that they could replicate the delineations.

David Plane (University of Arizona) asked about changes anticipated for 2020 in both classifications. Fitzsimmons responded that the Geography Division at the Census Bureau is responsible for the urban and rural classifications, and Michael Ratcliffe, who is responsible for this program, would address the workshop later in the agenda (see Chapter 7). The con-

cern heard most frequently about the metropolitan-micropolitan program relates to combined statistical areas, Fitzsimmons noted. These areas were offered as an extra service for data users, and they afford agencies larger, more encompassing areas to work with than individual metro and micro areas. Combined statistical area designations, however, have confused some users because not all individual metropolitan and micropolitan statistical areas qualify to be parts of combined statistical areas. He stated that this concern probably needs to be resolved before the next decennial review and updating. If the issue is primarily a communication challenge, he said, then a better job of education is needed. Alternatively, OMB could think about other approaches to defining the areas.

David McGranahan (ERS/USDA) observed that, at one point, “urban character” was used to determine whether a county was part of a metropolitan area. After it was dropped, many more counties were included in metropolitan areas. Fitzsimmons explained that the standards preceding 2000 used a sliding scale in determining outlying county qualifications. Commuting patterns were taken into account, but the weaker the commuting ties displayed, the more the measure of “metropolitan character” played a role. He said that measures of metropolitan character were based on estimates of population growth, population density, and percentage of the population that was urban. In an earlier time, an estimate for the percentage of the population in agricultural occupations was also used.

Fitzsimmons said that a key concern for OMB and the interagency committee that advised that agency on the standards for the 2000 round of review was to clarify and simplify conceptually the basis for outlying county qualification. The committee and OMB debated this issue at length, and public comment was taken into account. In the end, the standards provided for much more strongly functional areas, based on commuting ties for determination of outlying county qualification, and discarded other measures. As a result, the standards became more conceptually consistent, simpler, and shorter, Fitzsimmons said.

STATEMENT BY JOHN CROMARTIE⁷

Cromartie outlined the goals of his presentation: to provide descriptions of each ERS rural-urban classification system and its historical context; highlight differences and key similarities in criteria, data choices, and geography; and discuss the reasoning behind key decisions in the development of each classification system.

Cromartie focused much of his presentation on the key differences among the four ERS rural-urban classification codes (see Table 2-1). He

_______________________

⁷Cromartie (2015) prepared a paper for the workshop summary (see Appendix B).

TABLE 2-1 Economic Research Service (USDA) Rural-Urban Classification Codes

SOURCE: Prepared by John Cromartie for his presentation. Based on data from the USDA Economic Research Service. Available: http://www.ers.usda.gov/topics/rural-economypopulation/rural-classifications.aspx [October 2015].

noted they differ in terms of geography, number of categories, criteria used, and initial release dates.

Though the four classifications are very different, he pointed out their key similarities. First, they are anchored to the metropolitan concepts and the metro-nonmetro dividing line. Rural equals nonmetro; thus, the 50,000 population threshold is a key dividing line, making the Census Bureau’s urbanized areas a key construct. Urbanized areas form the basis of metro areas, and they form the beginning points for all ERS classifications. Second, within nonmetro areas, there are two dimensions to the rural-urban continuum—size and proximity. Counties are classified in terms of their own urban size and by their proximity to nearby metro areas. These two dimensions are also part of all four classifications.

Rural-Urban Continuum Codes (RUCC)

The Rural-Urban Continuum Codes (RUCC), commonly known as the Beale Codes, are a nine-level county classification first created for an ERS report (Hines, Brown, and Zimmer, 1975). This report documented socioeconomic changes for nonmetro areas during the 1960s. The 1960s were the last period of massive rural to urban migration, Cromartie said, which fueled increasing metropolitan dominance, increasing rural diversity, declining farm towns, and increases in new growth centers. This report provided one of the few rationales for the transition to looking at the world from a metropolitan lens and why that was happening. One rationale behind these codes, he noted, was to differentiate diverging types of nonmetro space, such as farm areas where towns and villages were declining versus new growth centers proximate to metro areas.

In the RUCC, two dimensions characterize nonmetro counties: urban size and metro proximity. The dimensions of rural America have changed over time along this continuum, and the RUCC has been successful in explaining socioeconomic conditions in rural America. Changes in settlement patterns and criteria for defining census-based urban areas and metro areas have reduced the number of nonmetro counties by roughly one-quarter from 1970 to 2010, from 2,700 to 2,000. As expected with more metro counties, accessibility of the rural population to metro areas has increased: the share of adjacent counties has gone from 39 to 52 percent within the nonmetro category of counties.

Urban Influence Codes (UIC)

Cromartie next described a similar 12-level county-based classification system, called Urban Influence Codes (UIC), created in the 1990s by ERS staff. The initial six-level version was developed for an ERS report

documenting the rural crisis of the 1980s, similar to the report prepared in the 1970s documenting what had happened in the 1960s.

He identified four differences between the UIC and the RUCC/Beale Codes:

1. The UIC emphasizes adjacency to metropolitan areas. Adjacency drove nonmetro population growth in the 1980s to a greater degree than during the previous two decades.
2. The population size of the adjacent metro area was added as a key component of the UIC based on ERS research that showed that just adjacency to metro area was not enough. Distinctions by size drove population and job growth.
3. The UIC incorporates a size threshold for population of 1 million to classify counties in large and small metropolitan areas. In addition to a population size threshold of 1 million, the RUCC also used a smaller size threshold (under 250,000) for counties in metropolitan areas.
4. Size of the largest city was used to develop urban size categories instead of total urban population. This change provided alignment with central-place principles showing employment opportunities and service provision varying by city size.

The updated UIC classification based on the 2000 decennial census has 12 categories, 2 for metropolitan counties and 10 for nonmetropolitan counties. (See background paper in Appendix B for a description of the 12 categories; see also Table 2-1.) Categories 3 to 7 are all adjacent to metro areas, and the rest are not. The UIC is complex and not as popular as the Beale Codes, Cromartie said, but it offers a different perspective.

Rural-Urban Commuting Area (RUCA) Codes

The Rural-Urban Commuting Area (RUCA) code scheme is the first nationwide subcounty classification system widely adopted for research and policy. It was developed in the 1990s as part of an interagency agreement between ERS and the Office of Rural Health Policy of the Department of Health and Human Services (HHS). Counties were too large, especially in the West, to adequately target rural health programs or identify places that needed hospitals or help with providing health care. This classification system has 10 primary codes and 30 secondary codes.

In the 1990s, there was a growing need for a subcounty classification system in order to look at rural issues in more detail. In this period of urban sprawl and fragmentation, Cromartie said, there was a growing need to capture the increasing complexity of hierarchical relations and

patterns of shared influence. The Census Bureau and OMB sponsored papers and convened a “Metro 2000”conference in the 1990s to rethink underlying concepts of metro areas. Two of the four sponsored papers, Morrill (1995) and Frey and Speare (1995), proposed subcounty classifications. Building on this work, ERS collaborated with geographer Richard Morrill to develop the RUCA codes, through funding from the Office of Rural Health Policy.

Part of the complexity of the codes was because Morrill and ERS had different goals for these codes, Cromartie said. ERS was interested in the basic question of what metro areas would look like if one tried to adhere closely to the criteria but used census tracts instead of counties as building blocks. Morrill was interested in the idea that places have different functions, with an overlapping nature to rural and urban. According to Cromartie, Morrill thought that there was a hierarchy but that it was overlapping. For example, a little town outside a metro area could be both a bedroom community and its own employment center. As a result, a more complex classification was adopted, providing flexibility for researchers interested in analyzing a variety of settlement patterns and functional relationships between areas.

Cromartie explained the method used in developing the codes as follows: Replace counties with census tracts, and aggregate tracts to form urban area approximations; then, using data from a special tabulation of tract-to-tract commuting flows prepared by the Census Bureau, analyze commuting flows between rural tracts and urban areas. Ten primary codes were identified based on the direction of largest commuting flow. Thirty secondary codes were identified based on the direction of the second-largest commuting flow to depict the overlapping nature of the urban-rural hierarchy. (See background paper in Appendix B for a listing of the primary codes. The methodology is also documented in Morrill, Cromartie, and Hart [1999].)

Frontier and Remote (FAR) Codes

As Cromartie highlighted, the Frontier and Remote Codes is a four-level grid-based classification developed in the 2000s. Grid analysis is a new approach using data that have been downcast to very small ½ x ½ kilometer grid cells. After analysis, results are aggregated up to larger geographic units. Accessibility/remoteness is defined in relation to the time it takes to travel by car to the edges of nearby urban areas and not by adjacency. It describes territory characterized by some combination of low population size and a high degree of geographic remoteness.

The motivation for the development of FAR Codes was twofold, he said. First, demand for a geographically detailed delineation of frontier

areas grew in federal policy circles, especially among rural health specialists, as programs emerged with the legislative mandate to improve access to health care in frontier areas. Second, research by Mark Partridge and others (Partridge et al., 2008a, 2008b, 2008c, 2008d) showed the economic and demographic costs of remoteness were increasing, and the variability of rural well-being was still strongly tied to the structure of the urban hierarchy. It was not just remoteness from any city, but the size of the city mattered, which led to creating levels of remoteness.

The methods for creating these levels of remoteness are fairly straightforward, according to Cromartie. For each of the approximately 25 million grid cells covering the entire United States, the census urban area designation and 2010 block-level population were added to the data record, along with the road network and posted speed limits. For each grid cell, distance was calculated as travel time by car to the edge of a nearby urban area in four urban population size classes: 2,500–9,999; 10,000–24,999; 25,000–49,999; and 50,000 or more. Four FAR levels were identified at the grid level, based on different urban classes, then aggregated to zip code areas based on population.

These concepts come from central place ideas, explained Cromartie, so FAR-1 represents frontier areas with access to high-end services, while FAR-4 represents the most remote areas based on access to low-end central services, with two intermediate codes. The codes nest, in that all FAR-4 areas are also FAR-3, FAR-2, and FAR-1 areas; all FAR-3 areas are also FAR-2 and FAR-1 areas; and all FAR-2 areas are also FAR-1 areas.

Once the grid level analysis is done, grids are aggregated to zip codes. FAR-1 areas have a majority of their populations living 60 minutes or more from urban areas of 50,000 or more. Based on decennial census 2010 data, FAR-1 represents 52 percent of the land area, but just 4 percent of the population. These percentages are down quite a bit from the ERS analysis in 2000 because of the addition of new urbanized areas.

The FAR-2 level adds a 25,000 population threshold and a 45-minute drive: these zip code areas have a majority of population living 60 minutes or more from urban areas of 50,000 or more people and 45 minutes or more from urban areas of 25,000–49,000 people. The FAR-3 level adds a 10,000 population threshold and a 30-minute drive: it includes zip code areas with the majority of population living 60 minutes or more from urban areas of 50,000 or more people; 45 minutes or more from urban areas of 25,000–49,999 people; and 30 minutes or more from urban areas of 10,000–24,999 people.

The FAR-4 level adds the smallest towns of 2,500 or more people, with a 15-minute drive: it includes zip code areas with the majority of population living 60 minutes or more from urban areas of 50,000 or more people; 45 minutes or more from urban areas of 25,000–49,999 people; 30 minutes

or more from urban areas of 10,000–24,999 people; and 15 minutes or more from urban areas of 2,500–9,999 people. It has 1 percent of the population but 35 percent of the land area.

Summary

In closing, Cromartie listed some questions that he said need to be addressed:

• Is it important to maintain the different perspectives offered by the two ERS county-level classifications?
• If not, which elements should be given priority?
• Are there ways to simplify the RUCA Codes and still maintain the multilevel, overlapping hierarchy they represent?
• Should ERS consider applying the grid-based methodology, currently used to define very remote areas, to a more comprehensive classification system?
• Are there broader applications for the measurement of distance using detailed travel-time analysis?
• What is the historical context today that is relevant in terms of design? For example, there is a lot of talk about growing spatial inequalities in rural areas. How can ERS capture that? There is the emergence of new trends in big-city downtowns, which will have an impact on migration trends for rural areas, such as people considering city centers as viable retirement places.

OPEN DISCUSSION OF CROMARTIE’S PRESENTATION

David Brown (Cornell University) provided background to the ERS report referred to by Cromartie (Hines, Brown, and Zimmer, 1975). For the 1950 decennial census, Duncan and Reiss (1950) described social characteristics of urban and rural areas. For the 1960 decennial census, a census monograph, Hathaway, Beegle, and Bryant (1968), served the same purpose. For the 1970 census, there were no plans for a census monograph series. ERS was asked to develop a publication that showed how various aspects of social and economic structure had changed across the nation’s geography. According to Brown, this was important because there was no intention at that time to develop a classification scheme for any other purpose than that publication. He noted that the authors were surprised when many people and agencies started to use the scheme. He explained the scheme became known as the Beale Codes because after the original publication, Calvin Beale, updated the codes with new data, distributed it, and answered questions about it.

Brigitte Waldorf (Purdue University) observed the two dimensions in Rural-Urban Continuum Codes, size and adjacency, are continuums; in the coding, she asked why size had priority over adjacency in the 1970s when the codes were developed. She also asked Cromartie to explain how grid sequences are aggregated to zip code levels in the FAR coding.

Cromartie responded that FAR coding is done with population and road layers overlaid on the grid. Initially, 2010 census population at the block level is downcast to grids using weighted area interpolation. Each grid is given a population, so that the total is equal to the total U.S. population. Then each grid is identified by whether it meets the criterion for the FAR-1 code. Then these FAR-1 grid cells are aggregated to zip code areas, overlaying those on the grids and simply tabulating the populations of the FAR-1 grids that fit into each zip code.⁸ This gives a percentage of the population that is FAR-1 for each zip code. A vast majority of zip codes are either all FAR-1 or not. Those zip codes with less than 100 percent and more than 50 percent FAR-1 population are also considered FAR-1 zip codes. This analysis is repeated to identify zip code areas that satisfy the criteria for the other three FAR levels. The state summary populations that ERS provides are for the total zip code population in each FAR category.

Ken Johnson (University of New Hampshire) asked about the stability of zip code boundaries in remote rural areas and asked about using the grid instead. Cromartie responded that a major reason is that it is hard to share data for 25 million grid cells. One can share maps, which ERS is considering. Although not currently on the ERS data product website, a link to an interactive map will allow users to look at the grid results. He said it would be fairly straightforward to devise a system whereby an address can be typed in to see whether it is FAR-1 or not, based on grid-level analysis.

Cromartie pointed to a problem with post offices closing, which leads to changes in zip code boundaries every year. Census tracts also change, but only once every 10 years. Although there are comparability checks between the two, there are issues, he said. ERS uses census tracts for RUCA Codes, but some communities of interest need to base their data on something different. Zip codes are useful for survey data because people know their zip codes but not their census tracts. On the other hand, he said, census data are easier to use with census tracts. Zip codes and census tract data are both useful, he concluded.

Thomas Johnson (University of Missouri) commented that the FAR and several other codes describe a geography, and it is possible to tally the population in that geography. He asked about other kinds of data

_______________________

⁸ERS is also considering publishing the FAR Codes for other geographies, such as census tracts and Zip Code Tabulation Areas (ZCTAs), the census approximation for zip codes.

associated with such geographies, and suggested that the process of estimating population solves the problem of preserving anonymity. He queried whether it would be possible to report data like income and other variables for those geographies. Cromartie responded that ERS does not report data for grids in particular. Other research projects at ERS have used this method, and they may have downcast some data on housing and food. He said he was not familiar with the level of accuracy but expressed concern about trying to downcast data for geographic units smaller than census blocks.

Mark Shucksmith (Newcastle University) referred to Fitzsimmons’ point that these classifications were to be used purely for statistical purposes. He said that he sees at least two purposes for which classifications might be used: for policy and for analysis. In thinking about how classifications might change, it is important to consider both uses, he said. Shucksmith said he was also struck by Cromartie’s point that in the 1959 and 1969 data, there was a very strong correlation of rurality with poverty, which is disappearing now. He noted the question that follows is the usefulness of rurality in analyzing inequality and poverty, or if other variables better reflect inequality that should be considered in classification systems.

John Pender (ERS/USDA) asked about considering income level, or income per capita of a proximate city, when thinking about access or market potential. He suggested that an area close to a wealthy city, even controlling for population size, might have more economic opportunities than one close to a poor city. Cromartie responded that the area of focus he saw in the 1990s was total population size and how that size affected the impact of adjacency and population change. He noted that rapidly growing metro areas would have a bigger impact than ones not growing as fast. He said he had not looked at income, but employment opportunity might be another variable to consider. Employment opportunities would attract more people from rural areas.

Michael Woods (Aberystwyth University) noted the definition of FAR areas is parallel to work done in Wales. He said that the Welsh government defined “deep rural areas” as more than 30 minutes travel from a settlement of 10,000 people. Those areas were analyzed through a household survey conducted across rural Wales. He reported one of the key differences the survey found was that levels of dissatisfaction with both public and private services were significantly higher in deep rural areas than they were across the rural country as a whole. According to Woods, this illustrates the definition of “deep rural areas” picks up something useful.

Referring to a point by Partridge about space and accessibility, Woods asked to what extent it is important to start thinking about Internet,

broadband, and telecommunications access when thinking about accessibility. He observed he has not seen this type of access drawn into any classifications of rurality or prerurality. Cromartie observed that work on broadband accessibility is under way, as businesses are not attracted to an area without it. He said that the issue of different ways of looking at metro areas—including media markets, newspaper circulation, and TV—has been on the table for some time.

Partridge commented that using Canadian data, he and Rose Olfert looked at the cause of spread effects—namely, rural areas near urban areas that were growing, or potentially growing, taking advantage of urban-led growth. They found that population size was first order, growth and employment opportunities were second order, but third order was whether income growth spreads out wealth. He reported that in a statistical sense, the spread effect was significant, but in terms of the magnitude, the income effects were rather small.

Danielle Rhubart (Pennsylvania State University) observed analysts use much data from the five-year American Community Survey estimates. She observed that most of the data are not at the zip code level, and wondered whether the FAR Codes could be aggregated into a dataset at the county level. She asked about any available dataset that provides the proportion of a county that lives in a FAR area and a proportion of county land that is designated as a FAR area. Cromartie responded that on the analysis file, ERS currently has four geographies, and the plan is to aggregate to census tracts, ZCTAs, and counties, as well as to zip codes. The question is how to design a data product that is not too confusing. Now ERS provides zip code-level data. One can download a zip code file and have the FAR identification for those areas. He asked whether ERS should also put out county-level, tract, and ZCTA files, or whether confusion would become a problem. He said he sees value in providing all of these geographies.

Goetz referred to Cromartie’s comment about the relationship between FARs and their applications in the health sector. In a region where he works, he said, there are many concerns about rural areas treated differently because of where they fall on the continuum. Looking at the FAR Codes, he said, the West dominates. He asked about conditions under which a FAR Code would be applicable in the Northeast, which also has rural areas, but where, except for northern Maine, they seem to be disappearing. Cromartie noted a region in northwest Pennsylvania identified as FAR-1. He said one thing he likes about FAR Codes is that there are frontier areas in the East. He noted that with standard definitions of frontier that use county-level measures of density, no counties in the East qualify. When the FAR Codes were first released, he said, ERS got some protest from people in the West not used to sharing the frontier

area designation with Eastern states. He noted the conditions of roads in many places make a difference.

Logan stated he was surprised at the use of census tract commuting data because of a lack of confidence in the quality of the data. He said that if the data were used to determine programs needed in a particular census tract, he would not trust them because the standard errors are huge, although he added he does not see a problem if the purpose is to get a general idea about patterns. An advantage of the FAR scheme, he said, is that the distances are accurate and easy to calculate for very small areas. To him, the downside is that the FAR relies on the assumption that accessibility to services is totally dependent on distance. On average this may be true, he said, but perhaps not for any particular area. If the goal is to draw programmatic conclusions for particular areas, deciding whether an area gets program money or not, he said he worried about the assumption that distance translates into accessibility to services. Cromartie responded that this distinction is helpful for ERS, which provides an overview or analysis of conditions in rural areas. Urbanized areas are key to all of these classifications, he said, and they also form the basis of the main USDA definition that is used for rural development programs.

Plane wondered whether the federal government is really the entity to be defining what is meant by rural, given that the country has different historical settlement and regional patterns, and yet a federal definition has to be consistent across the country. It was certainly an issue with the old metropolitan definitions, which got quite complicated, Plane said. Coming up with a consistent definition is a challenge, he added.

This page intentionally left blank.