Unequal hopes and lives in the USA: optimism, race, place, and premature mortality


The 2016 election highlighted deep divisions in the USA, and exposed unhappiness and frustration among poor and uneducated whites. The starkest marker of this unhappiness is the rise in preventable deaths and suicides among the middle aged of this cohort. In contrast, minorities have much higher levels of optimism, and their life expectancies continue to rise. Low-income respondents display the largest differences, with poor blacks by far the most optimistic, and poor rural whites the least. African Americans and Hispanics also have higher life satisfaction and lower stress incidence than poor whites. The gaps across racial groups peak in middle age, at the nadir of the U-curve of age and life satisfaction. We explored the association between our subjective well-being data and the Centers for Disease Control and Prevention (CDC) mortality data. We find that the absence of hope among less than college-educated whites matches the trends in premature mortality among 35–64-year-olds. Reported pain, reliance on disability insurance, low labor force participation, and differential levels of resilience across races all have mediating effects in the desperation-mortality associations. We also explore the role of place, and map the states associated with higher/lower indicators of well-being for these different cohorts. The matches between indicators of well-being and of mortality suggest that the former could serve as warning indicators of ill-being in the future, rather than waiting for rising mortality to sound the alarms.

This is a preview of subscription content, access via your institution.

Fig. 1
Fig. 2
Fig. 3
Fig. 4


  1. 1.

    For example, the high material costs of being poor in Latin America in the 1970s, which included paying as much as 18 times more per unit of water and electricity, with inferior health outcomes (Adrianzen and Graham 1973).

  2. 2.

    Gelman and Auerbach (2016) posit that these trends are driven in part by aggregation bias at the older ages of the 45–54 cohort.

  3. 3.

    Obtained through NBER at http://www.nber.org/data/seer_u.s._county_population_data.html.

  4. 4.

    Source: http://factfinder.census.gov/faces/tableservices/jsf/pages/productview.xhtml?pid=PEP_2015_PEPANNRES

  5. 5.

    This proportion has increased very slightly over time: from 85 to 85.5% from 2010 to 2014.

  6. 6.

    In additional specifications (Section 3 and Appendix 2), we use national-level weights (doubling the number of MSAs), with no meaningful changes.

  7. 7.

    GH halved the number of daily interviews to 500 in 2013, decreasing the number of weighted MSAs.

  8. 8.

    In Appendix 2, we employ alternative definitions for poor, middle-income, and rich individuals, including that from the Census Bureau. The results we obtain are quantitatively similar to those in our main specification.

  9. 9.

    This percentage and the previous one do not use Gallup’s survey weights. The corresponding shares of weighted respondents are 15 and 24%.

  10. 10.

    The age dummies have ranges that contain a similar number of observations and generally match the age brackets present in the other databases that we used, such as the CDC’s Compressed Mortality File.

  11. 11.

    The inclusion of (MSA)*(Year) interactions, month of interview dummies, other types of survey weights, and no survey weights at all, does not meaningfully change in the coefficients. See Appendix 2.

  12. 12.

    While the logit and ordered logit models are technically the appropriate specification, it has become accepted practice to use OLS in happiness regressions, for ease of interpretation, as long as the results are very close. In our case, the OLS specifications yield slightly more significant coefficients for some variables of interest, but the patterns are otherwise identical and the choice of model does not affect our conclusions. We report the logit and ordered logit results in Appendix 3.

  13. 13.

    However, we include both year and MSA dummies in every specification.

  14. 14.

    We also explored reported depression and found that it was also highest among low-income whites. A more comprehensive account, however, would require a separate study. Depression and happiness are distinct emotional states. While positive emotional states, such as happiness and smiling, tend to track closely, negative states—stress, anger, and depression—track differently, with depression the most distinct. See Stone and Mackie (2013).

  15. 15.

    We computed Figure 1 using the coefficients from column (2), the BPLA regression without BPL as a control. When we use current BPL as a control (column (3)), we get slightly lower gaps between poor blacks and poor whites and they decrease less as income increases.

  16. 16.

    These race-income heterogeneities are also very strong when generating separate regressions by year (Table 7, Appendix 1).

  17. 17.

    The social support and anger questions were asked in 2008–2012 and 2010–2013, respectively. Therefore, the time period under consideration differs for those two cases.

  18. 18.

    We also collapsed the data at the MSA level, using MSA fixed effects to control for non time-varying MSA-specific unobservables. This reduces the significance of some variables, but the main ones hold.

  19. 19.

    While we find a sharp drop in the life satisfaction and optimism of Democrats and Independents in weeks following the 2016 election, a partial recovery seems to be underway by the end of the year (https://www.brookings.edu/blog/up-front/2017/02/02/the-trump-unhappiness-effect-nears-the-great-recession-for-many/). We do not yet have the data to test if there is a longer-term negative effect of trends since then—and plan to do so going forward. The evidence above, though, suggests that this is not a finding that is explained by short-term events.

  20. 20.

    Graham and Pettinato (2002) coined the term “happy peasants and frustrated achievers” to describe such optimistic poor individuals in many poor countries over a decade ago.

  21. 21.

    These use national-level, rather than MSA-level, survey weights.

  22. 22.

    National Center for Health Statistics. Compressed Mortality File, 2008–2015 (CD-ROM Series 20, No. 2 U) Vital Statistics Cooperative Program. Hyattsville, Maryland. 2016.

  23. 23.

    The three decades considered (35–44, 45–54, 55–64) all had similar “composite” mortality rates.

  24. 24.

    The calculation is: exp(−0.087 ∗  log (1.50)). The log represents a 50% increase and (log (1.50) = 0.40547); the product equals approximately − 0.04. The mean optimism or expected future life satisfaction is 7.86, so the change above corresponds to approximately 0.5% of this mean value.

  25. 25.

    Regression results available from the authors.

  26. 26.

    This is not the case in Table 6, but is indeed the case when using a logit estimation framework (see Appendix 3).

  27. 27.

    Social Security Advisory Board: http://www.ssab.gov/Disability-Chart-Book. These are not age-adjusted numbers.

  28. 28.

    We thank Henry Aaron for raising this.

  29. 29.

    For the distribution of broadband, see: https://www.broadbandmap.gov/technology.

  30. 30.

    In this case, minorities comprise only African Americans and Hispanics.

  31. 31.

    We omitted income variables as otherwise state dummies would disproportionately pick up the disadvantageous state-level aspects, such as higher costs of living (Oswald and Wu 2011).

  32. 32.

    We excluded states with less than 50 observations/year for the group in question.

  33. 33.

    The U.S. has the world’s highest per capita consumption of opioids: http://www.painpolicy.wisc.edu/country/profile/united-states-america.

  34. 34.

    40% of Medicare recipients are unaware of being on a government program (Kuziemko et al. 2015).

  35. 35.

    See, e.g., https://www.whatworkswellbeing.org/ .

  36. 36.

    When regressing the household size variable on income group (recall that Gallup’s income variable assigns respondents to income brackets, coded from 0 to 10), a coefficient of 0.080 is obtained. This would mean that, on average and imposing a linear progression, an increase of 1 in the income group is associated with an increase of 0.08 in the household size.

  37. 37.

    Steven Ruggles, Katie Genadek, Ronald Goeken, Josiah Grover, and Matthew Sobek. Integrated Public Use Microdata Series: Version 6.0 [dataset]. Minneapolis, MN: University of Minnesota, 2015.

  38. 38.

    Respondents whose reported household size is larger than 10 are dropped from the analysis (951 observations).

  39. 39.

    More precisely, 19 and 20% of the (unweighted) observations corresponded to the poor and to the rich groups, respectively. Upon application of the sampling weights, these percentages changed to 27 and 14%, respectively.

  40. 40.

    See, for example: https://www.census.gov/data/tables/time-series/demo/income-poverty/historical-poverty-thresholds.html.

  41. 41.

    Under this specification, rich respondents are defined in the same way as in third alternative of Section a). The results do not meaningfully change if the rich are classified under the criterion used for the base specification (i.e., the rich group corresponds to the respondents whose reported household income is above $120,000/year; these results are not displayed but are available from the authors, upon request).


  1. Adrianzen B, Graham GG (1973) The high costs of being poor. Arch Environ Health 28(6):312–315

    Google Scholar 

  2. Andrews R, Casey M, Hardy B, Logan T (2017) Location matters: historical racial segregation and inter-generational mobility outcomes. Econ Lett 158:67–72. https://doi.org/10.1016/j.econlet.2017.06.018

    Article  Google Scholar 

  3. Angelini V, Casi L, Corazzini L (2013) Life satisfaction of immigrants: does cultural assimilation matter? J Popul Econ 28:817–844

    Article  Google Scholar 

  4. Assari S, Lankarani M (2016) Depressive symptoms are associated with more hopelessness among white than black older adults. Front Public Health 4(82):1–10

    Google Scholar 

  5. Black S, Furman J, Rackstraw E, Rao N (2016) The long-term decline in US prime-age male labour force participation. VoxEU.org, July 6

  6. Blanchflower D, Oswald A (2004) Well-being over time in Britain and the USA. J Public Econ 88(7-8):1359–1386. https://doi.org/10.1016/S0047-2727(02)00168-8

    Article  Google Scholar 

  7. Blanchflower D, Oswald A (2008) Is well-being U-shaped over the life cycle? Soc Sci Med 66(8):1733–1749. https://doi.org/10.1016/j.socscimed.2008.01.030

    Article  Google Scholar 

  8. Blanchflower D Oswald A (2018) Unhappiness and pain in modern America: a review essay and further evidence on Carol Graham’s happiness for all? J Econ Lit, forthcoming

  9. Case A, Deaton A (2015a) Rising morbidity and mortality in midlife among white non-Hispanic Americans in the 21st century. Proc Natl Acad Sci 112(49):15078–15083. https://doi.org/10.1073/pnas.1518393112

    Article  Google Scholar 

  10. Case A Deaton A (2015b) Suicide, age, and wellbeing: an empirical investigation. Center for Health and Wellbeing, Princeton University. http://www.nber.org/papers/w21279

  11. Case A Deaton A (2017) Mortality and morbidity in the 21st century, Brookings Papers on Economic Activity, Spring 2017

  12. Cherlin A (2016) Why are white death rates rising? New York Times. February 22

  13. Chetty R, Hendren N, Kline P, Saez E (2014) Where is the land of opportunity? The geography of intergenerational mobility in the United States. Q J Econ 129(4):1553–1623. https://doi.org/10.1093/qje/qju022

    Article  Google Scholar 

  14. Chetty R, Stepner M, Abraham S, Lin S, Scuderi B, Turner N, Bergeron A, Cutler D (2016) The association between income and life expectancy in the United States, 2001–2014. J Am Med Assoc 315(16):1750–1766. https://doi.org/10.1001/jama.2016.4226

    Article  Google Scholar 

  15. Clark A (2006) A note on unhappiness and unemployment duration. IZA Discussion Papers, No. 2406, October

  16. Clark A, Oswald A (1994) Unhappiness and unemployment. Econ J 104(424):648–659

    Article  Google Scholar 

  17. De Neve JE (2013) In: Helliwell J, Layard R, Sachs J (2013) World Happiness Report, 2013. New York: Earth Institute Press

  18. Demyank Y, Hyrshko D, Luego-Prado, MJ, Sorensen B (2017). “The Rise and Fall of Consumption in the 00’s: A Tangled Tale”, Working Paper, Federal Reserve Bank of Cleveland, (December). www.cepr.org/active/publications/discussion_papers/dp.php?dpno=12522

  19. Dursun B, Cesur R (2016) Transforming lives: the impact of compulsory schooling on hope and happiness. J Popul Econ 29(3):911–956

    Article  Google Scholar 

  20. Dwyer-Lindgren L, Bertozzi-Villa A, Stubbs RW, Morozoff C, Kutz MJ, Huynh C, Barber RM, Shackelford KA, Mackenbach JP, van Lenthe FJ, Flaxman AD, Naghavi M, Mokdad AH, Murray CJL (2016) U.S. county-level trends in mortality rates for major causes of death, 1980-2014. J Am Med Assoc 316(22):2385–2401. https://doi.org/10.1001/jama.2016.13645

    Article  Google Scholar 

  21. Eberstadt N (2016) Men without work: America’s invisible crisis. Templeton, W.C. Pennsylvania

    Google Scholar 

  22. Gelman A, Auerbach J (2016) Age aggregation bias in mortality trends. Proc Natl Acad Sci 113(7):E816–E817. https://doi.org/10.1073/pnas.1523465113

    Article  Google Scholar 

  23. Graham C (2008) Happiness and health: lessons—and questions—for policy. Health Aff 27(2):72–87. https://doi.org/10.1377/hlthaff.27.1.72

    Article  Google Scholar 

  24. Graham C (2017) Happiness for all? Unequal hopes and lives in pursuit of the American dream. Princeton University Press, Princeton 2017

    Google Scholar 

  25. Graham C, Pettinato S (2002) Frustrated achievers: winners, losers, and subjective well-being in new market economies. J Dev Stud 38(4):100–140. https://doi.org/10.1080/00220380412331322431

    Article  Google Scholar 

  26. Graham C, Ruiz-Pozuelo J (2017) Happiness, stress, and age: how the U-curve varies across people and places. J Popul Econ 30(1):225–264. https://doi.org/10.1007/s00148-016-0611-2

    Article  Google Scholar 

  27. Graham C, Eggers A, Sukhtankar S (2004) Does happiness pay? An initial exploration based on panel data from Russia. J Econ Behav Organ 55:319–342

    Google Scholar 

  28. Herbst C (2013) Welfare reform and the subjective well-being of single mothers. J Popul Econ 26(1):203–238. https://doi.org/10.1007/s00148-012-0406-z

    Article  Google Scholar 

  29. Isenberg N (2016) White trash: the 400-year untold history of class in America. Viking, New York

    Google Scholar 

  30. Jackson J (2015) The role of well-being measures in minority aging research. Presentation to National Institutes of Aging Conference on Well-Being and Aging, Orlando, November 18

  31. K Kawashima-Ginsberg, Sullivan F (2017). Sixty percent of rural Millennials lack access to a political life. The Conversation, March 27

  32. Keyes C, Simoes E (2012) To flourish or not: positive mental health and all-cause mortality. Am J Public Health 102(11):2164–2172. https://doi.org/10.2105/AJPH.2012.300918

    Article  Google Scholar 

  33. Krekel C, Tiefenbach T, Ziebarth NR (2015) How natural disasters can affect environmental concerns, risk aversion, and even politics: evidence from Fukushima and three European countries. J Popul Econ 28:1137–1180

    Article  Google Scholar 

  34. Krueger A (2017) Where have all the workers gone: an inquiry into the decline of the U.S. labor force participation rate. Brookings Papers on Economic Activity Fall 2017, forthcoming

  35. Krugman P (2015) Despair, American style. New York Times, November 9, A19

  36. Kuziemko I, Norton M, Saez E, Stantcheva S (2015) How elastic are preferences for redistribution? Evidence from randomized survey experiments. Am Econ Rev 105(4):1478–1508. https://doi.org/10.1257/aer.20130360

    Article  Google Scholar 

  37. Oswald A, Wu S (2011) Well-being across America. Rev Econ Stat 93(4):1118–1134. https://doi.org/10.1162/REST_a_00133

    Article  Google Scholar 

  38. Pierce JR, Schott PK (2016) Trade liberalization and mortality: evidence from U.S. counties. Finance and Economics Discussion Series, Federal Reserve Board, Washington, D.C.

    Google Scholar 

  39. Porter E (2015) Education gap widens between rich and poor. New York Times, September 23, B1

  40. Reardon S, Portilla X (2015) Recent trends in socioeconomic and racial school readiness gaps at kindergarten entry. Center for Education Policy Analysis Working Papers, No. 15–02, Stanford University

  41. Ruffing K (2017) Decline in labor-force participation not due to disability programs. Center on Budget and Policy Priorities Blogs, August 25

  42. Ryff C (2015) Varieties of well-being and their links to health. Presentation to National Institutes of Aging Conference on Well-Being and Aging, Orlando, November 18

  43. Schwandt H (2016) Unmet aspirations and an explanation for the age-U shape in well-being. J Econ Behav Organ 122:75–87. https://doi.org/10.1016/j.jebo.2015.11.011

    Article  Google Scholar 

  44. Shiels M et al (2017) Trends in premature mortality in the USA by sex, race, and ethnicity from 1999 to 2014: an analysis of death certificate data. Lancet 389(10073):1043–1054. https://doi.org/10.1016/S0140-6736(17)30187-3

    Article  Google Scholar 

  45. Steptoe A, Deaton A, Stone A (2015) Subjective well-being, health, and ageing. Lancet 385(9968):640–648. https://doi.org/10.1016/S0140-6736(13)61489-0

    Article  Google Scholar 

  46. Stone A, Mackie C (2013) Subjective well-being: measuring happiness, suffering, and other dimensions of human experience. National Research Council of the National Academies, Washington, DC http://www.nap.edu/catalog.php?record_id=18548

    Google Scholar 

  47. Tavernise S (2016) Black Americans see gains in life expectancy. The New York Times, May 8

  48. Trisi D (2016a) Safety net cut poverty nearly in half last year. Center on Budget and Policy Priorities Blogs, September 14

  49. Trisi D (2016b) Three essays on poverty and social welfare policy. PhD Dissertation, University of Maryland, College Park

Download references


The authors are, respectively, Leo Pasvolsky Senior Fellow at the Brookings Institution and College Park Professor, University of Maryland, and PhD student, University of Maryland. We thank Andrew Oswald and Eddie Lawlor, as well as Alice Rivlin, Alan Blinder, Belle Sawhill, Bill Galston, Mike O’Hanlon, Bradley Hardy and other participants at a Brookings “restoring the middle class” seminar, for very helpful comments. They also appreciate the suggestions of an anonymous reviewer. Graham acknowledges the generous support from a Robert Wood Johnson Foundation pioneer award, and Pinto from a flagship fellowship at UMD.


This study was funded by grant # 74378 from the Robert Wood Johnson Foundation.

Author information



Corresponding author

Correspondence to Carol Graham.

Ethics declarations

Conflict of interest

The authors declare that they have no conflict of interest.

Additional information

Responsible editor: Erdal Tekin


Appendix 1 Race-income heterogeneities by year

Table 7 Base specification, but with yearly regressions (2010–2015)

Appendix 2 Robustness checks

  1. a)

    Household size adjustments

One potential concern regarding the results in Table 2 is that we assign respondents to income groups based on total household income. In the GH data, household size correlates positively to household income,Footnote 36 which would introduce a bias in the estimated coefficients for the income and interaction terms. In our baseline specification, we did not adjust for household size, for two main reasons. One is the high share of missing values for this variable (25% of the observations in our baseline specifications). A second reason is that, as mentioned in Section 3, the income variable in GH is not continuous and instead assigns respondents to 1 of 11 income brackets. Adjusting the reported household income to the household size would therefore require assigning respondents an income value, based on the bracket they report. This problem is further compounded by the fact that, with a categorical income variable, incomes are inevitably top-coded, which demands further assumptions regarding how to assign income to the households in the top bracket.

We address this concern with three different strategies. In the first case, we consider only the cases of one-person households, where no adjustment is necessary. In the second alternative, we exclude those in the top income bracket (i.e., respondents reporting pre-tax household income above $120,000/year), assign every other respondent the midpoint of the income bracket they reported, and adjust reported income by household size, on a per capita basis. In the final alternative, we do not exclude any respondent. For those not in the top income bracket, we applied the adjustment described in the previous alternative. We assigned those in the top income bracket a value based on data from the American Community Survey, obtained through IPUMS (Ruggles et al. 2015).Footnote 37

Table 8 displays the results when following the first alternative. The magnitude of the indicator variables for income groups among poor white respondents increases slightly (see rows for “Poor household”). Nevertheless, the racial heterogeneities remained very stark. For instance, among the poor and holding everything else constant, African Americans score nearly 1.1 points higher on the 0–10 optimism scale than whites and are 13 percentage points less likely to have experienced stress the previous day (see rows for “Black” and “(Poor household)*(Black)”).

Table 8 Base specification, only one-person households (2010–2015)

As mentioned above, the second and third alternatives adjust the reported pretax income for household size. This required additional assumptions. We assigned those in income brackets below the top to the midpoint. We assigned those in the top income bracket the average of households whose total pretax income exceeds $120,000/year, based on estimates using data from Ruggles et al. (2015). For every year in the 2008–2015 period, we identified households reporting pretax income above the $120,000/year threshold, computed the corresponding average income, and assigned this yearly amount to the respondents in the top bracket. We then converted all incomes into per capita amounts, by dividing the total household income by the household size.Footnote 38 Finally, we reassigned the three income categories to reflect this new per capita income variable. We specified thresholds such that we would again obtain approximately 20% of observation in the rich group, another 20% in the poor, and the remaining in the middle-income group.Footnote 39 This resulted in a maximum threshold of $12,499 per person for the poor group and in a minimum of $54,000 per person for the rich group.

The second alternative differs from the third only in the choice to exclude those assigned to the top income bracket in the GH data, which substantially reduces the number of respondents in the rich group. Table 9 displays the estimates obtained when using this alternative. As before, the race-income heterogeneities remain quantitatively large despite a slight decrease in the optimism gap: poor African Americans are now 0.83 points higher than poor whites in the optimism scale.

Table 9 Base specification, income adjusted by household size (excludes respondents in the top income bracket)

Table 10 displays the estimates obtained when using the third alternative, which includes the respondents in the top income bracket. As in the previous cases, this approach generates large race-income heterogeneities between African Americans and whites.

Table 10 Base specification, income adjusted by household size (respondents in the top income bracket are included)
  1. b)

    Alternative measure of poverty following US Census Bureau

Two possible objections to the robustness checks conducted in the previous subsection are that the thresholds chosen are relatively arbitrarily and that the definition of poverty used implicitly ignores any type of equivalence scale. Regarding the latter aspect, it means that the income needed for a household to be above the poverty threshold is always linearly proportional to the household size, ignoring any aspect related to its composition or the age of its members. An alternative to address both issues, then, is to use the poverty thresholds that the US Census defines every yearFootnote 40 and correspondingly classify respondents as poor.Footnote 41 Table 11 displays the results for this specification. As before, there are no meaningful differences in the race-income heterogeneities.

Table 11 Base specification, income adjusted by household size and poverty thresholds as defined by the Census Bureau
  1. c)

    Exclude MSAs with smaller numbers of poor African American respondents

Another concern about the base specification results is that the results could be driven by the within-MSA variation in MSAs with very few African Americans, particularly poor ones. Table 12 displays the results obtained when running the base specification under different thresholds for the minimum number of low-income African Americans per MSA, per year. As panel A to panel C illustrate, there are again no meaningful differences in magnitude and significance levels across the different thresholds.

Table 12 Base specification, using thresholds for minimum number of Poor African Americans by MSA
  1. d)

    Include month and (MSA) × (year) dummies

A possible objection to the specification laid out in Eq. (1) is that, by using year and MSA dummies separately (i.e., without adding their interaction), we are imposing a parallel time trend on all MSAs. If the MSAs happened to follow heterogeneous time trends during the period under analysis, the absence of interaction dummies could bias our estimates. Similarly, the time of the year of the interview might be correlated with our variables of interest. Table 13 below displays the results when we include both month and (MSA) × (year) dummies. The coefficient estimates and significance are nearly unchanged, suggesting that neither factor was in fact introducing a meaningful bias into our estimates.

Table 13 Base specification, plus month and (MSA) × (year) dummies
  1. e)

    Robustness to use and type of survey weights

The base specification estimates in Table 2 use MSA-level weights. A possible concern is that the results may be sensitive to the type of survey weights, or simply to their use.

Table 14 below displays the results when the national-level survey weights are used. I this case, we are no longer restricted to the 196 MSAs for which we have MSA-level survey weights at some point during the 2010–2015 period, and as a result, our sample increases and encompasses nearly all the existing MSAs. The coefficient estimates for our variables of interest, however, suffer no relevant change.

Table 14 Base specification, with national-level survey weights

Table 15 also uses this enlarged sample of respondents located in any MSA, but instead uses no weights. The differences to Table 2 are small and the coefficient estimates are often of a higher magnitude.

Table 15 Base specification, no survey weights

Appendix 3 Ordered logit and logit estimation

As mentioned in the main text, we re-estimate the main tables of the article under ordered logit and logit specifications. Tables 16 to 21 below show that our findings are robust to the choice of estimation framework.

Table 16 Re-estimation of Table 1, using ordered logit and logit models
Table 17 Re-estimation of Table 2, using ordered logit and logit models
Table 18 Re-estimation of Table 3, using ordered logit and logit models
Table 19 Re-estimation of Table 4, using ordered logit and logit models
Table 20 Re-estimation of Table 5, using ordered logit and logit models
Table 21 Re-estimation of Table 6, using ordered logit and logit models

Appendix 4 The geography of stress and pain, by race group

Figure 5 below displays the maps for stress and pain, which are not in the main text.

Fig. 5

The geography of stress and pain, by race groups

Figure 6 shows the corresponding boxplots of state coefficients for each of the five mapped variables. Although the geographical patterns are different for whites and minorities, the dispersion in absolute terms is similar for both groups, except for life satisfaction, where location seems to matter substantially more for minorities.

Fig. 6

The distribution of state coefficients for life satisfaction, optimism, worry, stress, and pain, by race group

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Graham, C., Pinto, S. Unequal hopes and lives in the USA: optimism, race, place, and premature mortality. J Popul Econ 32, 665–733 (2019). https://doi.org/10.1007/s00148-018-0687-y

Download citation


  • Well-being
  • Optimism
  • Stress
  • Premature mortality
  • Resilience

JEL classifications

  • D6
  • I3
  • I14
  • I19