The Effect of Presidential Service on Life Expectancy Mark Borgschulte∗ UC Berkeley January 2014

Abstract Presidents of the United States of America appear to suffer a significant loss of natural life from their service. Using runner-up presidential candidates as the counterfactual, presidents lose an average of 3.8 years of natural life expectancy; including assassinations increases the lost life to 6.7 years. I argue the rigors of the office are the most likely explanation for the difference in natural lifespan. This finding complicates our understanding of the health effects of labor supply and rank-based explanations of the socioeconomic gradient in life expectancy. JEL No.: I14; J14

∗

PhD Candidate, Department of Economics, University of California, Berkeley. I would like to thank David Card, Angus Deaton, Ronald Lee, and Ken Wachter for their comments, and Natalie Goldberg and Allen Gurdus for their excellent research assistance. Email: mark at econ.berkeley.edu.

1

Introduction

The health effects of rank within a social and bureaucratic hierarchy have been proposed as an explanation for the socioeconomic gradient in health and life expectancy, most famously in the Whitehall study.1 Despite the extensive subsequent literature on social inequalities and health,2 causal evidence on the relationship between rank and lifespan remains scarce, especially as it is difficult to disentangle the effects of working conditions from selection into rank and occupation.3 A fundamental feature of the socioeconomic gradient in health and life expectancy is that it extends throughout the distribution, a fact which motivates much of the interest in rank-based explanations, and away from income and poverty as the causal forces behind health inequalities. Studies on primates find negative effects of low rank; however, they also suggest the effects of high rank may be inversely related to stability in the social hierarchy.4 To date, we have no systematic evidence on life expectancy at the top of human hierarchies, despite the importance of these issues for our understanding of inequality, bequests and estate taxation, and the labor supply of the most productive workers in the economy.5 In this paper, I investigate the reported rapid aging of US presidents, an apparent counterexample to rank as the cause of superior life expectancy. If promotion to high rank leads to longer life (as the rank-lifespan theory predicts), then presidents should exhibit superior life expectancy to both the general population and other political officeholders, since they are the highest ranking member of a number of hierarchies in the US government. On the other hand, conventional wisdom based on observation holds that presidents age more rapidly than other men of their age; in particular, they exhibit gray hairs, wrinkles and other age-related changes to facial features.6 The office of US president is uniquely well-suited for such an investigation, not only due to its prominence 1

From Cutler et al. (2006): “Outside of economics, the currently dominant theory of health differentials is that the poor health of low status individuals is caused by ‘psychosocial stress’—the wear and tear that comes from subordinate status and from having little control over ones own life.” See Marmot et al. (1991), Marmot (1994) and Marmot (2004) on Whitehall. 2 For example, see McEwen (1998) on allostatic loads, the proposed mechanism behind rank-based explanations. 3 Recent work in economics on the causal mechanisms of Whitehall include Anderson and Marmot (2011) and Case and Paxson (2011); also, see Stringhini et al. (2010). 4 As in Sapolsky (2005). 5 For example, Kopczuk and Saez (2004) use the life expectancy of college-educated whites (estimated in Brown et al. (2002)), in the absence of detailed data on life expectancy at the top of the income distribution. 6 A Google search of “presidential aging in office” uncovers articles and photo series documenting this visual effect published by Time magazine, the Washington Post and CNN, among others. See Olshansky (2011) and Hanna (2011) for discussions of this evidence.

1

in the US political hierarchy, but also the long history of contested democratic elections and stable 4-year term of service. These features allow me to establish counterfactual life expectancy through the comparison of US presidents to electoral runners-up.7 Against the predictions of the theory, but consistent with conventional wisdom, I find a shorter natural life expectancy of US presidents relative to runners-up in presidential elections. Runners-up outlive winning candidates by an average of 3.8 years, treating assassinated and still-living presidents as censored observations. Expanding the sample to include deaths by assassination increases the estimated effect to 6.7 years, meaning the loss of natural life expectancy has historically been larger than the loss to the risk of assassination. Winning candidates have survived an average of 16 years, so these estimates corresponds to a 20% reduction in natural life expectancy. The statistical significance of the results depends on the full sample, but is otherwise robust to a variety of controls and alternative specifications. In sum, I find the historical evidence is consistent with the observed rapid aging of presidents. Presidential runners-up may outlive presidents due to either the effects of presidential service or the (s)election of shorter-lived candidates in the final round of the electoral process. Previous studies on the socioeconomic gradient struggled to disentangle the causal effect of rank from the unobservable factors that predict it, as most promotion processes select for qualities associated with longer life. Crucially for my comparison, the selection effect works against the finding: if electors prefer longer-lived candidates, the comparison of winners to losers will understate the true costs of service. Historical accounts and previous research suggest healthier candidates possess an advantage in the election—whether through the campaign process or a direct electoral preference for health8 —implying the estimates are biased towards finding a life expectancy advantage for presidents. To be precise regarding the size of this bias, the estimates here should be adjusted by the (inherently unobservable) pre-election advantage in life expectancy of winning candidates. Overturning the monotonicity finding of previous research on the socioeconomic gradient complicates rank-based explanations of the socioeconomic gradient in life expectancy, but may also provide support for the hypothesized causal mechanism. As a causal pathway from rank to health, the Whitehall literature has focused on increased job control and predictability that comes with higher rank, diminishing work-related stress, and hence, 7

I define the runner-up to be the highest popular vote finisher who did not become president, or the in absence of a popular vote, the highest electoral vote-getter. Usually, but not always this candidate came in “second-place.” 8 See Acemoglu (2005) and Besley and Persson (2009).

2

a decreasing the risk of cardiovascular and other diseases.9 Although the medical literature has not conclusively shown stress to reduce lifespan, research has demonstrated the short-term damage caused by allostatic loads and repeated exposure to stress-related hormones.10 The job of US president may be particularly exposed to decreased job control and low predictability, especially during times of national emergencies. Motivated by the observed signs of aging among US presidents, previous research has investigated the hypothesis of a link between presidential service and life expectancy attributed to exposure to stress.11 In Section 4, I draw on evidence from secondary analyses and historical accounts to argue that work-related stress is the most likely explanation for these findings. Although work-related stress is an important hypothesis for the shorter life expectancy of US presidents, exposure to other risk factors may confound the relationship between service and aging-related deterioration of health and lifespan. While the detrimental health effects of presidential service correspond to effects observed at the top of primate hierarchies, I find little support for a corresponding relationship between the stability of the president’s position and the magnitude of the effect. I find no relationship between the margin of victory and life expectancy of the candidates, and no smaller effect for reelection (which should represent more stable political environments). Another possibility is that presidents travel and interact with many people, and may come into contact with infectious diseases at higher rates than had they lost the election. As well, the rigors of the office may lower immunity, and hence, make presidents more susceptible to infectious disease.12 I find little evidence of an effect of moving to Washington, D.C., as vice-presidents show no loss of life expectancy relative to runner-up vice-presidential candidates, but I cannot rule out all work-related exposures. An analysis of cause of death would require a larger sample size, and throughout the paper I measure the combined effect of service on longevity. The remainder of the paper is organized as follows. Section 2 discusses the selection of presidential candidates, and previous research on life expectancy at the top of hierarchies. Section 3 reports the main results. Section 4 discusses prospective mechanisms and robustness. Section 5 concludes. 9

See Vaananen et al. (2003), Anderson and Marmot (2011), Moslehi et al. (2012) for evidence on the links between job control, stress and life expectancy. 10 For example, see McEwen (1998), Epel et al. (2004). 11 I discuss Olshansky (2011) and selection into candidacy in Section 2. 12 Indeed, two of the four presidents to die of natural causes while in office fell victim to infectious disease, William Henry Harrison (pneumonia) and James K. Polk (gastroenteritis).

3

2

Presidents and Life Expectancy at the Top

2.1

Previous Work on Superstar Life Expectancy

Using an alternative counterfactual, Olshansky (2011) estimates a near zero effect of presidential service on aging by comparing presidents to other men of their birth cohort, conditional on survival to the age of election. In essence, this strategy assumes presidents can be compared to the average man of their birth cohort, conditional on survival to the age of election. This assumption would hold if presidents were chosen at random from the population, or more generally, the process of selection into the presidency were unrelated to factors that predict longer life. The comparison of presidents to the average man of their age and cohort ignores the socioeconomic gradient and attendant advantages possessed by presidents; to take a simple example from the US Census, literacy rates did not reach 90% for men age 50-60 until 1930. There are other obvious presidential characteristics that distinguish these men from the general population in ways that have been associated with differential life expectancy: marriage rates are around 75% for men age 50-60 before 1950, while 84% of presidential candidates have been married; the foreign-born comprised 30% of the US population in 1900, while presidents are natural born citizens by law; and around 10% of men aged 50-60 have identified as black since Emancipation, while no candidate before Barack Obama acknowledged African ancestry. Presidential candidates have been noted for their relative height, another predictor of high income and long life through much of US history.13 Put simply, the class of men from which the president is drawn has much higher life expectancy than the average man of their age in the population. The potential for negative effects of fame, superstardom and tournaments have also been studied, including the potential for effects on lifespan. Superstars, according to Rosen (1981), arise from systems with highly skewed distributions of income, market share and public attention. The tournament structure discussed by Rosen is reflected by the multiple rounds of election over a politician’s lifetime. Historically, winning presidential candidates are dramatically distinguished from losing candidates and other “near presidents,” receiving far greater prestige and celebrity after the election, as well as the returns in social and political capital to serving as chief executive. Malmendier and Tate (2009) study superstar CEOs and find negative effects of celebrity on company performance. Similar in spirit to this paper, Rablen and Oswald (2008) and Becker et al. (2008) study the effects of elections and prestige on longevity. Rablen and Oswald (2008) compares Nobel prize winners to a matched group of similar nominees, finding small (1-1.5 13

See Persico et al. (2004), Fogel (2004).

4

year) gains in life expectancy from winning a Nobel.14 Becker et al. (2008) study the life expectancy effect of election to the Baseball Hall of Fame by comparing those just elected to those just under the threshold. Becker et al. finds those who are elected live longer than both those who narrowly miss election and those not considered, but those who narrowly miss election have shorter life expectancy than those not considered. In sum, previous literature has explored linkages between high rank and lifespan without producing a consistent account of the causal effects of promotion.

2.2

Data and the Counterfactual Experiment

In contrast to the previous work on presidential life expectancy, I focus my analysis on the difference between presidents and their closest comparison group, the “runners-up” in presidential elections. This comparison implicitly adjusts for the characteristics which predict selection into candidacy, as all members of the sample have been selected in this way. The comparison of presidents to runners-up allows me to reduce the potential explanations for the shorter life expectancy of presidents to either the effects of service, or an electoral advantage of shorter lived-candidates. I return to the possibility of an electoral preference for characteristics associated with shorter life expectancy in Section 4. In the analysis, I focus on the outcome of the election as the random event; where ambiguities arise, I take the runner-up from the candidates eligible for the presidency on the popular ballot.15 Data on the birth and death dates of presidential candidates is available from numerous sources, and an Online Appendix reports the runners-up and dates of birth and death of the candidates. Table 1 summarizes the population of candidates, treating each candidacy as an individual observation. US presidential candidates are 56.02 years old on average. Winners (irrespective of cause of death) are 56.67 years years old, 1.31 years older than runners-up. A t-test fails to reject the null of no difference in mean age at election between winners and runners-up (t = 0.94; p = 0.35). The full sample size is 112 candidacies, composed of 72 individuals, 57 of whom have died of natural causes, with 4 assassinations representing 6 candidacies; no losing candidate has been assassinated. The sample statistics include individuals who appear multiple times as candidates; in other words, the sample is candidacies, not individuals. For example, Andrew Jackson narrowly lost the 1824 election, an election in which he actually garnered 14

This paper surveys the medical and social science literature in greater depth than I do here. Thus, I exclude “technical” presidential candidacies such as Aaron Burr in 1800, and include Horace Greeley, the losing presidential candidate in 1872, who died before the electoral college could meet. Greeley has the shortest survival of any candidate, winner or runner-up. 15

5

more electoral and popular votes than the winner, John Quincy Adams, but failed to take a majority of electoral votes. This sent the election to the House of Representatives, which elected Adams. Following the defeat, Jackson went on to win the 1828 and 1832 elections, and serve a total of 8 years. In the empirical model, I treat Jackson as three times a candidate, losing once and winning twice.16 In the analysis, I use weighting and clustered standard errors to address the multiple candidacies problem; however, summary statistics and figures report raw data. Although the age differences between the candidates at the time of election are not statistically significant, selection on health and age requires additional discussion. The process of selection into candidacy explicitly considers the expected survival of the candidate; specifically, selection on the health of candidates increases with age, meaning the age-specific hazard rate will be a function of the age of the candidate at the time of the election. I call this the “Ronald Reagan problem,” after the oldest president at election.17 It is natural to believe that candidate health matters most over the term of service, so that candidates will be selected on a high probability of survival for an additional 4 years. For example, compare the expected age at death (at the time of selection into candidacy) of Ronald Reagan, elected for the second time at age 73, to Barack Obama, elected for the first time at age 47. For Reagan, an additional 4 years of life expectancy placed him close to the expected survival of a man of Barack Obama’s age (not accounting for the fact that Obama is or was a cigarette smoker). As a result, Reagan’s age (and health) was an important campaign issue.18 On the other hand, it is unlikely the electorate placed high weight on Obama’s survival from age 73 to 77. Due to the context, we should not expect Obama and Reagan to have the same age-specific mortality rates or life expectancy. In the empirical analysis with age at death as the outcome, I account for this by controlling for age at election, a term which is significantly different from 0 (no Ronald Reagan problem) and 1 (perfect selection on remaining life expectancy) at p < 0.01 in all specifications. 16

In a setting (US House Elections) with many more observations, Dal Bo et al. (2009) restricts attention to candidates standing for their first re-election. 17 Thank you to Ken Wachter for suggesting this name. 18 Reagan famously neutralized the issue of his age in a debate with Walter Mondale, saying “I will not make age an issue of this campaign. I am not going to exploit, for political purposes, my opponent’s youth and inexperience.” Mondale later cited this quip as the moment his chance of winning came to an end.

6

3 3.1

Effects of Presidential Election and Succession Graphical Results

As the primary visual test of the hypothesis that presidents age faster than electoral runners-up, I plot the survival of US presidents and runners-up using Kaplan-Meier survival curves. This method permits the inclusion of still-living and assassinated presidents as censored observations, so the figures represent all 112 candidacies. Figure 1 displays the primary result. In the upper panel, the survival curves of presidents and runners-up reveal the superior survival of runner-up candidates at all time horizons. Both curves display a S-shape consistent with selection on health at the time of election, and resemble a normal-distribution survival function. Survival does not appear sharply differentiated until after the 10th year, and the majority of the lost life years appear in the second and third decades following election. Such an effect is consistent with selection on a minimum survival time for candidates. A proportional hazards assumption appears reasonable. The lower panel plots survival estimates with a cubic age adjustment. To construct this, I first run a regression of survival on a cubic polynomial in age for candidates who have died of natural causes, then predict survival based on age for all candidates (note the sample is balanced on year of election; results are quite similar when including additional controls or using alternative polynomials in age). I plot the residual between this prediction and realized survival, with living candidates truncated as in the KaplanMeier estimates above; the smallest residual is set to zero. This regression adjustment has only a small effect on the survival curves, but does serve to narrow the distance between the curves by several years among those surviving longest. It does not appear that the small gap in age at election between candidates affects the qualitative conclusion that runners-up have outlived presidents.

3.2

Censored Regression

My preferred regression specifications use the Tobit model, which can handle the Ronald Reagan problem through the inclusion of age controls, as well as the censoring of stillliving candidates’ survival times. The age controls serve to increase the precision of the estimates and correct for the (not significant) difference in age at election. Expected survival in Tobit follows a normal distribution, the tail of which closely matches the empirical distribution of expected human survival from the beginning of the Gompertz years of mortality through age 90, and the empirical pattern of presidential candidate

7

survival. Tobit has the additional advantage of producing linear estimates interpretable as lost years of life. In practice, these assumptions do not play a large role, and the results from a hazard model coincide with these estimates. The general specification for individual i, running for election in year t is 2 AgeAtDeathi,t = β0 + β1 [W ini,t ] + β2 Agei,t + β3 ElectY eari,t + β4 ElectY eari,t

+ [AgeAtDeathi,t − Agei,2013 |Ii = 1] + �i,t ,

(1)

where Ii is an indicator for survival until January 1, 2013, the point of right-truncation of the sample. The expectation, [AgeAtDeathi,t − Agei,2013 |Ii = 1], follows the assumed normal distribution, and is estimated in a first-stage that includes the same variables as the main specification. The linear age term and quadratic in election year were chosen by adding higher-order terms until the highest-order term became statistically insignificant; results are similar when using birth cohort in place of election year. The sample is weighted by the inverse of the candidate’s appearances to avoid placing undue weight on the individuals who appear multiple times. Standard errors are clustered by individual. Table 2 reports the main regression results. Column 1 reports the basic specification that does not control for previous terms served, finding an gap in conditional life expectancy of 4.5 years between winning and runner-up candidates. If service does exact a toll, we should prefer specifications which control for previous exposure; so, columns 2 through 5 add controls for previous terms served. The preferred specification is Column 2, in which winning candidates suffer a loss of 3.8 years of life expectancy. Previous terms served are associated with 2.6 years shorter life expectancy, consistent with the effects of presidential service cumulating over terms served. Although the previous terms served coefficient is not significant, we cannot reject the equality of the previous terms served coefficient and the main effect.19 Column 3 drops still-living candidates, finding a similar loss of life years (4.3 years), meaning the inclusion of still-living presidents (the censored observations) are not crucial to the result. Column 4 drops the sample weights, which leads to an improvement in precision that more than offsets a small decrease in the estimated effect, to 3.2 years of life lost. In the unweighted specification, previous terms reduce life expectancy by 3.0 years, slightly less than the estimated effect of winning the election. In all specification, the estimated effect of presidential service is significantly negative, with life expectancy reduced by around 4 years. The specifications reported here summarize the broad patterns in the data, and are 19

We would expect the previous terms coefficient to be smaller than the treatment effect, because the presidents with the worst health outcomes do not run for president again.

8

reasonably robust to alternatives. Election fixed effects increase the magnitude and significance of the effects. The effects are also quite similar when censoring the Ronald Reagan and Theodore Roosevelt observations as the time these presidents were shot (in failed assassination attempts). Attempts to add controls have had varying impacts on the main result. For example, controls for parent’s life expectancy are themselves far from significant (p > 0.5), but their inclusion does compromise the statistical significance of the main effect.20 On the other hand, controlling for height increases the magnitude of the main effect and its statistical precision in any model in which it is included. Perhaps surprisingly, height itself predicts a shorter life at half-a-year per inch (p < 0.01), almost exactly according to the estimates for healthy, modern males.21 Adding a dummy for military service leaves the main point estimate unchanged at -3.6 years (p = 0.067), though again, the dummy itself is not significant. I have also explored expanding the sample to candidates who finish 3rd and below in the electoral college, and vice-presidents. This changes the counterfactual experiment to one in which presidents are compared to men and women from lower in the political hierarchy, and point estimates fall to around 3 years of lost life. It is not clear if this decrease in the estimated effect is because the newly added sample members have lower life expectancy (i.e. are a poor counterfactual), or if the estimate is truly improved. Studying the extremes of distributions means dealing with small samples, and it is probably unwise to read too much into these patterns. What is clear is that the average survival of men who have been elected president has been shorter than those who have finished as runners-up, and this historical precedent is large enough to meet traditional levels of statistical significance.

4 4.1

Discussion Might the Electorate Prefer Shorter-lived Candidates?

I have documented the shorter life expectancy of US presidents in comparison to electoral runners-up. A primary concern with interpreting these results as a causal effect of 20 There are two reasons this model loses statistical power: parent’s life expectancy is unrealized for a number of modern presidents, decreasing the sample size; as well, life expectancy is endogenous for parents still-living at the time of election. 21 More on height: Height is the only covariate I have uncovered which is itself significant. Historically, height does not significantly predict electoral success in presidential elections; in my data, each inch of height is associated with 0.62% increase in the probability of victory, but this is far from statistically significant (p = 0.38). Nevertheless, recent research in political science argues such an association exists (Murray and Schmitz, 2011). See, for example, Samaras and Storms (1992) for evidence on the modern height-life expectancy relationship in well-nourished populations.

9

service is the selection effects in the final round of voting. In other words, are the lives of losing presidential candidates truly a good counterfactual for winners? In the introduction, I suggested we can put these issues aside due to an assumed electoral advantage for healthier candidates, which would imply the results are biased against the finding. Political economy models have emphasized the desirable incentives for long-lived rulers (Acemoglu (2005), Besley and Persson (2009)). Other research shows groups elect healthy members to lead them, and the performance and perception of leaders is related to their health (Little et al., 2007). The electorate’s preference for the candidate with superior life expectancy seems a reasonable assumption for the selection directly on health and life expectancy, however, it is possible that differences in other dimensions of the selection process into candidacy and the selection process in the general election can explain this result. There are a number of plausible reasons for an electoral preference for characteristics that are associated with shorter life expectancy. In general, it could be that selection along dimensions associated with longer life expectancy occurs in the nomination process, i.e. “class,” but the general electorate prefers candidate characteristics associated with a shorter lifespan i.e. “grit.” For example, the electorate may prefer candidate who have served in the military, or were born poor. Alternatively, it is possible the general election involves candidates competing on commitments to the electorate. If the candidate’s probability of winning is increasing in his commitments, and commitments have some power, candidates may end up competing on life-expectancy-reducing campaign promises: “I will do (a lot) in my first 100 days,” or “I will pass this controversial piece of legislation.” Controlling for previous terms served should answer most concerns regarding incumbency advantage, however, complex stories may be constructed. Strategically, parties might nominate longer-life expectancy candidates against incumbents, as they have the greatest chance of re-appearing. Based on this list, there are plausible reasons to suspect the electorate may favor characteristics associated with shorter-lived candidates, however, we must weigh these secondary electoral motives against the the direct factors favoring healthy candidates. Presidential candidates certainly desire to appear healthy, for example, releasing the results of a medical exam is standard in modern elections. A frequently-cited turning point in the 1960 presidential election occurred during a televised debate, when Richard Nixon appeared haggard and pale (i.e. less healthy) than the younger John F. Kennedy, Jr. To take another historical example, during the 1900 campaign William Jennings Bryant traveled 19,000 miles and gave 546 speeches, while his opponent, Theodore Roosevelt,

10

traveled 21,000 miles and gave 673 speeches (Cherney (1985); the difference in number of appearances likely resulted from the superior fund-raising successes of the Republican candidate). Even though retail campaigning did not become standard until the late 19th century, the health of a candidate could only help gather support through travel and longer work-days. History is replete with examples of men who were denied nomination due to health or age. Although Franklin Delano Roosevelt was elected 4 times while in a wheelchair, the truth is that he did not reveal his affliction to the public until late in his presidency. I interpret the weight of the historical evidence as supporting the assumption that health, particularly conditioned on age, predicts electoral success. Another important aspect of the counterfactual that I have not discussed is the effect of losing. Becker et al. (2008) find a negative effect on life expectancy of narrowly missing election to the Baseball Hall of Fame. While it is possible to construct a story in which losing presidential candidates experience a windfall gain in life expectancy (in some sense, this is the finding of the paper), it is more plausible that candidates make significant investment in the nomination and campaign process that yield smaller returns upon losing. To the extent that the experience of losing negatively affects life expectancy, the comparison made in this paper understates the loss of life resulting from service. Finally, even if the findings are explained by electorate’s preference for candidates with characteristics correlated with shorter life expectancy, the result remains a counterexample to the rank-lifespan correlation. If democratic elections favor candidates with shorter life expectancy, then the correlation is reversed in this setting. Might certain hierarchies promote leaders with ex-ante shorter life expectancy? As far as I know, this hypothesis has not been explored in the literature.

4.2

Mechanisms

If the negative health effects can be assigned to service, which aspects of presidential service are most likely to explain them? Based on Figure 1, the loss of presidential life to natural causes may begin soon after service, however, the majority of deaths occur in the years after the term served. This suggests the effect occurs through a drawdown of health capital in the process of service, consistent with reports of increased aging of presidents. Based on the evidence from other primates, I investigated the role of margin of victory in the estimates. If instability of the political hierarchy counteracts the effects of rank, we would expect candidates with smaller margins of victory to experience more lost life years. This is not the case: a higher vote share predicts a shorter life, and some of the shortest-lived presidents won by considerable margins. The sample size is far too small 11

to use a regression-discontinuity design, but what evidence we have does not support a strong role for stability in the political system on lifespan. These effects would also be consistent with environmental risks associated with service, such as increased exposure to disease. Presidents travel around the country and world, and meet with many people. Before the germ theory of disease, travel entailed exposure to water-borne disease and other infections. William Henry Harrison contracted pneumonia in his first weeks in office, when presidents solidify support by meeting with the political elites in the capital. As discussed, one potential mechanism I have ruled out is the physical move to the capital. This hypothesis is of particular interest, given the prevalence of malaria in the capital. If this was the primary mechanism, we would expect to see vicepresidents experience a loss of life. Nevertheless, it is likely that some portion of the effect can be explained by the exposure to additional risk factors beyond the rigors of the office, or represent the combination of such factors with work-related stress. Although I lack the statistical power to distinguish between mechanisms, in my view, the hypothesis of work-related aging remains the most plausible explanation for the findings. Two exemplars are the presidencies of Woodrow Wilson and Franklin Delano Roosevelt, the presidents who served during the world wars in the twentieth century. Both men traveled extensively as a result of the conflicts and worked themselves to the point of exhaustion. Near the conclusion of each wars, both presidents’ health failed, leading to their death. This anecdotal evidence is consistent with a model in which presidents face a variable demand for their services, and whether through altruistic motivations or unbreakable commitments, respond with an increase in work effort leading to a loss of life expectancy. The potential to select men willing to make these sacrifices seems to be exactly what democratic elections are intended to accomplish.

5

Conclusion Jimmy Carter on the Curse of Tippecanoe (presidents elected in years divisible by 20 between 1840 and 1960 died while in office): “I’m not afraid. If I knew it was going to happen, I would go ahead and be president and do the best I could, for the last day I could.” -October 2, 1980, at a presidential campaign rally in Dayton, Ohio

In this paper, I establish a stylized fact: on average, losing presidential candidates outlive presidents who die of natural causes by around 4 years. I argue this difference arises from the causal impact of presidential service, based on the assumption that the electorate 12

prefers longer-lived candidates. To the extent this assumption holds, these quantitative estimates understate the negative effects of presidential service on life expectancy. I cannot assign the effect to one particular mechanism, but emphasize the likely role of work-related stress and aging. That presidents are willing to sacrifice life expectancy in exchange for the returns to service is not a surprise, given the known risks of assassination. What is novel about this result is the loss of natural life expectancy. To my knowledge, this is the first paper to corroborate the reversal of the natural life expectancy gradient at the top of a human hierarchy. This finding complicates rank-based explanations for the socioeconomic gradient in life expectancy, by both supporting the proposed mechanism in this literature, while also providing a clear counterexample at the logical extreme of the theory. From the evidence presented here, I conclude that the mechanism of work-related stress may have sufficient force to explain large-scale patterns in life expectancy, but that the patterns of exposure to work-related stress do not support the reduction to direct relationship between social status and life expectancy, as the rank-lifespan theory implies. It is important to note that a defining characteristic of the socioeconomic gradient is that it extends throughout the socioeconomic distribution, so that the extension to the upper tiers of hierarchies is implied. While the US presidency is clearly an unusual job, we might expect the same effects to be found among other high-profile, high-stress occupations. Given the difficulties in gathering data on the very top of the distribution, some of the best evidence on the lives of these individuals is likely to come from historical examples, such as this one. The broadest findings of this investigation speak to a larger role for health in models of labor supply at older ages, particularly for high-achieving individuals. In their labor supply decisions, presidents choose to sacrifice not just leisure, but also a portion of their time endowment. This is obvious from the risk of assassination; however, this investigation finds that risk of assassination has accounted for less than one-half of the years lost to presidential service. The modeling conveniences of assuming a fixed lifespan or an infinite horizon cannot capture the tradeoff documented here, suggesting that for certain categories of workers, models of the retirement decision must account for the interaction of career, extensive-margin behavior and lifespan. As well, the returns to presidential service may or may not include increased consumption, but consumption differentials between winners and losers are unlikely to compensate presidents for the years of lost life. If presidents expect to lose 5 out of 20 remaining life years, and we assume log utility over consumption and no discounting (i.e. quite conservative assumptions), then this sacrifice would require to a 212% increase in consumption in the remaining years to

13

leave an individual indifferent. This compensating differential suggests we look beyond strictly consumption-based utility, to elements such as prestige and ego rents, risk-seeking behavior, dynastic concerns and altruism as the motivations of presidents. I conclude that research on the labor supply behavior among high-achieving older workers should treat health, lifespan, extensive margin choices, selection into occupations and careers, and beyond-consumption utility as essential objects. Finally, to the degree that work-related stress is truly concentrated in the lower-tiers of the income distribution, there may be scope for a more general role for health in the modeling of labor supply.

References Acemoglu, D. (2005). Politics and economics in weak and strong states. Journal of Monetary Economics 52 (7), 1199 – 1226. Anderson, M. and M. Marmot (2011). The Effects of Promotions on Heart Disease: Evidence from Whitehall. The Economic Journal 121, 1–35. Becker, D. J., K. Y. Chay, and S. Swaminathan (2008). Mortality and the Baseball Hall of Fame: An Investigation into the Role of Status in Life Expectancy. Working Paper. Besley, T. and T. Persson (2009). The Origins of State Capacity: Property Rights, Taxation, and Politics. American Economic Review 99 (4), 1218–44. Brown, J., J. Liebman, and J. Pollet (2002). Estimating Life Tables That Reflect Socioeconomic Differences In Mortality, Chapter Appendix, pp. 447–458. University of Chicago Press. National Bureau of Economic Research. Case, A. and C. Paxson (2011). The Long Reach of Childhood Health and Circumstance: Evidence from the Whitehall II Study. The Economic Journal 121 (554), F183–F204. Cherney, R. W. (1985). A Righteous Cause. Little, Brown and Company. Cutler, D., A. Deaton, and A. Lleras-Muney (2006). The Determinants of Mortality. Journal of Economic Perspectives 20 (3), 97–120. Dal Bo, E., P. Dal Bo, and J. Snyder (2009). Political Dynasties. The Review of Economic Studies 76 (1), 115–142.

14

Epel, E. S., E. H. Blackburn, J. Lin, F. S. Dhabhar, N. E. Adler, J. D. Morrow, and R. M. Cawthon (2004). Accelerated Telomere Shortening in Response to Life Stress. Proceedings of the National Academy of Sciences of the United States of America 101 (49), 17312–17315. Fogel, R. W. (2004). The Escape from Hunger and Premature Death, 1700-2100. Cambridge University Press. Hanna, J. (2011, August). Do Presidents Age Faster in Office? CNN Online Article; Accessed January 7, 2014. Kopczuk, W. and E. Saez (2004). Top Wealth Shares in the United States, 1916-2000: Evidence from Estate Tax Returns. National Tax Journal 57, 445–487. Little, L. M., B. L. Simmons, and D. L. Nelson (2007). Health Among Leaders: Positive and Negative Affect, Engagement and Burnout, Forgiveness and Revenge. Journal of Management Studies 44 (2), 243–260. Malmendier, U. and G. Tate (2009). Superstar CEOs. The Quarterly Journal of Economics 124 (4), 1593–1638. Marmot, M. (1994). Social Differentials In Health Within and Between Populations. Daedalus 123:4, 197–216. Marmot, M. (2004). The Status Syndrome: How Social Standing Affects Our Health and Longevity. New York: Holt. Marmot, M., S. Stansfeld, C. Patel, F. North, J. Head, I. White, E. Brunner, A. Feeney, M. Marmot, and G. Smith (1991). Health Inequalities among British Civil Servants: the Whitehall II Study. The Lancet 337 (8754), 1387 – 1393. Originally published as Volume 1, Issue 8754. McEwen, B. S. (1998). Protective and Damaging Effects of Stress Mediators. New England Journal of Medicine 338 (3), 171–179. Moslehi, J., R. A. DePinho, and E. Sahin (2012). Telomeres and Mitochondria in the Aging Heart. Circulation Research 110 (9), 1226–1237. Murray, G. R. and J. D. Schmitz (2011). Caveman Politics: Evolutionary Leadership Preferences and Physical Stature. Social Science Quarterly 92 (5), 1215–1235. 15

Olshansky, S. J. (2011). Aging of US Presidents. JAMA: The Journal of the American Medical Association 306 (21), 2328–2329. Persico, N., A. Postlewaite, and D. Silverman (2004). The Effect of Adolescent Experience on Labor Market Outcomes: The Case of Height. Journal of Political Economy 112 (5), pp. 1019–1053. Rablen, M. D. and A. J. Oswald (2008). Mortality and Immortality: The Nobel Prize as an Experiment into the Effect of Status upon Longevity. Journal of Health Economics 27 (6), 1462 – 1471. Rosen, S. (1981). The Economics of Superstars. The American Economic Review 71 (5), pp. 845–858. Samaras, T. and L. Storms (1992). Impact of Height and Weight on Lifespan. Bulletin of the World Health Organization 70 (2), 259. Sapolsky, R. M. (2005). The Influence of Social Hierarchy on Primate Health. Science 308 (5722), 648–652. Stringhini, S., S. Sabia, M. Shipley, E. Brunner, H. Nabi, M. Kivimaki, and A. SinghManoux (2010). Association of Socioeconomic Position With Health Behaviors and Mortality. JAMA: The Journal of the American Medical Association 303 (12), 1159– 1166. Vaananen, A., S. Toppinen-Tanner, R. Kalimo, P. Mutanen, J. Vahtera, and J. M. Peiro (2003, September). Job Characteristics, Physical and Psychological Symptoms, and Social Support as Antecedents of Sickness Absence among Men and Women in the Private Industrial Sector. Social Science & Medicine 57 (5), 807–824.

16

Table 1: Summary Statistics Age at Election Mean SD N All 56.02 7.38 112 Winners 56.67 6.56 56 Win, Natural Cause 57.25 6.52 50 Runners-up 55.36 8.13 56 t-test (N =112) 1.31 1.40 t=0.94 Age at Death All 72.95 11.56 97 Winners 70.93 11.60 49 Win, Natural Cause 73.25 10.29 43 Runners-up 75.01 11.26 48 Notes: Age at Election and Death, Presidential Candidates in 1789-2008 Elections. Ages separately reported for presidents who died of natural causes, the primary sample of interest. Winning candidates (irrespective of cause of death) have been 1.31 years older than losers, on average; a t-test reveals this difference is not significant.

17

Table 2: Age at Death (1) Winner -4.47* (2.09) Previous Terms Age at Election Tobit Weights Censor Assassinations Candidacies Individuals Deaths

-0.52** (0.14) X X X 112 72 57

(Survival Past Election) (2) (3) (4) -3.81* -4.31* -3.21* (1.89) (2.02) (1.61) -2.63 -2.15 -2.97* (1.77) (1.94) (1.50) -0.55** -0.51** 0.49** (0.14) (0.15) (0.12) X X X X X X X 112 91 112 72 57 72 57 57 57

(5) -6.66** (2.22) -1.93 (1.71) -0.69** (0.16) X X 112 72 61

Notes: Top 2 finishers, US Presidential Candidates in 1789-2008 Elections. Specifications (1)-(4) consider deaths from natural causes (treating assassinated and still-living candidates as censored observations); (5) adds assassinations (still-living candidates remain censored). Tobit model (all except (3)) accounts for right-censoring due to still-living candidates. Identical results on the main effect would be obtained taking survival past election as the outcome, due to linear control for age at election; all models also includes quadratic in election year. Observations weighted by the inverse of number of appearances in samples (1)-(3) and (5); standard errors clustered by individual. Significance levels: ** p<0.01, * p<0.05, † p<0.1.

18

Figure 1: Kaplan-Meier and Age-adjusted Survival

Kaplan−Meier survival estimates 1.00

Presidents Runners−up

0.75

0.50

0.25

0.00 0

10

20 30 Years Since Election

40

50

Age−adjusted survival estimates 1.00

Presidents Runners−up

0.75

0.50

0.25

0.00 0

10

20 30 Years Survived, age−adjusted

40

50

Notes: The figures plot the empirical probability a president or losing candidate survives the given years. Assassinated presidents and still-living candidates are treated as censored observations. Cubic age-adjustment accounts for slightly older age of winners, and selection on health as age rises.

19

Data Appendix (For Online Publication)

20

Table A.1: Presidential Candidates Birth and Name Year Outcome Birth Date Washington 1789 Winner 22feb1732 Adams 1789 Runner-up 30oct1735 Washington 1792 Winner 22feb1732 Adams 1792 Runner-up 30oct1735 Adams 1796 Winner 30oct1735 Jefferson 1796 Runner-up 13apr1743 Jefferson 1800 Winner 13apr1743 Adams 1800 Runner-up 30oct1735 Jefferson 1804 Winner 13apr1743 Pinckney 1804 Runner-up 25feb1746 Madison 1808 Winner 16mar1751 Pinckney 1808 Runner-up 25feb1746 Madison 1812 Winner 16mar1751 Clinton 1812 Runner-up 02mar1769 Monroe 1816 Winner 28apr1758 King 1816 Runner-up 24mar1755 Monroe 1820 Winner 28apr1758 Adams 1820 Runner-up 11jul1767 Adams 1824 Winner 11jul1767 Jackson 1824 Runner-up 15mar1767 Jackson 1828 Winner 15mar1767 Adams 1828 Runner-up 11jul1767 Jackson 1832 Winner 15mar1767 Clay 1832 Runner-up 12apr1777 VanBuren 1836 Winner 05dec1782 Harrison 1836 Runner-up 09feb1773 Harrison 1840 Winner 09feb1773 VanBuren 1840 Runner-up 05dec1782 Polk 1844 Winner 02nov1795 Clay 1844 Runner-up 12apr1777 Taylor 1848 Winner 24nov1784 Cass 1848 Runner-up 09oct1782 Pierce 1852 Winner 23nov1804 Scott 1852 Runner-up 13jun1786 Buchanan 1856 Winner 23apr1791 Fremont 1856 Runner-up 21jan1813 Lincoln 1860 Winner 12feb1809 Breckinridge 1860 Runner-up 16jan1821

21

Death Dates Death Date 14dec1799 04jul1826 14dec1799 04jul1826 04jul1826 04jul1826 04jul1826 04jul1826 04jul1826 16aug1825 28jun1836 16aug1825 28jun1836 11feb1828 04jul1831 29apr1827 04jul1831 23feb1848 23feb1848 08jun1845 08jun1845 23feb1848 08jun1845 29jun1852 24jul1862 04apr1841 04apr1841 24jul1862 15jun1849 29jun1852 09jul1850 17jun1866 08oct1869 29may1866 01jun1868 13jul1890 15apr1865 17may1875

Table A.2: Presidential Candidates Name Year Outcome Lincoln 1864 Winner McClellan 1864 Runner-up Grant 1868 Winner Seymour 1868 Runner-up Grant 1872 Winner Greeley 1872 Runner-up Hayes 1876 Winner Tilden 1876 Runner-up Garfield 1880 Winner Hancock 1880 Runner-up Cleveland 1884 Winner Blaine 1884 Runner-up Harrison 1888 Winner Cleveland 1888 Runner-up Cleveland 1892 Winner Harrison 1892 Runner-up McKinley 1896 Winner Bryan 1896 Runner-up McKinley 1900 Winner Bryan 1900 Runner-up Roosevelt 1904 Winner Parker 1904 Runner-up Taft 1908 Winner Bryan 1908 Runner-up Wilson 1912 Winner Roosevelt 1912 Runner-up Wilson 1916 Winner Hughes 1916 Runner-up Harding 1920 Winner Cox 1920 Runner-up Coolidge 1924 Winner Davis 1924 Runner-up Hoover 1928 Winner Smith 1928 Runner-up Roosevelt 1932 Winner Hoover 1932 Runner-up Roosevelt 1936 Winner Landon 1936 Runner-up Roosevelt 1940 Winner Willkie 1940 Runner-up

22

Birth and Death Dates (con’t) Birth Date Death Date 12feb1809 15apr1865 03dec1826 29oct1885 27apr1822 23jul1885 31may1810 12feb1886 27apr1822 23jul1885 03feb1811 29nov1872 04oct1822 17jan1893 09feb1814 04aug1886 19nov1831 19sep1881 14feb1824 09feb1886 18mar1837 24jun1908 31jan1830 27jan1893 20aug1833 13mar1901 18mar1837 24jun1908 18mar1837 24jun1908 20aug1833 13mar1901 29jan1843 14sep1901 19mar1860 26jul1925 29jan1843 14sep1901 19mar1860 26jul1925 27oct1858 06jan1919 14may1852 10may1926 15sep1857 08mar1930 19mar1860 26jul1925 28dec1856 03feb1924 27oct1858 06jan1919 28dec1856 03feb1924 11apr1862 27aug1948 02nov1865 02aug1923 31mar1870 15jul1957 04jul1872 05jan1933 13apr1873 24mar1955 10aug1874 20oct1964 30dec1873 04oct1944 30jan1882 12apr1945 10aug1874 20oct1964 30jan1882 12apr1945 09sep1887 12oct1987 30jan1882 12apr1945 18feb1892 08oct1944

Table A.3: Presidential Name Year Truman 1948 Dewey 1948 Eisenhower 1952 Stevenson 1952 Eisenhower 1956 Stevenson 1956 Kennedy 1960 Nixon 1960 Johnson 1964 Goldwater 1964 Nixon 1968 Humphery 1968 Nixon 1972 McGovern 1972 Carter 1976 Ford 1976 Reagan 1980 Carter 1980 Reagan 1984 Mondale 1984 Bush 1988 Dukakis 1988 Clinton 1992 Bush 1992 Clinton 1996 Dole 1996 Bush 2000 Gore 2000 Bush 2004 Kerry 2004 Obama 2008 McCain 2008

Candidates Outcome Winner Runner-up Winner Runner-up Winner Runner-up Winner Runner-up Winner Runner-up Winner Runner-up Winner Runner-up Winner Runner-up Winner Runner-up Winner Runner-up Winner Runner-up Winner Runner-up Winner Runner-up Winner Runner-up Winner Runner-up Winner Runner-up

23

Birth and Death Dates (con’t) Birth Date Death Date 08may1884 26dec1972 24mar1902 16mar1971 14oct1890 28mar1969 05feb1900 14jul1965 14oct1890 28mar1969 05feb1900 14jul1965 29may1917 22nov1963 09jan1913 22apr1994 27aug1908 22jan1973 02jan1909 29may1998 09jan1913 22apr1994 27may1911 13jan1978 09jan1913 22apr1994 19jul1922 21oct2012 01oct1924 14jul1913 26dec2006 06feb1911 05jun2004 01oct1924 06feb1911 05jun2004 05jan1928 12jun1924 03nov1933 19aug1946 12jun1924 19aug1946 22jul1923 06jul1946 03mar1948 06jul1946 11dec1943 04aug1961 29aug1936