How Should We Define “Rural” When Investigating Rural Tobacco Use in the United States?

Abstract

Purpose:

Investigations into rural tobacco-related disparities in the U.S. are hampered by the lack of a standardized approach for identifying the rurality—and, consequently, the urbanicity—of an area. Therefore, the purpose of this study was to compare the most common urban/rural definitions (Census Bureau, OMB, RUCA, and Isolation) and determine which is preferable for explaining the geographic distribution of several tobacco-related outcomes (behavior, receiving a doctor’s advice to quit, and support for secondhand smoke policies).

Methods:

Data came from The Current Population Survey Tobacco Use Supplement. For each tobacco-related outcome, one logistic regression was conducted for each urban/rural measure. Models were then ranked according to their ability to explain the data using Akaike information criterion (AIC).

Results:

Each definition provided very different estimates for the prevalence of the U.S. population that is considered “rural” (e.g., 5.9% for the OMB, 17.0% for the Census Bureau). The OMB definition was most sensitive at detecting urban/rural differences, followed by the Isolation scale. Both these measures use strict, less-inclusive criteria for what constitutes “rural.”

Conclusions:

Overall, results demonstrate the heterogeneity across urban/rural measures. Although findings do not provide a definitive answer for which urban/rural definition is the best for examining rural tobacco use, they do suggest that the OMB and Isolation measures may be most sensitive to detecting many types of urban/rural tobacco-related disparities. Caveats and implications of these findings for rural tobacco use disparities research are discussed. Efforts such as these to better understand which rural measure is appropriate for which situation can improve the precision of rural substance use research.

Introduction

A substantial body of health research indicates that rural populations in the United States (U.S.) are a disadvantaged group.^1-3 Generally, this disadvantage is identified by contrasting rural individuals with urban (or sometimes suburban) individuals. For instance, one recent study found that, compared to those in urban areas, individuals living in rural areas had higher rates of lung and bronchus, colorectal, oral and pharynx, larynx, and cervical cancers.⁴ Contributing to these disparities are meaningful differences in tobacco usage: adults who live in rural areas of the U.S. are more likely to smoke cigarettes and use smokeless tobacco than their urban counterparts.^5-8 Research investigating the reasons for this tobacco-use difference is emerging, and indicates factors ranging from socioeconomic status, to medical access, to support for secondhand smoke policies, to local culture.^9-13

As investigations into rural tobacco-related disparities in the U.S. continue to develop, this body of work is hampered by the lack of a standardized approach for identifying the rurality (and, consequently, the urbanicity) of an area. Specifically, multiple definitions of urban/rural exist and little is known about the heterogeneity among them. The measures are also only modestly correlated with each other.¹⁴ Research on other health outcomes in the U.S.^15-23 has either discussed or tested the strengths and weaknesses of these various definitions; yet to our knowledge, the issue has not been explored for the case of tobacco. This is important because, due to their variability, these diverse definitions can produce inconsistent research findings or generate situations where the measure being used is inappropriate for the research questions. Further, as the distribution of government funds, services, and other resources are often allocated based on rural status, inaccurate classifications could adversely impact how disparities are addressed. Tobacco-disparities research could be improved substantially with the development of methodological guidelines that clarify when various definitions of urban/rural should be used.

As shown in Table 1, U.S. classification systems for defining “urban” and “rural” incorporate different combinations of factors (population density, adjacency to urban areas, and commuting flows to urban areas). They also focus on different geographical units (e.g., counties or Census tracts). Perhaps the simplest system, the Census Bureau classifies census blocks as belonging to Urbanized Areas and Urbanized Clusters based primarily on population density, with all remaining areas classified as Rural. The Office of Management and Budget (OMB) classifies counties as Metropolitan, Micropolitan, or Non-Core (“rural”) based on population density and commuting to adjacent urbanized areas (these three levels, can, in turn, be broken down further). Third, the U.S. Department of Agriculture’s Economic Research Service (USDA-ERS) has developed a scheme of Rural-Urban Commuting Areas (RUCAs), which classifies census tracts as Metropolitan, Micropolitan, Small Town, and Rural based on urban density and commuting flows (these levels can likewise be broken down further). The common use of these three definitions—Census Bureau, OMB, and RUCA—is encouraged by their frequent inclusion in large, national datasets and by the fact that their scoring information is publicly available. Yet researchers have recently also developed and validated an additional measure, the Isolation scale,¹⁴ which classifies census tracts along a continuous scale of rurality based on distance from population-dense areas.

Table 1.

Various classification systems for defining “urban” and “rural” in the United States.

Type of Urban/Rural Definition	Measure Details
U.S. Census Bureau’s Urban and Rural Areas	Classifies Urbanized Areas and Urbanized Clusters. All other areas are considered Rural. Definitions are determined based on population density.
Office of Management and Budget (OMB)’s Core Based Statistical Areas and Non-Core Areas	Classifies areas as Metropolitan; Micropolitan, and Non-Core (“rural”). Categories can be further subdivided into 9 rural-urban continuum codes (RUCCs). Definitions are determined based on population density, adjacency to urban areas, and commuting flow.
U.S. Department of Agriculture, Economic Research Service (USDA-ERS)’s Rural-Urban Commuting Areas (RUCA)	Classifies areas a Metropolitan, Micropolitan, Small town, and Rural. Categories can be further subdivided into 10 primary and 21 secondary codes. Definitions are determined based on population density and commuting flow (i.e., the size and direction of worker commuting flows to urban areas).
Isolation	Classifies urban/rural on a continuum. Values are determined by distance to an urban center and selection of the delta parameter.

All of these four definitions of “rurality” have their own strengths and weaknesses. For example, the Census Bureau’s definition, which employs a high threshold of population density to identify “urban,” does little to delineate degrees of rurality. The measure tends to over-classify areas as rural, often lumping suburban and rural together. In contrast, the OMB definition tends to under-classify areas as rural because it defines areas at the county level. Thus, a county that contains both heavily-populated and sparsely-populated regions will be classified as metropolitan. Moreover, its metropolitan designations are made complex by complicated, and sometimes politically-motivated rules.²⁴ The RUCA definition was developed as a reaction to some of these limitations, and therefore uses tract-level classifications and accounts for commuting flows (i.e., from home to workplace) to estimate proximity to urban centers. However, this approach might be insufficient for capturing the complexities of rurality that contribute to health outcomes—which may depend on more than just where residents work, but the distance or length of time it takes to connect to health services. Finally, the Isolation scale has the advantage of being a continuous measure, but by using distance between tract centroids to approximate distance from urban centers, cannot account for features such as highways or rivers that might further support or restrict access to health relevant resources in its current form. Further, unlike the more simple measures described above, there can be different permutations of the measure, and selection depends on the sensitivity desired by the researcher.

Due to the many varying definitions in the U.S. of urban/rural (Census Bureau, OMB, RUCA, and Isolation), as well as their strengths and weaknesses, it is necessary to determine which definition of “rural” is preferable for investigating rural tobacco use. Doing so would support future investigations into rural tobacco-related disparities in the U.S. Therefore, the purpose of this study was to compare urban/rural definitions and determine which better explain the geographic distribution of several tobacco-related factors. As the literature investigates numerous outcomes when defining rural risk, this study was designed to examine multiple tobacco-related outcomes: behavior, receiving a doctor’s advice to quit, and support for secondhand smoke policies.

Methods

Data Sources

Publicly-Available Urban/Rural Codes:

Data for the three commonly-used urban/rural definitions (Census Bureau, OMB, and RUCA) were obtained from their agency’s website (Census Bureau, OMB, and USDA-ERS, respectively). The most recently-available year of classification was selected (2010 for the Census Bureau, 2013 for the OMB, and 2010 for the USDA-ERS). Census Bureau and USDA-ERS data were organized at the Census Tract level. OMB codes were organized at the county level. Tract-level data for the Isolation scale were obtained from the creator’s website (http://doogan.us/isolation/).

The Current Population Survey Tobacco Use Supplement (TUS-CPS):

Data concerning individuals’ tobacco-related behaviors, receipt of a doctor’s advice, and support for secondhand smoke policies came from the restricted use files of the 2014-2015 TUS-CPS. The CPS-TUS collects nationally representative data from approximately 240,000 non-institutionalized civilian U.S. adults aged 18 years and older from all 50 states and the District of Columbia. The supplement focuses on participants’ tobacco-related experiences, with demographic measures available from the core CPS. A complex sample design was used to select respondents;²⁵ briefly, counties or groups of contiguous counties within a state were grouped into primary sampling units (PSUs), and housing units were selected within those PSUs. About two-thirds of surveys were completed by phone and the remaining third are through in-person interviews.

Measures

Urban/Rural Definitions:

Census tracts were classified according to each of the four urban/rural definitions. For the Census Bureau definition, tracts were coded as Urban (areas where the majority of the population resided in an Urbanized Area or Urbanized Cluster—i.e., areas with at least 2,500 people) or Rural (all other tracts). For the OMB definition, tracts were classified according to whether the county was designated as either Metropolitan (areas of 50,000 or more people), Micropolitan (area with at least 10,000 and less than 50,000 people), or Non-core (all other areas). For the RUCA definition, tracts were classified into either Metropolitan (codes 1.0-3.0; area with 10% or more of primary flow to an Urbanized Area), Micropolitan (codes 4.0-6.1; area with 10% or more of primary flow to a large Urbanized Cluster), Small Town (codes 7.0-9.2; area with 10% or more of primary flow to a small Urbanized Cluster), or Rural (codes 10.0-10.6; area with primary flow to a tract outside an Urbanized Area or Cluster). Finally, for the Isolation definition, tracts were classified according to the continuous version of the scale, which assigns scores to census tracts along a continuum of rurality based on their access to population-dense areas. We tested different versions of the continuous scale that varied based on their delta (δ) parameter. As described further elsewhere,¹⁴ this δ parameter is used when calculating a census tract’s Isolation score and can be set anywhere between 0 and 1, with values closer to 1 providing a stricter, less-inclusive criteria for what constitutes “rural.” For this study, ten versions of the scale were originally investigated, where δ was set at values ranging from 0.80 to 0.98. As an example of how tracts were scored, for all versions assessed, tracts with the lowest Isolation scores (below 1.0) were located in the city of Chicago; tracts with the highest Isolation scores (a few decimals above 12.0) primarily consisted of remote parts of Alaska.

Behaviors:

Participants were asked whether they had smoked 100 cigarettes (approximately 5 packs) in their lifetime, and whether they now smoked cigarettes every day, some days, or not at all. Current smokers were defined as those who had smoked at least 100 cigarettes and now smoked some days or every day. Current menthol smokers were defined as current smokers who indicated “menthol” to the question “Do you usually smoke menthol or non-menthol cigarettes?”

Doctor advice:

Current smokers were asked if they had seen a medical doctor in the past 12 months. If so, they were asked: “During the past 12 months, did any medical doctor advise you to stop smoking?” (yes, no). Those who had not seen a medical doctor in the past 12 months were excluded from analyses.

Support for Secondhand Smoke Policies:

All participants were asked two questions assessing their support for secondhand smoke policies. The locations were “bars, cocktail lounges, and clubs” and “casinos” (allowed in all areas, allowed in some areas, not allowed at all). For each item, participants indicating that smoking should be ‘not allowed at all’ were coded as supportive. Participants were also asked about smoking rules inside their homes (no one is allowed to smoke anywhere inside your home, smoking is allowed in some places or at some times inside your home, smoking is permitted anywhere inside your home). Those who indicated that no one is allowed to smoke in the home were coded as supportive.

Demographic Characteristic:

Survey items assessed participants’ sex, age, race/ethnicity, and family income.

Analyses

All analyses were performed with SAS software (SAS Institute) by study team members who traveled to the Census Bureau Restricted Data Center (RDC). The restricted use files of TUS-CPS were first linked with the sample’s demographic data, and then organized at the census tract level. This set of tract-level data was then linked with the data the team brought into the RDC on the urban/rural measures to be investigated (Census Bureau, OMB, RUCA, and the various permutations of Isolation). All analyses were age-adjusted and weighted to reflect the U.S. population, and replicate weights²⁶ were used to calculate error variances.

Analyses began with descriptive statistics to examine the prevalence of population characteristics and tobacco-related outcomes across the urban/rural measures. For ease of interpretation, prevalence estimates examined on the continuous Isolation scale are presented in quartiles, with the highest quartile considered the most rural.

Analyses next assessed the differential utility of the urban/rural definitions for explaining variation in tobacco-related outcomes. Specifically, a series of logistic regressions were conducted on each outcome, one for each urban/rural measure. Isolation was included in its respective logistic regression models as a continuous variable and all other urban/rural measures were include as dummy variables to represent their categories. Analyses controlled for income, age, race/ethnicity, and gender. Models were ranked according to their ability to explain the data using Akaike information criterion (AIC), a likelihood-based measure of relative fit.²⁷ This allowed us to determine the quality of a model for the given set of data, relative to each of the other models. Due to RDC restrictions on how many results could be exported, we could not use additional measures of model fit (e.g., BIC). Likewise, due to RDC restrictions, the a priori decision was to only select and report the top three permutations of the Isolation scale. Thus, for each tobacco-related outcome, this paper presents results for six urban/rural definitions (Census Bureau, OMB, RUCA, and the three highest rated permutations of Isolation), ranked from best (1) to worst (6) according to the AIC measure of relative model fit.

Results

Descriptive Statistics

Table 2 provides age-adjusted, weighted descriptive statistics for the sample. A little over half (51.8%) of participants were female, and nearly 65% were Non-Hispanic White. The table also provides prevalence estimates to indicate how populations were distributed across the various urban/rural definitions. Clearly, each definition provides very different estimates for the prevalence of the U.S. population that is considered “rural:” 17.0% for the Census Bureau, 5.9% for the OMB, and 3.0% for RUCA; as the Isolation scale was divided into quartiles, there were naturally around a quarter of participants in quartile 4 (the most rural) for the three permutations of the Isolation scale.

Table 2.

Population distribution and standard error by demographic and geographic classification: United States (age-adjusted and replicate weighted). Approximate unweighted N = 230,000, although sample size varies across variables due to missing data.

Characteristic	%	Standard Error
Age
18-24	12.62	0
25-44	34.05	0
45-64	34.39	0
65+	18.94	0
Family Income
< $20k	15.90	0.11
≥$20k to < $50k	30.38	0.16
> $50k	53.72	0.17
Gender
Female	51.81	0.01
Race/Ethnicity
Hispanic	15.47	0.01
Non-Hispanic White	64.98	0.02
Non-Hispanic African American	11.72	0.02
Non-Hispanic Other	7.84	0.02
Census Bureau Urban/Rural Classification
Urban	82.99	0.19
Rural	17.01	0.19
OMB Urban/Rural Classification
Metropolitan	84.45	0.28
Micropolitan	9.68	0.29
Non-core	5.87	0.20
RUCA Urban/Rural Classification
Metropolitan	84.19	0.28
Micropolitan	8.76	0.25
Small Town	4.06	0.14
Rural	2.98	0.10
Isolation Scale (δ=0.94)
Quartile 1	25.51	0.19
Quartile 2	25.23	0.25
Quartile 3	25.78	0.26
Quartile 4	23.48	0.24
Isolation Scale (δ=0.96)
Quartile 1	25.79	0.17
Quartile 2	25.21	0.30
Quartile 3	25.74	0.26
Quartile 4	23.26	0.25
Isolation Scale (δ=0.98)
Quartile 1	26.17	0.15
Quartile 2	24.68	0.36
Quartile 3	26.12	0.31
Quartile 4	23.04	0.32

Table 3 presents the prevalence for the tobacco-related outcomes assessed, broken down by the various levels of urban/rural status. For each urban/rural definition, rural individuals had the highest prevalence of current smoking, with estimates ranging from as low as 17.8% (Census Bureau) to as high as 19.9% (OMB). Menthol smoking, in contrast, was higher in urban areas.

Table 3.

Prevalence and standard error of tobacco-related outcome by urban/rural status: United States (age adjusted and replicate weighted).

Urban-Rural Classification	% Behavior				% Doctor Advice		% Support for Secondhand Smoke Policies
	Current Smoker	SE	Current Menthol Smoker*	SE	Doctor advice to quit*	SE	Smoking in bars, cocktail lounges, and clubs	SE	Smoking in casinos	SE	Smoking in the home	SE
Census Bureau
Urban	12.4	0.09	30.7	0.26	47.8	0.41	57.9	0.16	55.6	0.17	89.2	0.11
Rural	17.8	0.20	24.5	0.44	48.0	0.72	52.2	0.35	52.4	0.36	84.0	0.23
OMB
Metropolitan	12.3	0.09	30.3	0.25	48.2	0.40	57.9	0.16	55.7	0.17	89.2	0.11
Micropolitan	18.6	0.29	26.6	0.57	46.9	0.98	52.2	0.59	52.3	0.60	84.5	0.36
Non-core	19.9	0.38	25.3	0.77	46.4	1.30	50.3	0.68	50.3	0.68	81.4	0.48
RUCA
Metropolitan	12.2	0.09	30.4	0.25	48.4	0.40	57.9	0.17	55.7	0.17	89.2	0.11
Micropolitan	18.9	0.33	26.4	0.67	46.2	1.02	52.0	0.56	52.3	0.58	84.6	0.38
Small Town	19.8	0.49	25.8	0.89	46.1	1.33	49.7	0.87	49.6	0.84	81.9	0.58
Rural	19.1	0.47	22.7	0.96	45.8	1.69	54.1	0.81	53.4	0.82	82.9	0.56
Isolation Scale (δ=0.94)
Quartile 1	9.2	0.13	32.6	0.56	46.5	0.84	62.7	0.33	59.4	0.33	90.3	0.20
Quartile 2	11.9	0.15	31.5	0.47	49.6	0.73	58.4	0.29	55.6	0.29	89.9	0.18
Quartile 3	14.3	0.18	29.8	0.41	48.0	0.65	55.3	0.30	54.1	0.33	88.3	0.20
Quartile 4	18.2	0.18	25.4	0.37	47.3	0.61	51.2	0.30	51.2	0.28	84.4	0.22
Isolation Scale (δ=0.96)
Quartile 1	9.1	0.13	32.3	0.53	47.2	0.84	63.0	0.35	59.7	0.35	90.5	0.21
Quartile 2	11.8	0.16	31.4	0.49	49.2	0.72	58.6	0.29	55.8	0.31	89.9	0.19
Quartile 3	14.7	0.19	30.0	0.41	48.1	0.62	55.1	0.30	53.7	0.32	88.1	0.20
Quartile 4	18.2	0.19	25.5	0.37	47.0	0.61	50.9	0.31	51.1	0.31	84.4	0.23
Isolation Scale (δ=0.98)
Quartile 1	9.1	0.15	31.8	0.54	47.9	0.81	63.7	0.35	60.2	0.36	90.5	0.20
Quartile 2	11.9	0.15	30.9	0.49	50.0	0.74	57.8	0.31	55.0	0.33	89.7	0.19
Quartile 3	14.5	0.19	30.2	0.45	47.4	0.68	55.5	0.32	54.3	0.33	88.1	0.20
Quartile 4	18.3	0.19	25.9	0.35	46.8	0.59	50.5	0.31	50.6	0.31	84.7	0.23

Note: Analyses controlled for income, age, race/ethnicity, and gender.

^*Prevalence is estimated among current smokers (i.e., the denominators for prevalence calculations are number of current smokers). Approximate unweighted N = 230,000, although sample size varies across variables due to missing data.

Receipt of a doctor’s advice to quit was less consistent across urban/rural definitions. By the Census Bureau definition, there was little difference in advice by urban/rural status. By the OMB and RUCA definitions, receipt was lower among rural populations. And by the Isolation definition, the association between rurality and advice appeared somewhat curvilinear, with advice highest among the more suburban populations, and lowest among the most rural and urban populations. At all levels and for all definitions, receipt of a doctor’s advice to quit never exceeded 50% among current smokers.

Estimates of support for secondhand smoke policies were lower among rural populations as defined by the Census Bureau, OMB, and Isolation scale (for RUCA, prevalence was lowest among the Small Town category). In other words, support for smokefree policies tended to be lower among rural individuals.

Ranking by Relative Model Fit

Table 4 presents the regression results for all tobacco-related outcomes in terms of AIC value, as well as the rankings for lowest AIC value. Findings indicated that, across all outcomes, the OMB definition was consistently the best-ranked measure. Thus, the OMB was best at explaining variation in the tobacco-related outcomes. The most sensitive Isolation score (at δ=0.98) was generally the second-best measure, although it performed poorly for the home smoking ban outcome. Finally, the Census Bureau and RUCA definitions generally performed the worst, being ranked at 5^th and 6^th place for most outcomes.

Table 4.

AIC by tobacco-related outcome and rank within outcome: United States (age adjusted, replicate weighted). Each AIC value represents a separate logistic regression, such that six regressions were conducted per outcome. Approximate unweighted N = 230,000, although sample size varies across variables due to missing data.

	Behavior				Doctor Advice		Support for Secondhand Smoke Policies
	Current Smoker		Current Menthol Smoker		Doctor advise to quit		Smoking in bars, cocktail lounges, and clubs		Smoking in casinos		Smoking in the home
Rurality Definition	AIC	Rank	AIC	Rank	AIC	Rank	AIC	Rank	AIC	Rank	AIC	Rank
Census Bureau	526,600,000	6	72,760,000	6	53,550,000	6	636,900,000	5	640,700,000	5	463,500,000	2
OMB	487,000,000	1	66,410,000	1	48,210,000	1	597,100,000	1	601,000,000	1	426,700,000	1
RUCA	526,600,000	5	72,640,000	5	53,530,000	5	637,200,000	6	640,900,000	6	463,800,000	3
Isolation Score Category - δ=0.94	525,300,000	4	72,590,000	4	53,520,000	3	635,500,000	4	640,200,000	4	464,000,000	4
Isolation Score Category - δ=0.96	525,100,000	3	72,570,000	3	53,530,000	4	635,300,000	3	640,000,000	3	464,100,000	5
Isolation Score Category - δ=0.98	525,100,000	2	72,540,000	2	53,520,000	2	635,100,000	2	639,900,000	2	464,200,000	6

Note: AIC values were rounded per Census Bureau RDC policy (ranks were calculated prior to rounding). Analyses controlled for income, age, race/ethnicity, and gender.

Discussion

Using restricted-use data from the TUS-CPS, this study assessed multiple urban/rural taxonomies (Census Bureau, OMB, RUCA, and Isolation) to determine which is preferable for identifying rural risk for several tobacco-related health outcomes. These different definitions vary in how they classify “rural,” in how they are constructed, and in their inherent strengths and weaknesses. As illustrated by the descriptive analyses, they also differ in the extent to which they over- or under-classify “rural.” Side-by-side comparison among these measures generally tells a consistent story: in rural areas, current smoking is higher, menthol smoking is lower, doctors’ advice to quit is lower, and support for smokefree policies is lower. Yet the analyses also indicate that certain measures are more sensitive to detecting urban/rural differences. In particular, the OMB definition appears to be best at capturing urban/rural tobacco-related disparities, followed by the Isolation scale. It is worth noting that both these measures used strict, less-inclusive criteria for what constitutes “rural” and employed some assessment of geographic proximity/adjacency to urban areas. Thus, it is possible that policies using alternative definitions (i.e., those that over-classify rural) are siphoning resources from the rural areas that need them the most. It is also worth noting that the OMB definition was the only county-based (rather than tract-based) definition and may have better sensitivity to the impact of county-based policies. Although some research on other health topics has demonstrated how outcomes differ based on urban/rural definitions,^18,19,22,23 this is the first study, to our knowledge, that has rated their performance.

Findings additionally demonstrated the heterogeneity across urban/rural measures. Most striking was that each definition provided very different estimates for the prevalence of the U.S. population classified as “rural” (ranging from 3% for the RUCA to 23% for the Isolation scale). Likewise, although similar patterns of rural disparities appeared across the tobacco-related outcomes examined, there were also noteworthy variations. For example, unlike the other definitions, the RUCA definition found support for secondhand smoke policies to be most negative among the Small Town (vs. the Rural) category. This type of nuances aligns with other arguments that geography shouldn’t be thought of as a simple dichotomy (i.e., urban vs. rural) nor as a linear continuum.¹⁶ Overall, findings emphasize that the varying definitions of “rural" presents a consequential measurement issue with implications for research into rural health. There is a need for best-practice guidelines to direct researchers on which urban/rural definition is appropriate for which situations.

Strengths and Limitations

A strength of this study was its ability to pair multiple urban/rural definitions with multiple tobacco-related outcomes from a large, nationally-representative sample. However, due to the restricted nature of the TUS-CPS data, there were limitations to the types and quantity of analyses that could be conducted at the Census Bureau RDC. This study was thus unable to assess additional tobacco-related outcomes, such as cessation or e-cigarette use. We were also unable to assess additional urban/rural measures or versions of the measures, such as RUCC or NCHS. As data were cross-sectional, assumptions about causality should be made with caution.

We cannot identify what caused the relationship between an urban/rural measure and a particular tobacco outcome. The study was also unable to determine which elements of the urban/rural definitions (e.g., population density, commuting flow) made some superior to others. The measures themselves likewise had inherent differences in their construction (e.g., using census tracts vs. counties, dichotomous vs. four levels) making “apples to apples” statistical comparisons impossible. Yet as this was an applied endeavor, the study was able to provide practical information on the performance of existing urban/rural definitions that were, for the most part, widely in use. It is important to note that this study was based on the premise that a preferable definition of “rural,” in the context of tobacco research, is the one that best distinguishes urban vs. rural tobacco-related differences. Yet there are many factors involved in determining what is rural—population density, geography, industry, local culture, etc. The preferred measure ultimately depends on the research question being posed. Future work should assess additional tobacco-related outcomes, to determine whether certain types of measures are better suited to particular urban-rural definitions.

This study found that the best-fitting Isolation scale was one with a very high delta parameter (0.98), and thus a very strict, less-inclusive criteria for “rural.” Previous work, however, found a better fit with a lower delta parameter (0.86).¹⁴ This discrepancy could be partly due to non-linearity of some of the present study’s outcome measures, which would be better suited to categorized, rather than continuous, urban/rural measures in regression analyses. Further work is needed to continue investigating the Isolation scale, with attention not only to what outcomes are selected, but how they are assessed.

Conclusions

Overall, this study emphasizes how a single definition of “rural” does not exist; nor is there a standardized approach for when to use which urban/rural definition. Although the present findings do not provide a definitive answer for which urban/rural definition should be used across all fields of research, they do suggest that, for many tobacco-related outcomes, the OMB and Isolation measures could be preferable. Findings such as these should be used in the development of methodological guidelines that clarify when various definitions of urban/rural should be selected. By better clarifying which measures are appropriate for which situations, we can improve the precision of research on the factors contributing to rural health disparities, as well as how they could be addressed.