Atl Econ J (2013) 41:8991DOI 10.1007/s11293-012-9342-2ANTHOLOGYSocial Capital and Income Inequality in the UnitedStatesRati RamPublished online: 17 October 2012# International Atlantic Economic Society 2012Many scholars have explored in recent years various correlates and consequences ofsocial capital along with discussions of the concept. For example, relationship ofsocial capital with population happiness, health, income, economic growth, andhuman development has been researched by several scholars. However, very fewstudies have considered the relationship between social capital and income inequality.One exception to that is the recent work by Robison et al. (Journal of SocioEconomics, 2011) which proposed a theoretical link between social capital andincome distribution and conducted an empirical exploration for the U.S. states forthe census years 1980, 1990, and 2000. Their key measure of social capital wassomewhat narrowly focused on percent of households headed by a single female withchildren. Given the importance of the topic, it is of interest to work with a broader andmore common proxy for social capital and also to use more recent inequality indexes.The theoretical framework suggested by Robison et al. linked greater social capitalwith increased trade across individuals or households, which raises average incomebut has an ambiguous effect on income distribution. It is possible to propose a simplerconceptual reasoning that generates a sharper implication about income inequality. Itis reasonable to suggest that higher social capital or social trust is associated with astronger sense of fairness and consideration for others, particularly relative to givingand receiving compensation for market work. Such a sense of fairness and consideration may be expected to mitigate wage and income inequalities. It is thus plausible topostulate increased social capital as an equalizer. This short paper pursues theforegoing theme by using fairly good social capital data and the most recent information on income inequality for the U.S. states.Social capital data are taken from the compilation by Bjornskov (Applied Researchin Quality of Life, 2008). The numbers are for social trust, which is a primaryindicator of social capital at the macro level, and are averaged over the period19901998. The inequality measure is state-level household Gini index fromR. Ram (*)Department of Economics, Illinois State University, Normal, IL 61790-4200, USAe-mail: rram@ilstu.edu90R. RamAmerican Community Survey (ACS) for the years 20062010 available at the CensusBureau’s American Fact Finder. By way of a simple control variable, averagehousehold income from ACS for the year of 2005 is also taken from American FactFinder. To provide a feel for the data, simple descriptive statistics for the variables areshown below.MeanSDMin.Max.Social trust (19901998)31.373.2324.6937.97Gini index (20062010)0.450.020.410.54Mean household income (2005, 000$)61.768.7745.9887.47The following regression estimates show the association between logarithms ofsocial capital (LTRUST), income inequality (LGINI), and average household income(LY05), with robust t-statistics in parentheses.LGINI ¼ À0:594 À 0:227ðLTRUSTÞ þ 0:053ðLY05ÞðÀ1:80Þ ðÀ5:76Þð1:55ÞLGINI ¼ À0:085À0:207ðLTRUSTÞðÀ0:72Þ ðÀ5:95ÞR2 : 0:31R2 : 0:34N ¼ 48N ¼ 48The estimates show that social capital is a highly significant equalizer. 1 %increase in social capital (trust) is expected to lower the Gini index by about 0.20 %.The following observations should also be of interest.1. The role of income is marginal and lacks statistical significance. The sign on theterm indicates a tendency for income inequality to increase with income, supporting several studies that have documented increasing inequality since the 1970s.2. Although the estimates are reported here for logarithmic versions of the variables,an almost identical position emerges if the variables are entered linearly. Severalother variants of the model yield a very similar scenario.3. There is lack of indication of a significant quadratic relation between socialcapital and income inequality. If a quadratic term for trust is added, adjustedR2 goes down, neither LTRUST nor its square is significant at any meaningfullevel, and both t-statistics are below unity.4. The parsimonious model and the possibility of a feedback from income inequality can lead to a reasonable concern about the quality of the estimates. However,several considerations should mitigate that concern. First, the social-capitalvariable is for the period 19901998 and is temporally predetermined relativeto the Gini index, which is for 20062010. Second, Halbert White’s well-knowntest, which has the joint null hypothesis of homoscedastic error and nomisspecification, is not rejected. If squares of residuals from the model (withincome) are regressed on squares and cross-products of the regressors, chi-squarestatistic with six degrees of freedom is 5.05, which is not significant at anySocial Capital and Income Inequality91sensible level. Third, the well-known RESET test also suggests lack of a significant specification error. In a simple version of the test, when the square of thepredicted Gini is added to the regression, t-statistic for the squared term is 0.87.There is thus considerable evidence of social capital being a significant equalizerin the context studied. Policy measures that augment social capital might, besidesyielding other benefits suggested in the literature, also be expected to mitigate therising trend in inequality in the United States.Copyright of Atlantic Economic Journal is the property of Springer Science & Business Media B.V. and itscontent may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder’sexpress written permission. However, users may print, download, or email articles for individual use.RESEARCH AND PRACTICEIncome Inequality in Health at All Ages: A Comparisonof the United States and EnglandMelissa L. Martinson, PhDPopulation health is worse in the United Statesthan England, despite the much higher level ofhealth care spending in the United States.1—3Well-documented health differentials betweenthe 2 countries exist for a wide variety of healthmeasures at all ages.3 However, questionsremain about the extent of cross-country differences in health disparities, in particularwhether income inequalities in health arehigher in the United States than in Englandacross the life span.Health comparisons between the UnitedStates and England are interesting because,despite many societal similarities, there aredifferences in health care provision, socialprotection policies, and societal inequality between the 2 countries.4—6 In particular, previous studies have postulated that differencesin health care systems between the UnitedStates and England (as well as other Europeancountries) may account for the relativelypoorer health in the United States as well asthe greater health inequalities among Americans.2,7—9 Additionally, whereas both countrieshave liberal, residual welfare states, Britainhas slightly lower income inequality anda greater focus on alleviating poverty, particularly among children, although it still lagsbehind many other European countries.5,10A handful of studies have examined themagnitude of socioeconomic disparities inhealth in the United States and England, andthe results are decidedly mixed. Banks et al.concluded that income- and education-basedhealth gradients among older adults aresteeper in the United States than in England,1whereas Avendano et al. found that thewealth gradient for older adults is similar inthe 2 countries.2 A series of articles examiningthe income gradient in health among childrenhas produced conicting ndings.7,11,12 Onecomparative study of self-rated health by income, occupational, and poverty status in theUnited States and United Kingdom includedmiddle-age adults; it found better health in theObjectives. I systematically examined income gradients in health in the UnitedStates and England across the life span (ages birth to 80 years), separately forfemales and males, for a number of health conditions.Methods. Using data from the National Health and Nutrition ExaminationSurvey for the United States (n = 36 360) and the Health Survey for England (n =55 783), I calculated weighted prevalence rates and risk ratios by income level forthe following health risk factors or conditions: obesity, hypertension, diabetes,low high-density lipoprotein cholesterol, high cholesterol ratio, heart attack orangina, stroke, and asthma.Results. In the United States and England, the income gradients in health arevery similar across age, gender, and numerous health conditions, and are robustto adjustments for race/ethnicity, health behaviors, body mass index, and healthinsurance.Conclusions. Health disparities by income are pervasive in England as well asin the United States, despite better overall health, universal health insurance,and more generous social protection spending in England. (Am J Public Health.2012;102:20492056. doi:10.2105/AJPH.2012.300929)United Kingdom than in the United States,as well as a greater likelihood of healthimproving over time in the United Kingdom.13However, that study did not examine theincome gradient in health. No study to datehas compared socioeconomic gradients inhealth throughout the life span in the UnitedStates and England.There is reason to believe that the incomegradient in health is largest in middle to lateradulthood, because the income gradient inhealth widens with age among children andnarrows with age among the elderly because ofincreased mortality among low-income people.11,12,14,15 This widening of income disparitieswith age would likely be similar in the UnitedStates and England if the income gradientage pattern is being driven by a higher vulnerability to health shocks among low-incomeindividuals than among high-income individuals.However, if an increase in income disparitieswith age is due to low-income individuals lackof ability to respond to health shocks (e.g.,through lack of insurance), one would expectto see a more rapid increase in income-basedNovember 2012, Vol 102, No. 11 | American Journal of Public Healthhealth inequalities in the United States than inEngland because of the highly variable UShealth care system.The extent to which income gradients inhealth and health trajectories differ in the 2countries by age is not known, but it representsan important area of inquiry for understandingthe processes leading to the well-documentedcross-country differences in health. In thisstudy, I describe and compare the extent ofincome-based socioeconomic gradients inhealth in the United States and England frombirth to 80 years, for both females and males,using a large set of biological and self-reportedhealth measures. This study provides a comprehensive description of the magnitude ofincome inequalities in health in the 2 countries.METHODSThe National Health and Nutrition Examination Survey (NHANES) for the United Statesand the Health Survey for England (HSE) wereused in this study. Both are large, nationallyrepresentative health surveys that haveMartinson | Peer Reviewed | Research and Practice | 2049RESEARCH AND PRACTICEcomparable measures of health assessedthrough both physical examinations and interviews.The NHANES is a comprehensive surveyconducted by the National Center for HealthStatistics in the United States continuouslysince 1999.16 For the analyses presented here,I used data from years 1999 to 2006 of thecontinuous survey. Of the 41 474 participantsfrom 1999 to 2006, individuals aged olderthan 80 years were removed. Additionally,about 8% of the sample was missing incomedata. The nal analytic sample was 36 360.Sample sizes varied by health measure becausesome conditions were assessed only for certainage groups.The HSE is an annual cross-sectional surveyof private households in England conducted bythe Joint Health Surveys Unit of the NationalCentre for Social Research and UniversityCollege London.17—20 I used the 2003—2006surveys for these analyses because, starting in2003, weights became available making itpossible to pool multiple years of data whilemaintaining the representativeness of the English population. The number of respondentsin the 2003—2006 surveys was 71 717.The analysis sample excluded individuals olderthan 80 years. Approximately 19% of respondents had missing income data and wereexcluded from the analytic sample. The nalanalytic sample was 55 783. It is worth notingthat older adults were more likely to havemissing data on income in both the US andEnglish samples. Some biological measureswere collected from representative subsamplesand some questions were asked only of participants in certain age groups.including the United States and GreatBritain.22I dened hypertension as a mean systolicblood pressure of 140 millimeters of mercuryor higher, mean diastolic blood pressure of90 millimeters of mercury or higher, or reportsof current treatment of hypertension with prescription medication.23 I assessed diabetesfrom glycosylated hemoglobin tests (HbA1c6.5%).24 I categorized HDL as low (< 40 mg/dL),normal (40—59 mg/dL), or high (> 59 mg/dL);in addition, I used a binary measure of low vsnormal or high HDL.25 In the absence of a lowdensity lipoprotein cholesterol measure, I usedthe total-cholesterol-to-HDL-cholesterol ratio.26High cholesterol ratio was dened as a totalcholesterol-to-HDL-cholesterol ratio of 5:1 orabove, although results were not sensitive tothe ratio cutoff used. I used high C-reactivesensitivity protein, a biomarker for inammation,conditions were measured for individualsaged 12 years and older. An advantage of usingthe biological measures was the ability tocapture health risk among individuals who wereyoung and for whom illness was relatively rare.For adults, the categories of body mass index(BMI; dened as weight in kilograms dividedby the square of height in meters) werebased on the World Health Organizationsstandard.21 The categories were normal(BMI = 18.5—24.9 kg/m2), overweight (BMI =25—29.9 kg/m2), obese (BMI 30 kg/m2),and underweight (BMI < 18.5 kg/m2).Obesity was specically examined as a healthrisk. For children (through age 17 years),age- and gender-specic thresholds weredetermined using the International ObesityTaskforce denition of the BMI categories(normal, overweight, and obese), whichwas based on BMI curves in 6 countries,TABLE 1Sample Characteristics of Survey Respondents in the United States and England,by Income Tercile: US National Health and Nutrition ExaminationSurvey (19992006) and Health Survey for England (20032006)United States (n = 36 360)England (n = 55 783)LowMiddleHighLowMiddleHigh33.7 (0.4)35.1 (0.4)34.5 (0.4)36.1 (0.3)35.9 (0.3)35.3 (0.3)46.849.851.144.951.052.253.250.248.955.149.047.8Non-Hispanic White50.770.182.679.290.091.8Hispanic24.713.16.0NANANANANANA10.54.74.818.711.06.48.23.92.4Mean age (SE), yGender, %MaleFemaleRace/ethnicity, %aAsianNon-Hispanic Black5.95.84.92.11.41.0Measures of HealthCigarette smoking, %Other31.223.715.836.923.316.9There were several comparable healthmeasures based on physical examinations orlaboratory reports in the NHANES and HSE. Iincluded the following risk factors or conditionsin this study: obesity, hypertension, diabetes,low high-density lipoprotein (HDL) cholesterol,high cholesterol ratio, and high C-reactiveprotein. The NHANES and HSE documentation indicated that very similar protocolswere used in the 2 countries. Obesity wascalculated for respondents aged 4 to 80 years,C-reactive protein was measured for respondents aged 18 to 80 years, and the otherDrinking 5 d/wk (age 20 y), %No health insurance, %4.529.96.614.59.55.912.5NA18.7NA26.6NA012 y (US), 011 y (England)64.645.124.750.429.914.41315 y (US), 1213 y (England)27.033.330.423.126.317.58.421.644.926.643.868.12050 | Research and Practice | Peer Reviewed | MartinsonEducation, %16 y (US), 14 y (England)Note. NA = not applicable. Because obesity was categorized differently for those younger than 18 years than for adults andbecause C-reactive protein was assessed only for those at least 18 years old, the adolescent group was categorized as 1217years and the young adult group as 1834 years for measures of obesity and C-reactive protein. Unless otherwise noted, allgures pertain to individuals aged birth to 80 years.aHispanic ethnicity was not available for England (individuals who are Hispanic could have classied themselves in any of theracial groups). Asian race was not available for the United States (individuals who are Asian are included in the other race/ethnic category).American Journal of Public Health | November 2012, Vol 102, No. 11RESEARCH AND PRACTICETABLE 2Prevalence of Health Outcomes Among Female Respondents, by IncomeTercile and Age Group: US National Health and Nutrition Examination Survey(19992006) and Health Survey for England (20032006)United States (n = 36 360), %Health Outcome and Age, YearsEngland (n = 55 783), %LowMiddleHighLowMiddleHigh12190.60.30.50.00.00.0203435491.16.3**2.2*2.20.51.71.12.5**1.21.21.10.5506415.0**7.35.58.8**3.32.1658014.3**15.0**7.413.5**12.9**5.741115.3**10.910.29.36.35.6121717.3**18.0**10.510.86.09.0183435.5**31.6**21.419.2**13.0*8.43549506445.2**47.4**35.041.5*30.831.926.6**28.322.1**25.515.823.7658038.4*40.6**29.424.827.2*20.6121913.6**11.9*7.45.06.99.7203416.2**11.4*6.69.9**5.73.1354914.6**11.5*6.68.1**3.32.3506413.0**7.05.16.3**2.01.410.9**6.75.45.3*3.42.45.1**DiabetesObesityLow HDL6580High cholesterol ratio12192.43.63.34.8203412.0*6.0**12.3*7.88.7**4.92.3354921.8**14.411.412.2**7.35.3506421.8**18.6**12.117.0**11.811.0658019.3*15.413.415.911.914.41834354941.351.7**39.944.138.139.734.732.0**29.227.7*29.422.3506455.8**50.6*39.744.5**33.629.9658054.9*49.645.744.9*45.6*37.512190.40.30.51.21.10.520343.6*2.61.34.13.33.6354923.8*17.017.814.714.711.75064658053.0**78.9*50.0**76.138.670.148.0**70.5**33.865.530.061.2Birth311.7**6.44.92.91.61.041113.7*9.78.68.1*8.2*3.3121917.416.419.57.24.85.2203417.413.415.87.55.96.2354915.815.614.39.7**6.05.05064658015.6**15.1*14.2*12.39.710.79.48.77.29.86.65.8High C-reactive proteinHypertensionAsthma ever diagnosedContinuedNovember 2012, Vol 102, No. 11 | American Journal of Public Healthto classify individuals as low risk (< 1 mg/L),medium risk (1—3 mg/L), or high risk (> 3 mg/L)and to create a binary measure of high vs low ormedium health risk.27—29The self-reported health conditions werebased on participants responses to standardsurvey questions. These were chosen becauseof comparability between the 2 data sets andwere used in previous research comparinghealth in the United States and England.3Responses indicated whether the individualwas ever told by a doctor that he or she hadhad a heart attack or angina, a stroke, or asthma(in England, the HSE simply asks whether theindividual has asthma). Except asthma, all ofthese measures were available for individualsat least 20 years of age. Asthma was availablefor all ages.Age Groups and Income MeasureI categorized age into broad groups thatcorrespond to the Centers for Disease Controland Preventions Stages of Life. The categorieswere as follows: infants (birth—3 years), children (4—11 years), adolescents (12—19 years),young adults (20—34 years), middle-age adults(35—49 and 50—64 years), and older-ageadults (65—80 years).The primary independent variable of interest in this study was income-based socioeconomic status, which I constructed from thefamily income variable available in boththe HSE and NHANES at present value, adjusted by the Organisation for Economic CoOperation and Developments (OECD) squareroot equivalence scale, and then divided intoequal terciles by using the sample weights.30The square root equivalence scale has beenused in OECD publications on internationalincome inequality and poverty since 2008.Additionally, because of the pooling of multipleyears of data in the NHANES and HSE, Iadjusted the measure for cost of living to theyear 2006 using the Consumer Price Indexesfor the United States and the United Kingdom.31,32 Use of terciles rather than absoluteincome adjusted for differences in averagelevels and the income distribution acrossthe 2 countries; previous studies of olderadults have used this method.1

Atl Econ J (2013) 41:8991DOI 10.1007/s11293-012-9342-2ANTHOLOGYSocial Capital and Income Inequality in the UnitedStatesRati RamPublished online: 17 October 2012# International Atlantic Economic Society 2012Many scholars have explored in recent years various correlates and consequences ofsocial capital along with discussions of the concept. For example, relationship ofsocial capital with population happiness, health, income, economic growth, andhuman development has been researched by several scholars. However, very fewstudies have considered the relationship between social capital and income inequality.One exception to that is the recent work by Robison et al. (Journal of SocioEconomics, 2011) which proposed a theoretical link between social capital andincome distribution and conducted an empirical exploration for the U.S. states forthe census years 1980, 1990, and 2000. Their key measure of social capital wassomewhat narrowly focused on percent of households headed by a single female withchildren. Given the importance of the topic, it is of interest to work with a broader andmore common proxy for social capital and also to use more recent inequality indexes.The theoretical framework suggested by Robison et al. linked greater social capitalwith increased trade across individuals or households, which raises average incomebut has an ambiguous effect on income distribution. It is possible to propose a simplerconceptual reasoning that generates a sharper implication about income inequality. Itis reasonable to suggest that higher social capital or social trust is associated with astronger sense of fairness and consideration for others, particularly relative to givingand receiving compensation for market work. Such a sense of fairness and consideration may be expected to mitigate wage and income inequalities. It is thus plausible topostulate increased social capital as an equalizer. This short paper pursues theforegoing theme by using fairly good social capital data and the most recent information on income inequality for the U.S. states.Social capital data are taken from the compilation by Bjornskov (Applied Researchin Quality of Life, 2008). The numbers are for social trust, which is a primaryindicator of social capital at the macro level, and are averaged over the period19901998. The inequality measure is state-level household Gini index fromR. Ram (*)Department of Economics, Illinois State University, Normal, IL 61790-4200, USAe-mail: rram@ilstu.edu90R. RamAmerican Community Survey (ACS) for the years 20062010 available at the CensusBureau’s American Fact Finder. By way of a simple control variable, averagehousehold income from ACS for the year of 2005 is also taken from American FactFinder. To provide a feel for the data, simple descriptive statistics for the variables areshown below.MeanSDMin.Max.Social trust (19901998)31.373.2324.6937.97Gini index (20062010)0.450.020.410.54Mean household income (2005, 000$)61.768.7745.9887.47The following regression estimates show the association between logarithms ofsocial capital (LTRUST), income inequality (LGINI), and average household income(LY05), with robust t-statistics in parentheses.LGINI ¼ À0:594 À 0:227ðLTRUSTÞ þ 0:053ðLY05ÞðÀ1:80Þ ðÀ5:76Þð1:55ÞLGINI ¼ À0:085À0:207ðLTRUSTÞðÀ0:72Þ ðÀ5:95ÞR2 : 0:31R2 : 0:34N ¼ 48N ¼ 48The estimates show that social capital is a highly significant equalizer. 1 %increase in social capital (trust) is expected to lower the Gini index by about 0.20 %.The following observations should also be of interest.1. The role of income is marginal and lacks statistical significance. The sign on theterm indicates a tendency for income inequality to increase with income, supporting several studies that have documented increasing inequality since the 1970s.2. Although the estimates are reported here for logarithmic versions of the variables,an almost identical position emerges if the variables are entered linearly. Severalother variants of the model yield a very similar scenario.3. There is lack of indication of a significant quadratic relation between socialcapital and income inequality. If a quadratic term for trust is added, adjustedR2 goes down, neither LTRUST nor its square is significant at any meaningfullevel, and both t-statistics are below unity.4. The parsimonious model and the possibility of a feedback from income inequality can lead to a reasonable concern about the quality of the estimates. However,several considerations should mitigate that concern. First, the social-capitalvariable is for the period 19901998 and is temporally predetermined relativeto the Gini index, which is for 20062010. Second, Halbert White’s well-knowntest, which has the joint null hypothesis of homoscedastic error and nomisspecification, is not rejected. If squares of residuals from the model (withincome) are regressed on squares and cross-products of the regressors, chi-squarestatistic with six degrees of freedom is 5.05, which is not significant at anySocial Capital and Income Inequality91sensible level. Third, the well-known RESET test also suggests lack of a significant specification error. In a simple version of the test, when the square of thepredicted Gini is added to the regression, t-statistic for the squared term is 0.87.There is thus considerable evidence of social capital being a significant equalizerin the context studied. Policy measures that augment social capital might, besidesyielding other benefits suggested in the literature, also be expected to mitigate therising trend in inequality in the United States.Copyright of Atlantic Economic Journal is the property of Springer Science & Business Media B.V. and itscontent may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder’sexpress written permission. However, users may print, download, or email articles for individual use.RESEARCH AND PRACTICEIncome Inequality in Health at All Ages: A Comparisonof the United States and EnglandMelissa L. Martinson, PhDPopulation health is worse in the United Statesthan England, despite the much higher level ofhealth care spending in the United States.1—3Well-documented health differentials betweenthe 2 countries exist for a wide variety of healthmeasures at all ages.3 However, questionsremain about the extent of cross-country differences in health disparities, in particularwhether income inequalities in health arehigher in the United States than in Englandacross the life span.Health comparisons between the UnitedStates and England are interesting because,despite many societal similarities, there aredifferences in health care provision, socialprotection policies, and societal inequality between the 2 countries.4—6 In particular, previous studies have postulated that differencesin health care systems between the UnitedStates and England (as well as other Europeancountries) may account for the relativelypoorer health in the United States as well asthe greater health inequalities among Americans.2,7—9 Additionally, whereas both countrieshave liberal, residual welfare states, Britainhas slightly lower income inequality anda greater focus on alleviating poverty, particularly among children, although it still lagsbehind many other European countries.5,10A handful of studies have examined themagnitude of socioeconomic disparities inhealth in the United States and England, andthe results are decidedly mixed. Banks et al.concluded that income- and education-basedhealth gradients among older adults aresteeper in the United States than in England,1whereas Avendano et al. found that thewealth gradient for older adults is similar inthe 2 countries.2 A series of articles examiningthe income gradient in health among childrenhas produced conicting ndings.7,11,12 Onecomparative study of self-rated health by income, occupational, and poverty status in theUnited States and United Kingdom includedmiddle-age adults; it found better health in theObjectives. I systematically examined income gradients in health in the UnitedStates and England across the life span (ages birth to 80 years), separately forfemales and males, for a number of health conditions.Methods. Using data from the National Health and Nutrition ExaminationSurvey for the United States (n = 36 360) and the Health Survey for England (n =55 783), I calculated weighted prevalence rates and risk ratios by income level forthe following health risk factors or conditions: obesity, hypertension, diabetes,low high-density lipoprotein cholesterol, high cholesterol ratio, heart attack orangina, stroke, and asthma.Results. In the United States and England, the income gradients in health arevery similar across age, gender, and numerous health conditions, and are robustto adjustments for race/ethnicity, health behaviors, body mass index, and healthinsurance.Conclusions. Health disparities by income are pervasive in England as well asin the United States, despite better overall health, universal health insurance,and more generous social protection spending in England. (Am J Public Health.2012;102:20492056. doi:10.2105/AJPH.2012.300929)United Kingdom than in the United States,as well as a greater likelihood of healthimproving over time in the United Kingdom.13However, that study did not examine theincome gradient in health. No study to datehas compared socioeconomic gradients inhealth throughout the life span in the UnitedStates and England.There is reason to believe that the incomegradient in health is largest in middle to lateradulthood, because the income gradient inhealth widens with age among children andnarrows with age among the elderly because ofincreased mortality among low-income people.11,12,14,15 This widening of income disparitieswith age would likely be similar in the UnitedStates and England if the income gradientage pattern is being driven by a higher vulnerability to health shocks among low-incomeindividuals than among high-income individuals.However, if an increase in income disparitieswith age is due to low-income individuals lackof ability to respond to health shocks (e.g.,through lack of insurance), one would expectto see a more rapid increase in income-basedNovember 2012, Vol 102, No. 11 | American Journal of Public Healthhealth inequalities in the United States than inEngland because of the highly variable UShealth care system.The extent to which income gradients inhealth and health trajectories differ in the 2countries by age is not known, but it representsan important area of inquiry for understandingthe processes leading to the well-documentedcross-country differences in health. In thisstudy, I describe and compare the extent ofincome-based socioeconomic gradients inhealth in the United States and England frombirth to 80 years, for both females and males,using a large set of biological and self-reportedhealth measures. This study provides a comprehensive description of the magnitude ofincome inequalities in health in the 2 countries.METHODSThe National Health and Nutrition Examination Survey (NHANES) for the United Statesand the Health Survey for England (HSE) wereused in this study. Both are large, nationallyrepresentative health surveys that haveMartinson | Peer Reviewed | Research and Practice | 2049RESEARCH AND PRACTICEcomparable measures of health assessedthrough both physical examinations and interviews.The NHANES is a comprehensive surveyconducted by the National Center for HealthStatistics in the United States continuouslysince 1999.16 For the analyses presented here,I used data from years 1999 to 2006 of thecontinuous survey. Of the 41 474 participantsfrom 1999 to 2006, individuals aged olderthan 80 years were removed. Additionally,about 8% of the sample was missing incomedata. The nal analytic sample was 36 360.Sample sizes varied by health measure becausesome conditions were assessed only for certainage groups.The HSE is an annual cross-sectional surveyof private households in England conducted bythe Joint Health Surveys Unit of the NationalCentre for Social Research and UniversityCollege London.17—20 I used the 2003—2006surveys for these analyses because, starting in2003, weights became available making itpossible to pool multiple years of data whilemaintaining the representativeness of the English population. The number of respondentsin the 2003—2006 surveys was 71 717.The analysis sample excluded individuals olderthan 80 years. Approximately 19% of respondents had missing income data and wereexcluded from the analytic sample. The nalanalytic sample was 55 783. It is worth notingthat older adults were more likely to havemissing data on income in both the US andEnglish samples. Some biological measureswere collected from representative subsamplesand some questions were asked only of participants in certain age groups.including the United States and GreatBritain.22I dened hypertension as a mean systolicblood pressure of 140 millimeters of mercuryor higher, mean diastolic blood pressure of90 millimeters of mercury or higher, or reportsof current treatment of hypertension with prescription medication.23 I assessed diabetesfrom glycosylated hemoglobin tests (HbA1c6.5%).24 I categorized HDL as low (< 40 mg/dL),normal (40—59 mg/dL), or high (> 59 mg/dL);in addition, I used a binary measure of low vsnormal or high HDL.25 In the absence of a lowdensity lipoprotein cholesterol measure, I usedthe total-cholesterol-to-HDL-cholesterol ratio.26High cholesterol ratio was dened as a totalcholesterol-to-HDL-cholesterol ratio of 5:1 orabove, although results were not sensitive tothe ratio cutoff used. I used high C-reactivesensitivity protein, a biomarker for inammation,conditions were measured for individualsaged 12 years and older. An advantage of usingthe biological measures was the ability tocapture health risk among individuals who wereyoung and for whom illness was relatively rare.For adults, the categories of body mass index(BMI; dened as weight in kilograms dividedby the square of height in meters) werebased on the World Health Organizationsstandard.21 The categories were normal(BMI = 18.5—24.9 kg/m2), overweight (BMI =25—29.9 kg/m2), obese (BMI 30 kg/m2),and underweight (BMI < 18.5 kg/m2).Obesity was specically examined as a healthrisk. For children (through age 17 years),age- and gender-specic thresholds weredetermined using the International ObesityTaskforce denition of the BMI categories(normal, overweight, and obese), whichwas based on BMI curves in 6 countries,TABLE 1Sample Characteristics of Survey Respondents in the United States and England,by Income Tercile: US National Health and Nutrition ExaminationSurvey (19992006) and Health Survey for England (20032006)United States (n = 36 360)England (n = 55 783)LowMiddleHighLowMiddleHigh33.7 (0.4)35.1 (0.4)34.5 (0.4)36.1 (0.3)35.9 (0.3)35.3 (0.3)46.849.851.144.951.052.253.250.248.955.149.047.8Non-Hispanic White50.770.182.679.290.091.8Hispanic24.713.16.0NANANANANANA10.54.74.818.711.06.48.23.92.4Mean age (SE), yGender, %MaleFemaleRace/ethnicity, %aAsianNon-Hispanic Black5.95.84.92.11.41.0Measures of HealthCigarette smoking, %Other31.223.715.836.923.316.9There were several comparable healthmeasures based on physical examinations orlaboratory reports in the NHANES and HSE. Iincluded the following risk factors or conditionsin this study: obesity, hypertension, diabetes,low high-density lipoprotein (HDL) cholesterol,high cholesterol ratio, and high C-reactiveprotein. The NHANES and HSE documentation indicated that very similar protocolswere used in the 2 countries. Obesity wascalculated for respondents aged 4 to 80 years,C-reactive protein was measured for respondents aged 18 to 80 years, and the otherDrinking 5 d/wk (age 20 y), %No health insurance, %4.529.96.614.59.55.912.5NA18.7NA26.6NA012 y (US), 011 y (England)64.645.124.750.429.914.41315 y (US), 1213 y (England)27.033.330.423.126.317.58.421.644.926.643.868.12050 | Research and Practice | Peer Reviewed | MartinsonEducation, %16 y (US), 14 y (England)Note. NA = not applicable. Because obesity was categorized differently for those younger than 18 years than for adults andbecause C-reactive protein was assessed only for those at least 18 years old, the adolescent group was categorized as 1217years and the young adult group as 1834 years for measures of obesity and C-reactive protein. Unless otherwise noted, allgures pertain to individuals aged birth to 80 years.aHispanic ethnicity was not available for England (individuals who are Hispanic could have classied themselves in any of theracial groups). Asian race was not available for the United States (individuals who are Asian are included in the other race/ethnic category).American Journal of Public Health | November 2012, Vol 102, No. 11RESEARCH AND PRACTICETABLE 2Prevalence of Health Outcomes Among Female Respondents, by IncomeTercile and Age Group: US National Health and Nutrition Examination Survey(19992006) and Health Survey for England (20032006)United States (n = 36 360), %Health Outcome and Age, YearsEngland (n = 55 783), %LowMiddleHighLowMiddleHigh12190.60.30.50.00.00.0203435491.16.3**2.2*2.20.51.71.12.5**1.21.21.10.5506415.0**7.35.58.8**3.32.1658014.3**15.0**7.413.5**12.9**5.741115.3**10.910.29.36.35.6121717.3**18.0**10.510.86.09.0183435.5**31.6**21.419.2**13.0*8.43549506445.2**47.4**35.041.5*30.831.926.6**28.322.1**25.515.823.7658038.4*40.6**29.424.827.2*20.6121913.6**11.9*7.45.06.99.7203416.2**11.4*6.69.9**5.73.1354914.6**11.5*6.68.1**3.32.3506413.0**7.05.16.3**2.01.410.9**6.75.45.3*3.42.45.1**DiabetesObesityLow HDL6580High cholesterol ratio12192.43.63.34.8203412.0*6.0**12.3*7.88.7**4.92.3354921.8**14.411.412.2**7.35.3506421.8**18.6**12.117.0**11.811.0658019.3*15.413.415.911.914.41834354941.351.7**39.944.138.139.734.732.0**29.227.7*29.422.3506455.8**50.6*39.744.5**33.629.9658054.9*49.645.744.9*45.6*37.512190.40.30.51.21.10.520343.6*2.61.34.13.33.6354923.8*17.017.814.714.711.75064658053.0**78.9*50.0**76.138.670.148.0**70.5**33.865.530.061.2Birth311.7**6.44.92.91.61.041113.7*9.78.68.1*8.2*3.3121917.416.419.57.24.85.2203417.413.415.87.55.96.2354915.815.614.39.7**6.05.05064658015.6**15.1*14.2*12.39.710.79.48.77.29.86.65.8High C-reactive proteinHypertensionAsthma ever diagnosedContinuedNovember 2012, Vol 102, No. 11 | American Journal of Public Healthto classify individuals as low risk (< 1 mg/L),medium risk (1—3 mg/L), or high risk (> 3 mg/L)and to create a binary measure of high vs low ormedium health risk.27—29The self-reported health conditions werebased on participants responses to standardsurvey questions. These were chosen becauseof comparability between the 2 data sets andwere used in previous research comparinghealth in the United States and England.3Responses indicated whether the individualwas ever told by a doctor that he or she hadhad a heart attack or angina, a stroke, or asthma(in England, the HSE simply asks whether theindividual has asthma). Except asthma, all ofthese measures were available for individualsat least 20 years of age. Asthma was availablefor all ages.Age Groups and Income MeasureI categorized age into broad groups thatcorrespond to the Centers for Disease Controland Preventions Stages of Life. The categorieswere as follows: infants (birth—3 years), children (4—11 years), adolescents (12—19 years),young adults (20—34 years), middle-age adults(35—49 and 50—64 years), and older-ageadults (65—80 years).The primary independent variable of interest in this study was income-based socioeconomic status, which I constructed from thefamily income variable available in boththe HSE and NHANES at present value, adjusted by the Organisation for Economic CoOperation and Developments (OECD) squareroot equivalence scale, and then divided intoequal terciles by using the sample weights.30The square root equivalence scale has beenused in OECD publications on internationalincome inequality and poverty since 2008.Additionally, because of the pooling of multipleyears of data in the NHANES and HSE, Iadjusted the measure for cost of living to theyear 2006 using the Consumer Price Indexesfor the United States and the United Kingdom.31,32 Use of terciles rather than absoluteincome adjusted for differences in averagelevels and the income distribution acrossthe 2 countries; previous studies of olderadults have used this method.1,2,33 I alsoadjusted the terciles by age group becauseof the uctuations in income throughoutthe life span.Martinson | Peer Reviewed | Research and Practice | 2051RESEARCH AND PRACTICETABLE 2ContinuedHeart attack20340.70.50.30.30.00.135492.31.41.21.0*1.5*0.250648.9**4.42.03.1*1.7658018.4**10.47.611.2*15.4**7.10.1Stroke20345.4**0.70.50.20.30.035492.5*1.60.60.50.40.350645.1**3.21.33.9**0.90.8658010.0**8.3*4.16.67.25.3Note. HDL = high-density lipoprotein cholesterol.*P