Research article

Identifying the insured and uninsured in rural America: an empirical discriminant analysis

  • Received: 05 February 2021 Accepted: 09 May 2021 Published: 17 May 2021
  • Purpose 

    This present study sought to investigate whether there were factors that could discriminate insured from uninsured rural Americans.

    Methods 

    Data for four groups were used: 34 uninsured, 102 government-insured (GP), 324 private- or employer-insured (PEP), and 96 both government- and private- or employer-insured (GPEP). A discriminant analysis was conducted on the four groups, using group membership as the dependent variable; age, education, income, attitude to insurance, emergency room visit, chronic disease prevalence were the independent variables.

    Findings 

    The analysis yielded three discriminant functions, however the only significant function was the one that discriminated the PEP-insured individuals from the other groups. About 48% of the cases were classified correctly with the significant discriminant function.

    Conclusion 

    The findings of this study can serve as a baseline for future research seeking to eradicate barriers to getting health insurance among the uninsured in rural America.

    Citation: Promise Tewogbola, Norah Aung. Identifying the insured and uninsured in rural America: an empirical discriminant analysis[J]. AIMS Public Health, 2021, 8(3): 421-427. doi: 10.3934/publichealth.2021032

    Related Papers:

    [1] Jacqueline C Wiltshire, Kimberly R Enard, Edlin Garcia Colato, Barbara Langland Orban . Problems paying medical bills and mental health symptoms post-Affordable Care Act. AIMS Public Health, 2020, 7(2): 274-286. doi: 10.3934/publichealth.2020023
    [2] Luke Stanisce, Donald H Solomon, Liam O'Neill, Nadir Ahmad, Brian Swendseid, Gregory J Kubicek, Yekaterina Koshkareva . Transportation considerations in underserved patient populations receiving multidisciplinary head and neck cancer care. AIMS Public Health, 2024, 11(4): 1125-1136. doi: 10.3934/publichealth.2024058
    [3] LacreishaEjike-King, RashidaDorsey . Reducing Ex-offender Health Disparities through the Affordable Care Act: Fostering Improved Health Care Access and Linkages to Integrated Care. AIMS Public Health, 2014, 1(2): 76-83. doi: 10.3934/publichealth.2014.2.76
    [4] Swapna Reddy, Matthew Speer, Mary Saxon, Madison Ziegler, Zaida Dedolph, Siman Qaasim . Evaluating network adequacy of oral health services for children on Medicaid in Arizona. AIMS Public Health, 2022, 9(1): 53-61. doi: 10.3934/publichealth.2022005
    [5] Rizwana Biviji, Nikita Vora, Nalani Thomas, Daniel Sheridan, Cindy M. Reynolds, Faith Kyaruzi, Swapna Reddy . Evaluating the network adequacy of vision care services for children in Arizona: A cross sectional study. AIMS Public Health, 2024, 11(1): 141-159. doi: 10.3934/publichealth.2024007
    [6] Aswin Sugunan, Rajasekharan Pillai K, Anice George . What is the coverage of your health insurance plan? An audit of hospital billing. AIMS Public Health, 2024, 11(4): 1013-1029. doi: 10.3934/publichealth.2024052
    [7] Lixian Zhong, Yidan Huyan, Elena Andreyeva, Matthew Lee Smith, Gang Han, Keri Carpenter, Samuel D Towne, Sagar N Jani, Veronica Averhart Preston, Marcia G. Ory . Predicting high-cost, commercially-insured people with diabetes in Texas: Characteristics, medical utilization patterns, and urban-rural comparisons. AIMS Public Health, 2025, 12(1): 259-274. doi: 10.3934/publichealth.2025016
    [8] Larrell L. Wilkinson, Saundra H. Glover, Janice C. Probst, Bo Cai, Lisa T. Wigfall . Psychological Distress and Health Insurance Coverage among Formerly Incarcerated Young Adults in the United States. AIMS Public Health, 2015, 2(3): 227-246. doi: 10.3934/publichealth.2015.3.227
    [9] Kelly Graff, Ye Ji Choi, Lori Silveira, Christiana Smith, Lisa Abuogi, Lisa Ross DeCamp, Jane Jarjour, Chloe Friedman, Meredith A. Ware, Jill L Kaar . Lessons learned for preventing health disparities in future pandemics: the role of social vulnerabilities among children diagnosed with severe COVID-19 early in the pandemic. AIMS Public Health, 2025, 12(1): 124-136. doi: 10.3934/publichealth.2025009
    [10] Bethany M. Kwan, Aimee B. Valeras, Shandra Brown Levey, Donald E. Nease, Mary E. Talen . An Evidence Roadmap for Implementation of Integrated Behavioral Health under the Affordable Care Act. AIMS Public Health, 2015, 2(4): 691-717. doi: 10.3934/publichealth.2015.4.691
  • Purpose 

    This present study sought to investigate whether there were factors that could discriminate insured from uninsured rural Americans.

    Methods 

    Data for four groups were used: 34 uninsured, 102 government-insured (GP), 324 private- or employer-insured (PEP), and 96 both government- and private- or employer-insured (GPEP). A discriminant analysis was conducted on the four groups, using group membership as the dependent variable; age, education, income, attitude to insurance, emergency room visit, chronic disease prevalence were the independent variables.

    Findings 

    The analysis yielded three discriminant functions, however the only significant function was the one that discriminated the PEP-insured individuals from the other groups. About 48% of the cases were classified correctly with the significant discriminant function.

    Conclusion 

    The findings of this study can serve as a baseline for future research seeking to eradicate barriers to getting health insurance among the uninsured in rural America.



    Individuals resident in rural regions of the US find it challenging to access healthcare due to the fewer healthcare providers in rural regions, limited health insurance coverage provided by employers, long distance to healthcare facilities, among other reasons [1],[2]. The Affordable Care Act (ACA) was enacted to address some of these challenges and consequently make healthcare more accessible to individuals, irrespective of whether they were resident in urban or rural areas. This was achieved by making provisions to ensure that insurance companies were unable to use price discrimination tactics to discourage people with pre-existing medical conditions from purchasing insurance coverage. In addition, the qualification process for accessing government-funded insurance for individuals below the poverty line was made to be less restrictive. All health insurance plans were also expanded to provide additional health benefits, such as mental health services, as well as preventative care [3].

    Despite the strides made by the ACA to improve access to healthcare, people continue to remain uninsured. For instance, Cha and Cohen [4] reported that 14.5% of adults in the US were uninsured for reasons ranging from not wanting coverage, to finding the signing up process too difficult. In addition, in the 2018 data obtained from a clinical data research network of community health centers (CHC), 23% of the 28 million patients were uninsured. Furthermore, in comparison with nonelderly individuals who are resident in urban areas, nonelderly individuals in rural America continue to be uninsured at a higher rate [5].

    Although the consensus is that, generally speaking, rural Americans have higher uninsured rates than their urban counterparts do, few studies have investigated the specific factors that predict whether a rural resident will be insured or not. This present study attempts to fill that void in knowledge by answering two questions: First, whether there are variables that can reliably predict whether a rural resident is insured or not; and second, based on the discriminant functions created by those variables, what proportion of cases will be correctly identified. An understanding of the factors that distinguishes the insured from the uninsured in rural America can be useful in assisting policymakers and key decision makers in further reducing barriers to obtaining health insurance among the uninsured. This will be crucial to seeing the number of uninsured rural Americans drop, even as more Americans enjoy access to quality healthcare, regardless of their location.

    Secondary data was obtained from a predominantly white, rural county in a Midwestern state, where the local health department had conducted a health needs assessment survey. After dropping participants with missing responses, the total usable N was 556 (444 women, 110 men and 2 unspecified; 34 uninsured, 102 government-insured (GP), 324 private- or employer-insured (PEP), and 96 both government- and private- or employer-insured (GPEP)).

    The needs assessment survey was designed and administered by the county health department. After providing consent, participants completed either the written or electronic form of the survey. The survey was conducted to fulfil an administrative requirement, rather than for the sole purpose of research. As a consequence, assessments of the reliability and validity of measures were unavailable.

    Age: This describes the age of the participant. Choices ranged from “Less than 18 years” to “65+”. This was coded as “0” to “5”. Though data was collected as categorical, this variable was treated as continuous for analyses.

    Education: This describes the highest level of education attained by the respondent. Choices ranged from “High school without diploma” to “Graduate level degree”, which were coded as “1” to “7”. This variable was also treated as continuous for analytical purposes.

    Income: This describes the respondent's combined household income for the previous year. Choices ranged from “Less than $20000” to “$100000 or more” which was coded as “1” to “5”. This variable was also treated as continuous for analysis.

    Attitude towards insurance: This assesses the respondent's perception of top insurance-related problems in the community. The score was computed from respondents' answers to the following questions: (1) Whether the respondent sees the ability to pay for care as one of the most pressing problems in the community; (2) Whether the respondent sees a lack of health insurance as one of the most pressing health problems in the community; (3) Whether the respondent sees a lack of health insurance as one of the most pressing health problems in the community. The scores ranged from “0” to “3” depending on the number of the aforementioned questions the respondent answered in the affirmative.

    Emergency room visit: This asks how many times in the past 12 months the respondent or someone in the respondent's household had used a hospital emergency room. Choices available to respondents were “None”, “1–2 times”, “3–5 times” and “6 or more times” which was coded as “0”, “1”, “3” and “5” respectively. This variable was also treated as continuous for analysis.

    Chronic disease prevalence: This asks about the number of chronic diseases such as hypertension, diabetes, and obesity, which the respondent currently has. Scores are weighted based on their prevalence rates in the US as reported by Buttorff, Ruder, and, Bauman [6].

    Insurance status: This asks whether or not the respondent has insurance coverage, as well as the type of insurance coverage he/she has. The available categories were “uninsured”, “government-provided (GP) insurance”, “private- or employer-provided (PEP) insurance”, and “both government- and private- or employer-provided (GPEP) insurance”.

    A direct discriminant analysis was performed using six variables as predictors of membership in 4 groups. Predictors were age, highest education level, income, attitude towards insurance, time of most recent visit to hospital emergency room, and chronic disease prevalence scores. Groups were uninsured, government-provided (GP) insurance, private- or employer-provided (PEP) insurance, and both government- and private- or employer-provided (GPEP) insurance. A discriminant analysis was used instead of a multivariate analysis of variance (MANOVA), because the research focused on predicting membership in naturally occurring groups, rather than those formed by random assignment. IBM SPSS statistical package (v. 25) was used for all analyses.

    Three more cases were excluded from the analyses because they were missing at least one discriminating variable. However, the characteristics of the dropped cases did not significantly differ from the cases eventually analyzed. Missing data appeared to be randomly scattered throughout groups and predictors. For the remaining 553 cases, evaluations of assumptions of multicollinearity were satisfactory. Z-scores for skewness for highest education level, income, time of most recent visit to hospital emergency room, and chronic disease prevalence scores were greater than 3.3 and were consequently transformed to approach normality. Additionally, statistically significant homogeneity of variance-covariance matrices (p < 0.10) was observed.

    ANOVA results indicated that age (F(3,549) = 27.37, p < 0.001), highest education level (F(3,549) = 8.62, p < 0.001), income (F(3,549) = 26.72, p < 0.001), time of most recent visit to hospital emergency room (F(3,549) = 3.54, p < 0.05), and chronic disease prevalence scores (F(3,549) = 0.83, p < 0.001) all differed statistically between groups.

    Three discriminant functions were calculated, with a combined Wilks' Lambda = 0.729, X2(18, N = 553) = 172.87, p < 0.001, accounting for 27.1% of the relationship between the predictors and groups. After the removal of the first and then the second function, the combined Wilks' Lambda increased to 0.978, χ2(10, N = 553) = 12.33, p = 0.26, and then 0.992, χ2(4, N = 553) = 4.59, p = 0.33. Canonical R = 0.504 for the first discriminant function, 0.119 for the second discriminant function, and 0.091 for the third discriminant function. Eigenvalue scores for each of the three discriminant functions were 0.341, 0.014 and 0.008 which respectively accounted for 93.8%, 3.9% and, 2.3% of the between-group variability.

    The first discriminant function separated the PEP-insured from the other three groups, while the second discriminant function separated the uninsured from the other three groups. Since the Wilks' Lambda for the second and third discriminant functions were not significant, they are not be interpreted, although their functions at group centroids are shown in Table 1. The structure matrix of correlations between predictor and discriminant functions seem to suggest that the best predictors for distinguishing between the PEP-insured and the other groups (first discriminant function) are age (0.654), income (−0.648) and chronic disease prevalence scores (0.478). Because these variables are all related to events more likely to occur at a specific phase of life, this first dimension was labelled “chronophilic factors”. The PEP-insured were younger (mean = 3.85, SD = 1.03) than the uninsured (mean = 4.64, SD = 1.37), the GP-insured (mean = 4.90, SD = 1.51), and the GPEP-insured (mean = 4.74, SD = 1.41). Additionally, the PEP-insured were higher earners (mean = 3.37, SD = 1.61) than the uninsured (mean = 2.97, SD = 1.49), the GPEP-insured (mean = 2.43, SD = 1.64) and the GP-insured (mean = 1.95, SD = 1.12). Furthermore, the PEP-insured had lower chronic disease prevalence scores (mean = 6.55, SD = 10.48) than the uninsured (mean = 7.28, SD = 7.89), the GP-insured (mean = 13.73, SD = 13.95) and the GPEP-insured (mean = 13.25, SD = 13.31).

    Table 1.  Unstandardized canonical discriminant functions evaluated at group means.
    Group Function
    1 2 3
    Uninsured 0.136 0.461 −0.077
    GP-insured 0.884 −0.076 −0.121
    PEP-insured −0.472 −0.027 −0.006
    GPEP-insured 0.610 0.011 0.175

     | Show Table
    DownLoad: CSV

    Thus, the four groups differ most notably on their age, income and chronic disease prevalence scores. Individuals who are PEP-insured are the highest earners, followed by the uninsured, the GPEP-insured and the GP-insured. Also, individuals who are PEP-insured tend to be younger than all other insurance groups and are less likely to be bothered by chronic diseases when compared with other groups.

    As shown in Table 2, the classification analysis for the total usable sample of 553 respondents showed correct classification in 267 cases (48.3%). Although 58.4% of the respondents were PEP-insured, the classification scheme classified 35.1% of the respondents as PEP-insured (194/553 from the Cross-validation table). That said, PEP-insured respondents were still more likely to be correctly classified (60.1% correct classification) than the uninsured (48.5% correct classification), GP-insured (41.6% correct classification), and GPEP-insured (15.6% correct classification).

    Table 2.  Results of classification analysis.
    Group Predicted Group Membership
    Number of Cases Uninsured GP-insured PEP-insured GPEP-insured
    Uninsured 33 16 (48.5%) 6 (18.2%) 7 (21.2%) 4 (12.1%)
    GP-insured 101 24 (23.8%) 42 (41.6%) 18 (17.8%) 17 (16.8%)
    PEP-insured 323 64 (19.8%) 40 (12.4%) 194 (60.1%) 25 (7.7%)
    GPEP-insured 96 20 (20.8) 32 (33.3%) 29 (30.2%) 15 (15.6%)

     | Show Table
    DownLoad: CSV

    The main objective of this study was to investigate whether there were discriminant functions that could distinguish insured rural Americans from their uninsured counterparts. The analyzed data did not reveal any significant discriminant functions delineating the insured from the uninsured. However, the discriminant function separating the PEP-insured from other groups can provide some insights that predict membership in this group, even though it accounted for about 27% of the variance in the groups.

    The data analysis revealed that income, age, and prevalence of chronic diseases were the largest contributors to the discriminant function separating members in the PEP-insured group from the other groups. The analyzed data showed that the PEP-insured were the highest income earners. Since members in this group have higher levels of income in comparison to other groups, one can assume that they were not poor. On the other hand, members of the GP- or GPEP-insured groups were eligible for government-provided insurance because they were low-income earners, disabled, or elderly and are unable to command an income comparable to their counterparts in the PEP-insured group [7]. Studies have shown that poverty is one of the major contributors to a lack of insurance. For instance, Makowska [8] reported that 26% of the individuals below the US poverty line did not have health insurance. Wippold and colleagues [9] also provided evidence showing that people with low socioeconomic statuses struggle to “pay or co-pay...for services associated with health care”, which may further increase the likelihood of them delaying in seeking care as though they did not have insurance. Cha and Cohen [4] also reported that people's perception of their inability to afford coverage was a contributor to their status as uninsured.

    Members in the PEP-insured group were also younger than their counterparts in other groups. This is plausible from the economic perspective of rural employers who may prefer to employ younger people because of their strength and potential for prolonged productivity. In exchange for their commitment to pursue employment in a rural region, younger people may be incentivized with health insurance coverage. This, however, is inconsistent with Makowska [8] who reported that people aged between 26 and 34 years were the most likely to be uninsured in comparison to other age groups.

    Prevalence of chronic diseases was much lower in the PEP-insured group than in any other group. This may be because the accommodations of the PEP-insurance provides access to healthcare services that better helps this category of people to manage and ultimately cure their chronic diseases. Alternatively, members of the PEP-insured group may just be overall healthier than their counterparts in other groups, are potentially more productive economically, and are thus able to get insurance-providing employment. This is consistent with the suggestion of Fernandez-Lazaro and colleagues [10] that uninsured populations are more vulnerable to chronic diseases.

    The current study has some limitations. As mentioned earlier, non-researchers designed the needs assessment survey for the purpose of fulfilling an administrative requirement. As a result, the validity and reliability of the survey tool could not be determined. Furthermore, data was self-reported by the respondents and analyses were conducted with the assumption that the survey questions were answered honestly and accurately. In addition, some of the variables of interest were collected as categorical, but analyzed as continuous. This, and the violation of some statistical assumptions, could affect the findings of the study, particularly the proportion of cases that the significant discriminant functions were able to predict accurately. Future research with data from more counties will be useful in providing more insight about the discriminant functions that separate insured, rural Americans from their uninsured counterparts. In the same vein, further studies can investigate discriminant functions that predict non-insurance when more rural counties are taken into consideration. Similar studies can also be conducted to find out the discriminant functions that can predict rural insurance coverage between genders, or amongst age groups.

    Understanding the factors that discriminate insured rural Americans from their uninsured counterparts will prove to be vital to the creation of a future where everyone has access to healthcare. Although, the findings of this study did not show the factors that discriminated the uninsured from the insured, it did demonstrate that there were specific factors that distinguished the PEP-insured from others. By using this study as a baseline, and investigating further in future studies, policymakers and key stakeholders can design targeted interventions that can reliably increase insurance enrollment among uninsured, rural Americans.


    Acknowledgments



    The authors would like to thank Dr. Tamara Kang for introducing us to the statistical approach used in this article.

    Conflicts of interest



    The author declares no conflicts of interest.

    [1] Benitez JA, Seiber EE (2018) US health care reform and rural America: results from the ACA's Medicaid expansions. J Rural Health 34: 213-222. doi: 10.1111/jrh.12284
    [2] Gong G, Phillips SG, Hudson C, et al. (2019) Higher US rural mortality rates linked to socioeconomic status, physician shortages, and lack of health insurance. Health Aff 38: 2003-2010. doi: 10.1377/hlthaff.2019.00722
    [3] Huguet N, Angier H, Hoopes MJ, et al. (2019) Prevalence of pre-existing conditions among community health center patients before and after the Affordable Care Act. J Am Board Fam Med 32: 883-889. doi: 10.3122/jabfm.2019.06.190087
    [4] Cha AE, Cohen RA (2020) Reasons for Being Uninsured Among Adults Aged 18–64 in the United States, 2019. NCHS Data Brief 1: 1-8.
    [5] Foutz J, Artiga S, Garfield R (2017)  The role of Medicaid in rural America Washington, DC: Kaiser Family Foundation.
    [6] Buttorff C, Ruder T, Bauman M (2017)  Multiple chronic conditions in the United States Santa Monica, CA: Rand. doi: 10.7249/TL221
    [7] Berkowitz SA, Terranova J, Hill C, et al. (2018) Meal delivery programs reduce the use of costly health care in dually eligible Medicare and Medicaid beneficiaries. Health Aff 37: 535-542. doi: 10.1377/hlthaff.2017.0999
    [8] Makowska M (2018) Uninsured America: problem of lack of health insurance cover-age in the US. JPHNMR 253: 33-39.
    [9] Wippold GM, Nmezi N, Williams JL, et al. (2020) An Exploratory Study to Understand Factors Associated with Health-related Quality of Life Among Uninsured/Underinsured Patients as Identified by Clinic Providers and Staff. J Prim Care Community Health 11: 2150132720949412. doi: 10.1177/2150132720949412
    [10] Fernandez-Lazaro CI, Adams DP, Fernandez-Lazaro D, et al. (2019) Medication adherence and barriers among low-income, uninsured patients with multiple chronic conditions. Res Social Adm Pharm 15: 744-753. doi: 10.1016/j.sapharm.2018.09.006
  • Reader Comments
  • © 2021 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Metrics

Article views(2906) PDF downloads(113) Cited by(0)

Figures and Tables

Tables(2)

Other Articles By Authors

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog