Studying Leaders and Elites: The Personal Biography Approach

The last two decades have seen a revival in work that takes the role of individual leaders and elites seriously. This article surveys new research that explores how biographical factors influence their behavior. We call this literature the personal biography approach to political leadership. Our survey first lays out four mechanisms through which biographical characteristics might affect leader behavior. We then discuss the main findings, grouping them according to socializing experiences (e.g., education, military service, and prior occupation) and ascriptive traits (e.g., gender, race, and ethnicity). We also consider the methodological problems, especially endogeneity and selection effects, that pose challenges to this style of research. We conclude with an assessment of gaps in the literature and provide suggestions for future work in the biographical vein. 133 A nn u. R ev . P ol it. S ci . 2 02 0. 23 :1 33 -1 51 . D ow nl oa de d fr om w w w .a nn ua lr ev ie w s. or g A cc es s pr ov id ed b y N or th w es te rn U ni ve rs ity o n 11 /1 5/ 20 . S ee c op yr ig ht f or a pp ro ve d us e. PL23CH08_Roberts-A ARjats.cls April 25, 2020 22:15


INTRODUCTION
The conventional wisdom about the importance of individual leaders for politics has changed remarkably over time. Political science scholarship once largely accepted Thomas Carlyle's maxim that "the history of the world is but the biography of great men." In international relations (IR), influential scholars such as Hans Morgenthau and Henry Kissinger placed great weight on the role of leaders when explaining international conflict and cooperation. In comparative politics (CP), the profiling of political elites, particularly in communist nations such as China and Russia, was popular among early behavioralists (Leites 1951). In American politics (AP), the study of the presidency paid close attention to the character of those who held the office.
But the field eventually took theoretical and methodological turns that prompted most political scientists to shy away from analyzing individual leaders. IR scholarship began to emphasize the structure of the international system at the expense of leaders in the wake of Kenneth Waltz's landmark books. Some even came to view leader-oriented explanations as unscientific. Gartzke & Rohner (2011, p. 533), for example, cast "particularistic" theories of foreign policy, "housed in the aspirations of individual leaders," as "more the stuff of historical and biographical research than of social science." For CP and AP scholars studying legislators, early work generally yielded ambiguous results (Putnam 1976). The field consequently moved toward institutionalist and strategic approaches that, for the most part, discouraged research on individual elites. 1 The study of political leaders has recently staged a dramatic comeback. A pair of developments likely played a role in this resurgence. First, real-world events offered some tantalizing counterfactuals. Would the United States have invaded Iraq in 2003 if only a few hundred votes in Florida had gone to Al Gore instead of George W. Bush? Would America's current foreign policy agendaranging from trade policy to the NATO alliance-look different if someone other than Donald Trump were president? 2 Second, methodological innovations have allowed scholars to make more credible inferences about leaders and elites.
This article focuses on new leader-centric research that moves beyond the earlier "great man" approach or a general insistence that elites somehow matter. The more recent wave of work makes a bolder claim: The personal attributes and life experiences of individual leaders affect important political outcomes in systematic, predictable ways. This article offers a comprehensive assessment of this claim by (a) reviewing research on both international and domestic politics and (b) adopting a broad definition of the term leader that includes national and local executives, cabinet officials, and legislators.
The payoff from our cross-subfield assessment of what we call the personal biography approach to the study of leaders is threefold. First, we outline four distinct mechanisms-beliefs and values, competence and skills, material interests, and the perceptions of others-through which leader attributes might shape political outcomes. Regardless of the attribute or outcome under study, scholars almost always rely (at least implicitly) on one of these mechanisms, an insight that can help unify work on personal biography across subfields. Second, we uncover gaps where each subfield could expand the traits that it studies. For instance, work on legislators in domestic politics has paid far more attention to the role of race, ethnicity, and gender than has research on heads of state in international politics. Third, different subfields have developed different methodological toolkits for studying leaders and elites. Our survey provides examples of these different approaches that subfields might usefully borrow from each other.
The rest of the article is organized as follows. We first describe the intuition behind the personal biography approach and explain how it differs from other leader-centric research. We then discuss the particular elements of personal biography emphasized in recent work, ranging from socializing experiences such as education, military service, and prior occupation to ascriptive characteristics like age, gender, and racial/ethnic identity. Next, we highlight the daunting methodological challenges that arise when attempting to estimate the independent effect of leaders' backgrounds on political outcomes and outline several potential solutions. Finally, we point to key remaining questions and suggest a path forward.

THE PERSONAL BIOGRAPHY APPROACH
The intuition behind the personal biography approach is straightforward: The characteristics and prior experiences of political elites have meaningful effects on their behavior. This idea has widespread appeal. Journalists, pundits, and even analysts at the Central Intelligence Agency have spent a good deal of time speculating about how leaders' personal attributes inform their approaches to governing. But until recently, this sort of information was missing from most political science models. Leaders were treated as interchangeable "black boxes" that responded to external constraints and opportunities in similar ways. Thus, the defining characteristic of the personal biography approach is the unpacking of the black box to examine how individual-level attributes and experiences influence political outcomes.
The personal biography approach relies on an important but often unstated assumption: Leaders have at least some level of discretion. If leaders were entirely hemmed in by structural or strategic considerations, then biographical factors would be irrelevant-and the black box approach to political leadership would be sufficient. As discussed below, biographical variables may matter a great deal in the study of outcomes where elites have a relatively high level of autonomy (such as foreign policy or bill proposals) and may matter little when constraints are severe (such as roll call votes).
Prior to scouring the backgrounds of leaders and elites, scholars first need to explain why and how features of individuals' biographies might matter at all. We highlight four theoretical pathways commonly invoked by scholars working in the personal biography vein (see Figure 1). 3 1. Biographical factors may affect the core beliefs and values held by political leaders. These beliefs constitute politicians' worldviews and underpin many of their policy preferences and priorities. They are deeper than and distinct from preferences induced by electoral  Pathways connecting personal biography to leader behavior. Biographical factors may affect a leader's core beliefs and values, competence and skills, and material and personal interests, as well as how others perceive the leader. considerations and party discipline. While much work views these beliefs and values as exogenous, the personal biography approach posits that they are shaped by ascriptive characteristics given at birth and/or socializing life experiences accrued prior to assuming office. 2. Biographical factors may affect a politician's competence and skills, or as some scholars put it, the "quality" of the leader (Besley et al. 2011). These skills might help achieve outcomes where the benefits accrue narrowly to the leader (such as reelection) or to society more broadly (such as economic growth). Personal experiences also provide politicians with information and expertise about specific issue areas, which may induce leaders to devote more attention and effort to those areas (Burden 2007). 3. Biographical factors may shape leaders' incentives to pursue particular policy agendas (Alexiadou 2016). Here we have in mind personal interests that benefit the politician economically or electorally. For example, leaders whose career paths included spells in the financial sector might be inclined to help their former colleagues (possibly to ensure that a lucrative job awaits after their time in office ends) by pursuing profinance policies (Adolph 2013). More generally, prior professional networks and personal investments can influence leaders' preferences and skew their decisions in favor of their own interests or the interests of members of their networks. Alternatively, characteristics such as race or gender can affect the sorts of policies that politicians use to attract votes from different sectors of the population. 4. Biographical factors may affect how others perceive leaders. These perceptions, in turn, can indirectly shape leader behavior. Identifiable features from a particular leader's background can influence other people's beliefs about that leader's skills, status, and orientations. Female politicians, for example, may have to behave differently in order to earn the same level of respect from peers that male politicians gain as a matter of course.
It can be difficult to distinguish these mechanisms in empirical work. For example, are politicians with military backgrounds more or less likely to engage in war because military service inculcated them with martial values, made them more competent in using force, or endowed them with reputations for toughness in the minds of their rivals? An important research frontier involves identifying when and where the different mechanisms are at play.
The personal biography approach is fundamentally different from three other strands of research that, to some degree, concern political leaders. The first explores how leaders' desire to stay in power and/or avoid punishment shapes their behavior (Chiozza & Goemans 2004). This approach restores leaders to the center of analysis, but it does not unpack the black box of individual leader characteristics. The second examines the political psychology of leader decision making ( Jervis 2017, Kertzer & Tingley 2018. This research agenda compellingly illustrates the importance of individual biases in elite decision making, but it focuses on general human cognitive processes and limitations, not on a leader's attributes or experiences. The third involves the idiosyncratic, character-shaping life events that are often crucial for historians and biographers. While these factors are no doubt relevant in specific cases (and are an important source of new hypotheses), the work reviewed here shares a commitment to scouring leader biographies for shared, measurable attributes and experiences that have systematic effects. 4

TWO CATEGORIES OF LEADER BIOGRAPHY
Which elements of leader biography matter, and how are they connected to political outcomes? For the sake of parsimony, we group the various elements of personal biography into two core categories, socializing experiences and ascriptive characteristics, though the distinction is not airtight. 5 Socializing experiences are formative experiences into which individuals can select themselves, such as education, military service, and occupation. By contrast, ascriptive characteristics are qualities-such as age, gender, and racial and ethnic identity-that are conferred at birth and are difficult (or impossible) to change.
Research on these two types of characteristics occurs in silos roughly corresponding with political science's subfields. IR scholarship overwhelmingly examines the socializing experiences of heads of state, whereas work in CP and AP focuses more on ascriptively diverse legislatures. We bring these strands of research together in this article, but we also highlight several differences. One important distinction is methodological: Since political elites can select into socializing experiences, it is often hard to separate the independent effect of the experience from the underlying factors that led the individual to obtain it in the first place (we return to this issue later). By contrast, self-selection effects are rarely a concern with ascriptive traits. 6 In the following two sections, we consider the main studies that examine the effect of socializing experiences and ascriptive traits on the behavior of political leaders. Our aim is to survey the key findings and the mechanisms through which these features produce their observed effects. We also provide some information about the methods of these studies with particular emphasis on state-of-the-art designs such as natural experiments, regression discontinuity (RD) designs, and difference-in-differences (DiD) approaches that address endogeneity concerns. Finally, we identify important gaps in the literature on each trait.

Level and Type of Education
Education, particularly university-level education, is commonly considered a key formative experience that affects outcomes through all four mechanisms (see Figure 1). Since education is a form of human capital, many have argued that educational attainment (often measured with a dichotomous variable for a college degree) is a proxy for leader quality or competency. Thus, Besley et al. (2011) find that highly educated leaders are associated with higher rates of economic growth. Others link leaders' levels of educational attainment to their political survival (Yu & Jong-A-Pin 2016), nationwide levels of primary education (Diaz-Serrano & Perez 2013), and corruption (Efobi 2015).
However, another set of empirical findings on leaders' educational attainment has delivered mixed results. Using both natural experiments and RD, Carnes & Lupu (2016) consider a host of different dependent variables and different types of leaders. Regardless of the outcome or type of leader, they find little or no relationship between leaders' lack or possession of a college degree and their behavior in office. Similarly, Hira (2007) finds no effect of leaders' economic education 5 Some variables, such as religion and class, could arguably fit in both categories. Ultimately, context should shape how we think about these categories. Religious identities, for instance, appear far less "sticky" in the United States than in Northern Ireland. 6 However, due to the (relative) immutability of ascriptive characteristics, there is some debate over their suitability for making causal claims (Sen & Wasow 2016). We sidestep this debate by focusing on the causal effect of a leader with trait X being replaced by an equivalent leader with trait Y, rather than the effect of the trait per se.
www.annualreviews.org • Studying Leaders and Elites on rates of economic growth, Horowitz et al. (2015) show that higher education of leaders does not affect international conflict, and Dreher et al. (2009) conclude that the link between leaders' educational backgrounds and growth is "not robust." Others take a more nuanced view of leader education. This is most evident in research that sees education as working through values, interests, and perceptions (rather than competence). Simply measuring whether a leader holds a university degree can mask a great deal of heterogeneity, and only certain types of education may be theoretically relevant. Gift & Krcmaric (2017), for example, highlight a connection between Western-educated leaders and democratization. They argue that university education in the West socializes future leaders to embrace democracy and creates transnational linkages that alter the strategic calculus of democratization. Barceló (2018) borrows this logic to argue that Western-educated leaders are more peaceful as well. Studies of economic policy making have also linked a particular kind of educational background-graduatelevel training in US economics departments-to a range of outcomes, including capital account openness (Chwieroth 2007), social spending (Chwieroth 2010), the fiscal balance (Kaplan 2018), and the tightness of monetary policy (Adolph 2013).
Elites' educational backgrounds might also affect the way that others view them. Nelson (2017) shows that countries with policy makers trained in top-ranked US economics departments are perceived as more trustworthy by the International Monetary Fund and as a result receive more generous loans. Others focus on the benefits that flow to leaders with technical educational backgrounds in fields such as economics, finance, or management. Flores et al. (2013), for example, find that credit rating agencies bestow better creditworthiness ratings on countries headed by leaders with graduate training in economics and finance from elite institutions. Overall, this work illustrates the potential importance of (a) where the leader was educated and (b) the leader's field of study.
We see two issues at the frontier of research on leader education. First, does education actually proxy competence in any meaningful sense? Moving beyond blunt indicators of educational attainment to finer-grained indicators may shed light on which types of educational backgrounds, if any, correlate with leadership quality. Second, future work should devote attention to training in fields other than economics-which has attracted the lion's share of attention-and introduce outcomes beyond economic performance (such as social and cultural policies) that education might influence.

Military Experience
Most work on the effects of military experience focuses on how it shapes elites' values. One perspective sees military training as a process that socializes participants to become more comfortable with using force to resolve disputes. The second perspective reaches the opposite conclusion: Military experiences, particularly facing the horrors of war firsthand, should induce a healthy dose of caution in military leaders compared to their civilian peers. Gelpi & Feaver's (2002) study of military veterans in elite positions within the US government found qualified support for the second perspective. As the proportion of military veterans in the Cabinet or Congress increased, the United States was less likely to initiate international disputes. When it did enter disputes, however, it used higher levels of force. Similarly, Lupton (2017) found that veterans in the US Congress were more likely to vote to exercise oversight and limit troop deployments during the Iraq and Afghanistan wars.
Mirroring the trajectory of studies of education's effect on leaders' behavior, scholars have started disaggregating the "former soldier" category to produce more nuanced measures. Horowitz et al. (2015) show that leaders with military training but no direct combat experiencea subset of leaders who are likely to consider themselves experts in military affairs but who have never been personally exposed to the dangers of war-are the most war prone. Other scholars move beyond formal participation in national militaries and highlight the role of rebel experience. Rebels who succeed and gain control of the state may be particularly risk tolerant as leaders, a proposition confirmed by studies finding that when they become national executives, they are more likely to pursue nuclear weapons (Fuhrmann & Horowitz 2015) and initiate international disputes (Horowitz et al. 2015).
The move from the soldier/civilian distinction to finer-grained measures of types of military experiences has been productive, but still more can be done on this front. One promising avenue involves studying the characteristics of rebel leaders themselves (Cunningham & Sawyer 2019). Another concerns the long-lasting effects of battlefield exposure on bystanders or survivors. Ellis (2018) argues that childhood exposure to wartime violence (not as a combatant) is likely to increase a leader's propensity for risk taking as an adult, resulting in more frequent entry into militarized disputes.
Looking forward, scholars might explore the links between military experience and outcomes other than armed conflict. The literature has focused almost exclusively on how military experience shapes leaders' beliefs, their competencies, and rivals' perceptions of them in the security realm, but military service has been linked to other kinds of beliefs, including tolerance for social difference, that could be tied to outcomes beyond armed conflict ( Jennings & Markus 1977).

Occupation and Social Class
Early studies of the effect of social class on the values of leaders found weak and inconsistent effects (Putnam 1976). Carnes & Lupu (2015) argue that this was due to imprecise measures of class (which equated it with education or income rather than occupation) and misplaced measures of outcomes (most work focused on highly constrained roll call votes rather than less constrained actions like bill sponsorship). Introducing these changes, Carnes (2013) finds that working-class representatives in the United States are more liberal than other legislators, a result of holding different values. Among Latin American legislators, Carnes & Lupu (2015) conclude that class "matters some of the time." Carnes & Sadin (2015) and Grumbach (2015), meanwhile, disagree on whether parents' social class affects the behavior of US congresspersons.
Some recent work has extended the investigation of these factors to the realm of executives and-in line with expectations-finds evidence that a blue-collar background makes political elites friendlier to expansionary and redistributive economic policies. Hayo & Neumeier (2016) show that heads of OECD governments from lower class status backgrounds were more likely to increase public deficits. Similarly, Göhlmann & Vaubel (2007) find that central bankers with prior experience in organized labor tend to be more inflation friendly. All of these studies seem to rely on a values mechanism, given that the material rewards for supporting redistributionist policies are less than for supporting free market policies.
Most theories about the impact of professional backgrounds on political behavior, by contrast, emphasize the interest mechanism-politicians are expected to favor policies that benefit their former industry. Adolph (2013) found strong evidence that central bankers from the financial sector are more inflation averse. Financial sector experience also appears to influence fiscal policies: In a study of budget balances in German Länder, Jochimsen & Thomasius (2014) find that ministers with financial backgrounds pursue tighter fiscal policies and achieve lower budget deficits.
Several studies have looked at the behavior of businesspeople-politicians, who often tout their competence. Witko & Friedman (2008) found modest effects of business experience on probusiness activities in the US Congress using an ordinary least squares design. Beach & Jones (2017) use a stronger RD approach to analyze California city council elections and find little evidence www.annualreviews.org • Studying Leaders and Elites that election of a businessperson affects expenditures, revenues, budget surplus, and unemployment. Szakonyi (2020), by contrast, finds more pernicious effects using an RD design in Russia. He shows that businessmen-politicians benefited their own firms by providing access to bureaucratic favors. These studies cast doubt on the claim that businesspeople have the ability to run the government efficiently like a firm.
Law has also attracted attention from scholars working in the personal biography tradition. Matter & Stutzer (2015) find that lawyer-legislators in the US Congress and 17 state legislatures voted differently than other legislators on matters of tort reform but not on other issues. They argue that this is an interest effect rather than a result of greater knowledge or competence because it correlates with recent practice of the law and is increased by looming term limits.  (2020), who shows that countries in NATO spend less on defense when governed by businesspeople due to their more egoistic preferences.

Political Experience
Even as politicians tout their political experience as a qualification for higher office, little research has asked whether political experience affects performance. We have in mind here the accumulation of office-holding experience prior to assessing behavior in the office or term that is being studied. One might expect strong effects on competence, but Volden & Wiseman (2014) found only limited influence of prior legislative experience on the effectiveness of members of Congress. Two studies have looked at the effect of a number of experience variables on expert ratings of US presidents. Simon & Uscinski (2012) argue that experiences similar to the presidency or specific aspects of the presidency positively predicted performance, but Balz (2010) finds little evidence for such effects. Potter (2007), however, shows that US presidents are less likely to engage in violent conflicts as time passes in their term, presumably due to lessons learned from experience. One potentially fruitful area for future work is considering how the experience of a leader's advisors interacts with the leader's own experience (Saunders 2017).

ASCRIPTIVE CHARACTERISTICS Gender
Gender has been studied extensively among elites (typically legislators due to the paucity of female executive leaders). Most of this work has focused on the values held by female politicians. Their value orientations are manifested in the greater priority they give to areas such as healthcare, education, family, housing, and so-called women's issues both in the United States (Swers 2013, Volden et al. 2018) and elsewhere (Paxton & Hughes 2016, Schwindt-Bayer 2006. Several scholars have utilized natural or quasi-natural experiments to show that differences in the priorities of female representatives persist even after accounting for omitted-variable bias and endogeneity. Thus, Macdonald & O'Brien (2011) compare male and female US legislators who replace each other in the same district and find different priorities. Chattopadhyay & Duflo (2004) take advantage of the random assignment of one-third of Village Council head positions in India to show that women invest more in public goods linked to women's concerns. Clots-Figueras (2011) focuses on close elections for state legislative assemblies in India that pitted a man against a woman and also included reserved seats for lower castes. She finds that female legislators in lower-caste seats invest more than male legislators in health, early education, and woman-friendly laws, but higher-caste women do not.
Turning to competence, Volden & Wiseman (2014) find that female legislators are more effective than males in passing legislation when they are in the minority because of their greater ability to build consensus (males tend to obstruct when in the minority), but women are less effective than men when in the majority. This finding supports an older literature demonstrating that women prioritize a legislative style that Paxton & Hughes (2016) characterize as "getting things done" over exercising control. In a large comparative study with a particular focus on Argentina, Barnes (2016) finds that female legislators collaborate more than males, though she attributes it largely to their marginalization. Volden et al. (2018) similarly see female legislators' difficulties in passing bills related to women's issues as connected to the lack of women in committee leadership positions.
Using DiD and RD methods, Anzia & Berry (2011) show that female members of the US Congress are more effective at bringing money to their districts than men are and also sponsor more legislation. They attribute this higher competence to selection effects: Sex-based discrimination and female self-selection (underestimating their own competence) lead to more qualified women being elected. Turning to the developing world, Brollo & Troiano (2016) use an RD design and random corruption audits in Brazil to show that female mayors are less corrupt than males but also less successful in attracting contributions and winning reelection. Others find null results: An RD study of 5,500 mayoral elections in the United States concludes that gender has no effect on policy outcomes such as size of government, composition of spending and employment, and crime (Ferreira & Gyourko 2014).
Gender does not operate in a vacuum-perceptions are important. Reputational effects might be especially important in international politics, though Koch & Fulton (2011) do not find an effect of female heads of state on interstate conflict in the modern period. Dube & Harish (2020), however, use the natural experiment of hereditary succession to show that queens were more likely than kings to fight wars in the distant past. The theory that a "critical mass" of women in a legislature is key to their influence might fall in the category of perceptions as well, though research to date has been inconclusive (Childs & Krook 2008).
In sum, female politicians appear to have different values and political styles, and they might have higher levels of competence as well. Extending these results across a variety of contexts would help determine whether they are universal or are specific to certain cultures and institutional arrangements.

Race, Ethnicity, and Religion
Relevant work on leaders' race, ethnicity, and religion has mostly asked whether politicians favor their own identity groups. Scholars typically highlight a values mechanism (ethnic altruism) and/or an interest mechanism (quid pro quo) whereby politicians favor their own group in order to stay in power. There is some tension between those who take a positive view of descriptive representation and those who have a more negative view of so-called ethnic favoritism. In the former case, scholars often look at legislators or lower-level officials. In the latter, studies of national leaders are more common.
A long line of research on minority representation has examined how black representatives in the United States represent black interests (Tate 2004). Recent work has taken steps to separate the effects of the representative's race from the constituency they represent by looking at blacks who represent white districts (Grose 2011) or matching on district characteristics (Kopkin 2017); black representatives typically behave differently even after these adjustments. Two articles using DiD and RD approaches to assess the impact of black mayors, however, find few overall effects except for greater hiring of black police officers (Hopkins & McCabe 2012) and increases in black employment (Nye et al. 2015). Several studies use audit designs to show that legislators are more likely to respond to requests from same-race letter writers (Butler & Broockman 2011 on the US; McClendon 2016 on South Africa). Broockman (2013) even finds that black representatives are more likely than whites to respond to same-race requesters outside of their district, an interesting demonstration of the values (or ethnic altruism) pathway, as there are few material benefits to be gained from representing nonconstituents.
India has been a key site for such research due to its system of randomly assigned reserved seats for scheduled castes and minorities. Pande (2003) demonstrates that reserved seats are associated with redistribution favoring the politician's group. Bhalotra et al. (2013) uses an RD design to examine close state legislative electoral contests between Muslim and non-Muslim candidates and shows that Muslim officials are linked to higher child survival rates and improvements in educational attainment even among non-Muslims.
Research using the lens of ethnic favoritism often has a regional focus on Africa because of its social diversity. The standard design is a DiD approach, where changes in the ethnicity of presidents or representatives provide causal leverage. The independent variable of interest is typically whether a district shares the politician's ethnicity. Studies in this mode include those of Franck & Rainer (2012) on primary education and infant mortality in 18 African countries, Burgess et al. (2015) on road building in Kenya, and Jablonski (2014) on foreign aid distribution, again in Kenya. Two papers use nighttime light intensity to proxy development and find that leaders favor their birthplace (Hodler & Raschky 2014) and their own ethnic group (De Luca et al. 2018). A dissenting study shows evidence of ethnic disfavoritism, with coethnic farmers facing higher taxes because they are more effectively monitored (Kasara 2007). Most often, ethnic favoritism explanations are based on an electoral (or interest) logic-effects are stronger in democracies and near elections-though scholars sometimes invoke values.
Turning to religion and elite behavior, most work examines US congresspersons (Burden 2007, Edwards Smith et al. 2010. Evangelical religious affiliation, not surprisingly, is typically associated with voting on issues favored by the religious right, but few of these studies have utilized stronger means of causal inference. Intriguingly, an RD study of Turkish mayors finds that the election of an Islamist increases female secular education, presumably because it makes state schools more welcoming to conservative families (Meyersson 2014).
In sum, there is considerable evidence that leaders favor their coethnics and coracials, though there is a concern that scholars only study the outcomes where these characteristics are most likely to matter (Kramon & Posner 2013). Mechanisms based on leader values (ethnic altruism) and interests (quid pro quo) predominate, but institutional factors such as democracy and election timing create conditional effects. One research frontier involves extending this work to international politics, though examining race, ethnicity, and religion in a cross-national context is admittedly more difficult than doing so in a single-country study.

Age
There has been surprisingly little work on age, probably the easiest biographical trait to measure. Age can affect mental acuity (leader competence) and time horizons (which might be classified as an interest mechanism). Most research on competence has been qualitative and focused on particular individuals (Garrow 2000), though one can imagine using text-analysis techniques to track declines more generally.
One area that has received some attention is leader age and conflict initiation. Some believe that younger, untested leaders are brasher and thus likely to initiate conflict. Horowitz et al. (2015), however, find that conflict risk increases with leader age, which they attribute to shortening time horizons and a desire to secure a legacy. Studies of legislatures, by contrast, find little evidence of value changes over a legislator's lifetime; the voting records of members of Congress show a high degree of continuity (Poole 2007). Note that evaluating the effect of age is tricky, since it is hard to distinguish age from experience as well as period and cohort effects. This is clearly an area where additional research could yield large gains.

Other Characteristics
The list of biographical characteristics that might affect elites' political behavior is lengthy. It is worth mentioning a handful of others that fall outside our article's main scope. In explaining why leaders fight, Horowitz et al. (2015) look at upbringing variables such as whether a leader's parents were divorced, whether leaders experienced childhood war trauma or poverty, and birth order effects. Only war trauma appears to have significant effects. Ramey et al. (2017) show that personality types affect many aspects of congressional behavior. Burden (2007) finds that specific traits such as personal tobacco use, having children in public schools, and membership in certain religious denominations affect the voting patterns of US congresspersons on policies related to those traits. Sexual identity may matter: Legislatures with a higher proportion of LGBT-identifying officials pass more gay-friendly laws (Reynolds 2013). Washington (2008), meanwhile, analyzes the effects of daughters on legislative behavior and finds strong effects across a range of voting outcomes. One can easily imagine other potential characteristics that might be analyzed, such as health (McDermott 2007), intelligence (Besley et al. 2017), and physical appearance (Todorov 2017).

METHODOLOGICAL CHALLENGES AND POTENTIAL SOLUTIONS
Most quantitative work linking leaders' biographical features to behavior or outcomes has used standard regression techniques. In this approach, political outcomes are modeled as a function of the characteristics of individual leaders (after including the standard control variables associated with each literature). A shortcoming of this approach is that it does not adequately account for endogeneity concerns. In this section, we outline three methodological challenges-the timing of leadership transitions, leader selection processes, and individual self-selection into background experiences-and highlight some potential solutions. These solutions are nested within the larger "credibility revolution" (Angrist & Pischke 2010) where design-based studies that search for natural experiments with "as if" randomization are increasingly popular.

The Timing of Leadership Transitions
One endogeneity concern is that structural forces-such as underlying political, economic, or social conditions in a country-determine when leadership transitions take place. If these conditions also affect the political outcomes under study, then any correlation between leaders and outcomes may be spurious. Jones & Olken's (2005) study was a watershed in the study of leaders because it provided a technique to address this problem. Specifically, they examined www.annualreviews.org • Studying Leaders and Elites as-if-random leadership transitions-cases where leaders died in office of natural causes or accidents-to investigate whether leaders matter for economic growth. In such cases, the timing of the transition from one leader to another is plausibly exogenous.
This approach hinges on a crucial assumption: The broader environment in which leaders operate is static before and after the unexpected leadership change. This may be a reasonable assumption in many cases, but one can identify at least some instances where it is problematic. An extreme case is Rwanda in 1994, when President Habyarimana's death in a plane crash not only caused a leadership transition but also unambiguously altered the broader environment any successor would inherit, since Habyarimana's death triggered genocide. Moreover, IR work on leadership turnover shows that new leaders are often "tested" by adversaries (Wolford 2007). Put differently, the general political environment may change precisely because there was a leadership transition. There is no easy fix for this issue, but rather than simply assume it away, scholars could start by explaining why this concern is minimal for the outcome under study. A distinction between policy choices (which leaders can control more easily) and policy outcomes may be a place to start.

Leader Selection
Even if the timing (the "when") of the leadership transition is exogenous, the selection of the next leader (the "who") likely is endogenous to political competition and the interests of powerful actors. Moreover, biographical traits often play a key role in leader selection, as evidenced by campaigns' nearly universal appeals to prior political, military, and business experiences, not to mention the visibility of a politician's gender, race, and ethnicity. Consequently, observed correlations between leader characteristics and political outcomes may be due to strategic leadership selection. For instance, if leaders with military experience are selected because potential conflicts loom on the horizon, then their experience may not actually be the cause of conflict.
A related selection effect, termed the "Jackie (or Jill) Robinson effect" (Anzia & Berry 2011), suggests that if particular types of individuals face internal or external barriers to entering politics, they might not be representative of their broader group. For example, the barriers women face may mean that female representatives are on average more competent than their male counterparts. This dynamic can even influence groups not directly implicated: Besley et al. (2017) found that the introduction of gender quotas in Sweden increased the quality of male politicians by removing "mediocre men." The standard experimental solution to the selection problem is unavailable because scholars cannot assign leaders with different traits to office, though researchers have had some success using audit studies in which they send requests to representatives from constituents who share and do not share their traits (Butler & Broockman 2011, McClendon 2016. Natural experiments with as-if-random assignment may be the next best way to assess the causal effect of certain types of leaders. The case that has been mined most intensively is India's system of randomly reserved seats for women and minorities (Chattopadhyay & Duflo 2004, Clots-Figueras 2011. Pushing this approach even further, Dube & Harish (2020) use monarchical rules of succession (where women only take power in the absence of a male heir) to investigate the relationship between gender and war.
One can imagine using other institutional rules for similar inferential purposes. Some that have begun to be used but have not been completely tapped include quotas, term limits, age restrictions, mandatory retirement rules, and salary changes (Carnes & Hansen 2016, Hughes et al. 2017, Dal Bó & Finan 2018, Hall 2019.
Given the rarity of genuine cases of random assignment, scholars have tried to engineer asif randomness with observational data. RD has become the preferred approach for this purpose.
The typical setup focuses on elections where a candidate with the biographical trait of interest narrowly wins or loses against a candidate without the trait. Researchers can thus assume that the selection of the particular politician was essentially random. The concern with this design is that it does not address the timing of leader transitions and that close elections are relatively rare events, especially at the level of national executives (Bertoli et al. 2019).
DiD approaches are another source of causal leverage. This method considers changes in leaders across different biographical traits and compares outcomes in similar cases without a change in leadership. Here the main limitation is the assumption of parallel trends across cases. To date, these methods have mainly been utilized to study the effects of race/ethnicity and gender, but they could plausibly be applied to other biographical traits. Looking ahead, scholars might use DiD and RD designs to compare those who were conscripted into fighting with those who were just too young to serve, or to investigate abrupt changes in the franchise that alter the character of elites. Finally, scholars may attempt to better model the process of leadership selection. There has been a boom of work on who runs for office and who wins (Lawless 2012, Hall 2019, which may offer theoretical insights that help account for selection biases.

Selection into Socializing Experiences
A final challenge is that future leaders typically self-select into background experiences. That is, individuals usually decide whether to join the military, pursue a particular kind of education, or work in a certain field. This makes it difficult to determine whether background experiences actually have a socialization effect on future leaders or whether individuals only select into experiences that are compatible with their preexisting values and beliefs. If the latter, then background experiences cannot be considered causal (though they could still help us predict leader behavior). Despite the pervasiveness of this problem, it has received surprisingly little attention in studies of leaders. Instead, it is common to simply assume that both self-selection and socialization are at play to some degree-but few studies attempt to figure out which is doing most of the work.
While self-selection is a daunting problem, there are steps scholars can take to minimize it. One would ideally find a natural experiment that as-if randomizes elites' biographical traits. Of course, such opportunities are extremely rare, but one example is Washington's (2008) study of how having daughters affects the voting records of members of Congress. The identification strategy assumes that the number of daughters is random (conditional on the total number of children). When such a clean research design is not feasible, scholars might follow the lead of Gift & Krcmaric (2017). They try to rule out the possibility that only future leaders who harbor democratic leanings in their youth will self-select into Western universities. The authors draw on a variety of sourcesranging from surveys of students studying abroad to the statements of foreign elites-to show that preexisting preferences for democracy do not necessarily drive the decision to study in the West. In our view, even imperfect solutions are preferable to ignoring the problem entirely.

REMAINING QUESTIONS AND THE PATH AHEAD
Despite the flurry of research on the personal biographies of leaders, several important issues deserve attention (in addition to the gaps already identified in the sections on empirical findings and methodology).
First, scholars should pay closer attention to the causal mechanisms that connect background experiences and traits to leader behavior. We have outlined four potential mechanisms through which personal biography might matter (Figure 1), but surprisingly few studies explicitly state the causal mechanism at work. Moreover, scholars rarely offer empirical tests that distinguish www.annualreviews.org • Studying Leaders and Elites between mechanisms (see Carnes 2013 for a positive example). Greater attentiveness to qualitative sources could be a useful remedy. In particular, we suggest that scholars look at leaders' personal accounts-memoirs, diaries, speeches, notes from meetings, and the like-to investigate whether they considered their background experiences relevant to their governing behavior or general worldviews. We should take such evidence with caution if there is reason to think leaders have incentives to misrepresent, but this approach can serve as a plausibility check: If quantitative evidence suggests a particular factor is important but key leaders themselves never mention it, there may be reason to doubt the hypothesized relationship between the trait and the outcome variable.
Second, several conceptualization and data questions have come to the fore. The move toward more nuanced measures of biographical traits is a welcome development, but it also raises issues of interpretation. Indeed, we are now at the point where the same background experience can have multiple meanings across different studies. Consider a hypothetical leader with a graduate degree in economics from an American university. What does that tell us about the leader? Some would focus on the transmission of neoliberal beliefs (Nelson 2017), others on the socializing effect of studying in a Western democracy (Gift & Krcmaric 2017), others on technocratic experience (Flores et al. 2013), and still others on general competence (Besley et al. 2011). Which measure is best? How would we know? Or can they all be insightful simultaneously?
The explosion of measures has also ignored interactions between traits, not to mention intersectionality. Baturo's (2017) data on heads of state show that there is considerable overlap between variables such as class, military experience, business careers, and education, among others. The next generation of scholars will need to find clever ways to tease out the effects of each one. Additionally, a number of biographical traits have been neglected in certain research agendas and could be the focus of more work. Ascriptive characteristics, such as race and gender, are typically missing from IR research on heads of state, while AP and CP work on legislators tends to neglect socializing experiences like education and military service. Age and political experience have been understudied in all types of work. The move toward more systematic data sets-such as the LEAD data set on national executives (Horowitz et al. 2015) and the Global Leadership Project on a broad range of elites (Gerring et al. 2019)-will be helpful in bridging these gaps.
Third, what are the policy and normative implications of these findings? While questions of descriptive representation, not to mention policies such as quotas, traditionally focused on ascriptive traits like gender and race, the results summarized here suggest that citizens should pay closer attention to socializing traits like occupation, education, and military service, which also have large effects on the behavior of elites and thus representation. Burden (2007) further suggests that the mechanisms of influence of these traits should affect our assessments of them. One might privilege competence mechanisms over interest ones, with values occupying a more ambivalent place. "Small-d" democrats might even question whether we should welcome these effects at all, since they imply that the actions of elites may diverge from the preferences of the mass public.
Fourth, it may be time to return to the structure/agency issue that initially motivated much of the research on leaders. To what extent and under what conditions can individual leaders overcome structural forces that constrain their independent influence on outcomes? Recent work on personal biography often ignores such questions. In fact, one of the more sophisticated empirical strategies used in the literature-exploiting as-if-random leadership transitions-explicitly tries to hold structure constant to test whether leaders have any marginal effect. This approach is admirable because it mitigates some of the inferential challenges associated with studying leaders, but this research design also makes it impossible to compare the extent to which leaders matter relative to the structural constraints in which they operate.
Future research might explore the domains in which leader-specific attributes generate meaningful effects (Burden 2007, Baturo & Elkink 2014. For instance, do national executives or lowerlevel political elites generally have more freedom of action? This is not only a question about which types of leaders are most hemmed in by structural forces but also a distinction between the outcomes that different subfields of political science tend to study. Much of the work on international politics focuses on low-probability/high-consequence outcomes such as war, whereas much of the work in domestic politics examines high-probability/low-consequence outcomes such as bill proposals or roll call votes. It may not be reasonable for a particular background experience to matter in the same way, if at all, in both contexts. The accumulation of studies using the personal biography approach may soon give us enough results to draw such context-dependent conclusions as well as move toward more general theories that integrate structure and agency.

DISCLOSURE STATEMENT
The authors are not aware of any affiliations, memberships, funding, or financial holdings that might be perceived as affecting the objectivity of this review.