4.1. Data and measurements

To test the main hypotheses, we analyze data covering UN agencies across—including funds and programs, specialized agencies, other entities and related organizations—from 1996 to 2022. Due to data availability, our sample includes 25 UN agencies.Footnote ⁴ Our dataset captures each agency’s staff composition by nationality, senior officials’ nationalities, country-specific voluntary contributions, and key economic and political characteristics of contributing nations.

First of all, we sourced UN staffing data from the United Nations Chief Executives Board for Coordination (CEB) Personnel Statistics reports for the years 1996–2022.Footnote ⁵ These annual reports provide detailed staffing information across all UN agencies. The data covers staff employed for durations of one year or longer. Although our hypotheses focus on international professional (P) staff, our empirical analysis also includes general service (GS) staff. International professional staff are globally recruited experts with specialized skills, often requiring advanced degrees, and their numbers reflect the representation and influence of their home countries. In contrast, general service staff, primarily responsible for support and administrative tasks, are typically hired locally. It is important to note that the data excludes consultants, who comprise a substantial portion of the UN workforce and often serve on short-term contracts (Seabrooke and Sending, Reference Seabrooke and Sending2020), but are not consistently reported in the official data.Footnote ⁶

We use Ln(P_staff) and Ln(GS_staff) to represent the logarithmic forms of the respective staff categories.Footnote ⁷ Given the nature of these categories, we are more interested in the former, as professional staff play a more significant role in organizations and are therefore more likely to represent the interests of their home countries. Let P_staff_increment (GS_staff_increment) denote the change of a country’s staff in an organization, i.e., the difference in Ln(P_staff) (Ln(G_staff)) between periods t and t − 1. For simplicity, we omit detailed subscripts for country, organization, and year when it does not cause confusion. Notice that P_staff_increment (GS_staff_increment), as a primary outcome variable, approximates the growth rate of the professional staff (general service staff).

Our second data source focuses on IO leadership. We compiled data on top officials from the UN Yearbook, which provides the names and titles of senior staff. We determined their nationalities by consulting directories, media reports, and historical documents. Due to inconsistencies in the UN Yearbook’s coverage of organizations and deputy positions, we conducted extensive searches across media sources and historical records to fill any gaps in the data. Moreover, as the UN Yearbook ceased publication after 2017, we supplemented the dataset with information for the years 2018–2022 through additional searches. For leadership roles, we consider both executive leaders of the organization and their deputies.Footnote ⁸ Our key explanatory variable, leadership, is binary: it equals 1 if a country has one or more than one citizen serving as a leader (either as head or deputy head) in the organization, and 0 otherwise.

For voluntary contributions from countries to organizations, we compiled UN financial data from the biannual or multiannual “Budgetary and Financial Situation of the Organizations of the United Nations System” reports. These reports provide detailed information on the annual voluntary contributions from each country to each organization.Footnote ⁹ The voluntary contributions, measured in millions of dollars, are transformed using a logarithmic scale, resulting in the variable Ln(contributions).

In our empirical analysis, we also control for the socioeconomic and political conditions of countries. We use Ln(working_population) to denote the logarithm of the working-age population (individuals aged 15–64), as defined by the OECD,Footnote ¹⁰ and Ln(GDP) to represent the logarithm of GDP. These two variables are sourced from the World Bank’s WDI database. We use the variable education, drawn from the UNDP’s HDI dataset, to capture the average years of schooling. Higher education levels indicate a greater potential supply of citizens qualified to work in IOs. In addition, we use economic_globalization and political_globalization to capture countries’ overseas interests, sourced from the KOF Globalization Index. The variable economic_globalization measures the size of overseas business activities, reflecting a country’s involvement in international trade and investment, while political_globalization captures countries’ participation in foreign affairs, including appointing diplomatic officers, engaging in peacekeeping missions, and cooperating with international NGOs. To capture the domestic political conditions of member countries, we include two additional variables from the V-Dem project dataset. The variable corruption measures the level of domestic political corruption in the public sector, and the variable VDem_index provides a measure of liberal democracy for each country.Footnote ¹¹

We summarize the descriptive statistics for the main variables in Table 1. Between 1997 and 2022, the average staff increment (P_staff_increment) was about 0.023, with a standard deviation of 0.251. While the number of staff remained relatively stable over time, there was considerable variation in growth rates across countries. The average leadership score is 0.014, indicating that only a small proportion of countries have nationals appointed as leaders in the organizations in our sample.

Table 1.

Descriptive statistics

Figure 1 displays the annual number of chief executive leadership positions in our sampled IOs held by nationals from the five permanent members of the UN Security Council, along with Germany and Japan.Footnote ¹² Russian nationals did not hold any such positions during the study period and therefore excluded. The figure shows that advanced economies such as the United States, France, and Japan tend to occupy more leadership roles, although the numbers fluctuate over time. In contrast, developing countries such as China have seen a notable increase in leadership representation since 2005. This trend is echoed in Figure 2, which presents the yearly count of top leaders by national income level. While high-income countries have consistently dominated IO executive leadership, there has been a steady rise in leaders from upper-middle-income countries in recent years, suggesting a gradual shift toward a more inclusive and representative global leadership landscape.

Figure 1.

Number of chief executive leaders from the P5 countries, Germany, and Japan.

Notes: Since Russia does not have any chief executive leaders in our working sample, it is not shown in the figure.

Figure 2.

Number of chief executive leaders in office by country income level.

Figure 3 illustrates how staff representation varies across countries based on their leadership role in IOs over time. We categorize countries into three groups: those with current UN agency leaders, those with former leaders only, and those that have never had leaders. The figure displays the total number of professional staff for each category. The three curves—ranked from top to bottom—represent the total staff for the “current leaders,” “former leaders,” and “no leaders” groups, respectively. These curves differ not only in absolute numbers but also in their slopes, reflecting varying rates of annual staff increment.

Figure 3.

Representation of professional staffs of countries with different leadership Status.

Professional staff numbers increase most rapidly in countries with current leaders. Countries with former leaders show moderate growth, while those without leaders maintain nearly flat rate. Having a current leader in an international organization appears to significantly boosts a country’s professional representation, suggesting leadership roles create enhanced career opportunities for nationals and advance the country’s organizational policy agenda. In the following sections, we will verify this patterns observed in Figure 3 through systematic data analysis.

4.2. Empirical strategy

In the baseline analysis, we use a two-way fixed-effects model with a DiD approach to estimate the impact of leadership on the growth of countries’ staff representation in UN agencies. The regression model is specified as follows:

(1)\begin{equation}\begin{array}{*{20}{c}} {staff\_incremen{t_{c,i,t}} = \beta \cdot leadershi{p_{c,i,t - 1}} + {X_{c,i,t - 1}} \cdot \Gamma + {\lambda _{c,i}} + {\mu _t} + {\varepsilon _{c,i,t}},\ } \end{array}\end{equation}

where the $c$ represents the country, $i$ the international organization, and $t$ the year. The variable $staff\_incremen{t_{c,i,t}}$ refers to the change in either professional staff (P_staff_increment) or general-service staff (GS_staff_increment) for country $c$ in year t within organization $i$, which is our outcome of interest. $leadershi{p_{c,i,t - 1}}$ indicates whether country c had at least one national serving as a senior official in organization $i$ during year $t - 1$. The staff increment is measured as the difference between year t and year t − 1, so the outcome variable has a one-year lag relative to the explanatory variable $leadershi{p_{c,i,t - 1}}$.

${X_{c,i,t - 1}}$ is a vector of control variables, including the logarithms of the country’s staff numbers in each organization for year t − 1, the log form of the working-age population, GDP, education, economic globalization, political globalization, and indicators of domestic political corruption and democracy, all for year t − 1.

${\lambda _{c,i}}$ represents dyadic (IO-country pair) fixed effects, controlling for time-invariant factors specific to the relationship between country $c$ and organization $i$, such as the advantages of founding member states. ${\mu _t}$ represents year fixed effects, controlling for common time trends affecting all dyads, such as policy shocks. Including these fixed effects ensures the analysis focuses on within-dyad variations over time while accounting for overall time-based changes. ${\varepsilon _{c,i,t}}$ is the error term. To address potential correlations in the error terms, we cluster standard errors by dyad. The coefficient of interest is $\beta $, which captures the average treatment effect of leadership on staff representation dynamics.

4.3. Baseline results

We report the baseline results from estimating Equation (1) in Table 2. In all four regressions, we control for country-IO and year fixed effects. Columns (1) and (3) show the results of the benchmark regressions, controlling only for staff size. Columns (2) and (4) additionally control for domestic conditions at the country level.

Table 2.

The impact of leadership on the increase of staff representation in IOs

*** Standard errors (in parentheses) are clustered at the dyadic (i.e., country-IO) level: *** p < 0.01, ** p < 0.05, * p < 0.1. All the explanatory variables are lagged by 1 year.

Overall, the results indicate that leadership in IOs significantly improves the representation of professional staff from the home country, but not general service staff. Specifically, the coefficient for leadership is 0.041 in column 1, significant at the 1% level, and changes slightly to 0.038 in column 2,Footnote ¹³ also significant at the 1% level. Our regression results suggest that leadership provides a significant advantage for a country in securing professional positions within IOs, supporting Hypothesis 1. To interpret the estimated coefficients, consider a hypothetical example in which a country has 100 staff members in an organization at period t. According to the estimate in column 2, the country’s staff in the IO at period t + 1 increases to approximately 104 due to its leadership position at period t, holding other factors constant.Footnote ¹⁴ If the leadership continues for five years, the country would gain a total of 20 additional staff members, representing a 20% increase. For general service staff, columns 3 and 4 show no significant relationship with leadership, as coefficients are not significantly different from zero. This pattern likely reflects two factors: the distinct nature of general service recruitment, where local hiring practices often play a key role in the selection process, and the focus of these positions on operational and administrative functions rather than policy-making activities.

4.4. Robustness checks and event-study estimates

The validity of two-way fixed effects estimation depends on the assumption of homogeneous treatment effect. To account for potential heterogeneous effects, we employ several alternative estimation strategies. Column 1 of Table 3 shows the result of the two-stage DiD estimation (Gardner, Reference Gardner2022). Column 2 displays the result of the stacked DiD model (Cengiz et al., Reference Cengiz, Dube, Lindner and Zipperer2019). Table 3 shows that the leadership effects remain positive and statistically significant across different heterogeneity-robust estimations. These robustness checks further support our hypothesis that leaders in IOs shape the dynamics of staff representation.

Table 3.

Additional robustness checks

*** Standard errors (in parentheses) are clustered at the dyadic (i.e., country-IO) level: ***p < 0.01, **p < 0.05, *p < 0.1. The variable leadership and other control variables are lagged by 1 year.

The validity of DiD estimates relies on the parallel-trend assumption. We present the event-study figures generated by several recently proposed estimators (Cengiz et al., Reference Cengiz, Dube, Lindner and Zipperer2019; Gardner, Reference Gardner2022) to check for potential pre-trends and examine the dynamic effects of leadership in IOs. Specifically, Figure 4 presents the event-study estimates with a 95% confidence interval, using the period −1 as the reference group. Regardless of the estimation methods used, the coefficients before the appointment of a leader (i.e., period −1) are close to zero and statistically insignificant. This suggests that there is no substantial pre-treatment trend, thereby confirming the parallel-trend assumption. Additionally, the coefficients turn positive shortly after leaders assume their positions for the first time, with some estimations showing a decrease in the return on representation after period 0.

Figure 4.

Event study: dynamic effects of leadership.

Note: The plots show the estimated dynamic effects of leadership for each year, along with 95% confidence intervals.

4.5. Mechanisms

In the following, we empirically examine two mechanisms through which leadership influences staff representation, testing Hypotheses 2 and 3. The first mechanism involves the relationship between leadership and organizational funding.

Since the 1980s, the increasing proportion of earmarked voluntary contributions has politicized UN funding, giving major donors greater control over programs (Weiss, Reference Weiss1982). This shift from collective core financing to individual restricted voluntary funds has led to what E. R. Graham (Reference Graham2015) terms the “bilateralization” of the UN. Contributing countries often gain the ability to provide personnel for funded programs (Gui, Reference Gui2024).

Leadership positions enable control over IO resources. Xu and Weller (Reference Xu and Weller2018) document how leadership candidates, supported by their governments, often promise member states increased aid in exchange for votes—a practice known as bringing a “dowry.” This “dowry” extends beyond bilateral aid to include development programs within IOs. Organizations, in turn, may favor leadership candidates capable of attracting substantial funding, creating a self-reinforcing cycle: leaders use their positions to encourage voluntary contributions from their home countries, which then strengthens their influence over staffing decisions to benefit their nationals.

Figure 5 illustrates the relationship between leadership roles and voluntary contributions to organizations. Similar to Figure 3, we compare voluntary contributions across three categories: countries with current leaders, those with former leaders only, and those without any leaders. As shown in Figure 5, countries with current leaders exhibit a marked upward trend in voluntary contributions, particularly after 2015, when contributions increased substantially. In contrast, countries with only former leaders or no leaders demonstrate limited growth in voluntary contributions. Though countries with former leaders maintain slightly higher average contributions than those without leaders, both curves remain relatively flat throughout the observed periods.

Figure 5.

Voluntary contributions by countries with different leadership Statuses.

Note: For each year, we calculate the average number of contributions across IO–country pairs within each of the three categories.

Table 4 presents the results of an empirical test of Hypothesis 2 regarding the relationship between leadership, voluntary contributions, and staff representation. Column (1) suggests that IOs receive significantly higher voluntary contributions from countries with leadership positions compared to other countries. Furthermore, column (2) shows that these increased voluntary contributions can further drive higher number of professional staff from the leaders’ home countries, while leadership maintains its direct impact on staff representation.

Table 4.

Leadership impact through bringing voluntary contributions

*** Standard errors (in parentheses) are clustered at the dyadic (i.e., country-IO) level: ***p < 0.01, **p < 0.05, *p < 0.1. All the explanatory variables are lagged by 1 year.

To test Hypothesis 3, which suggests that IO leaders promote cooperative interactions with their home countries to increase national representation, we construct proxy variables to capture interactions between IOs and countries using the GDELT dataset. This database records events reported in international news media since 1979 and can be used in studies of bilateral relations and diplomatic behavior. Based on original news sources, GDELT provides detailed attributes for each event, including the type of interaction, the parties involved, and event sentiment.Footnote ¹⁵

In the GDELT data, news reports that mention both an IO and a country are categorized into four types: verbal cooperation, material cooperation, verbal conflict, and material conflict. These categories capture different aspects of IO–state relationships. Since Hypothesis 3 centers on mechanisms that operate through positive interactions, we focus on those with cooperative characteristics, namely verbal and material cooperation. Material cooperation refers to tangible, resource-based collaboration. For instance, Germany’s financial contribution of EUR 1.3 million to the UNDP-supported cybersecurity project in Bosnia and Herzegovina represents such material cooperation, where both sides engage in a jointly implemented development program. This event is coded as material cooperation in the data (GDELT Event ID: 1076305628). Verbal cooperation includes public statements of support or mutual appreciation. For example, UNHCR publicly expressed gratitude to Italy for its long-standing commitment to receiving refugees from Libya via humanitarian corridors and for providing continuous financial assistance for Rohingya refugees in Bangladesh, supporting essential services such as health, education, and protection. This event is coded as verbal cooperation in the data (GDELT Event IDs: 1052494651; 1076129855). These cases demonstrate how cooperative interactions can range from, but not limited to, financial commitments to expressions of diplomatic goodwill, both contributing to the deepening of IO-state relationships. Of the 25 organizations in our working sample, 15 are involved in the above two types of cooperative interactions with countries: FAO, IAEA, ICAO, ILO, IMO, IOM, ITU, UNDP, UNESCO, UNHCR, UNICEF, UNWTO, UPU, WFP, and WHO.

Following the approaches of Amarasinghe (Reference Amarasinghe2022) and Chiba and Gleditsch (Reference Chiba and Gleditsch2017), we use the number of verbal and material cooperation events to construct two new variables. The first variable, cooperative_interaction1, is the logarithm of the total number of cooperation events—both verbal and material—between an IO and a country in a given year, capturing the frequency of their cooperative interaction. However, this measure does not account for the intensity or consequences of these interactions. To address this limitation, we construct a second variable, cooperative_interaction2, which is based on the Goldstein scores assigned to cooperative events between an IO and a country in the same year. The Goldstein scores reflect the potential significance of these events for the countries involved. We sum the Goldstein scores of all relevant events and then take the logarithm of the total.

It is important to note that both variables are based on media coverage and therefore serve only as approximations of actual IO and country interactions. While not perfect measures, they nonetheless capture certain cooperative dimensions of these relationships.

Columns 1 and 2 in Table 5 show that leadership in IOs enhances staff representation from their home countries through increased cooperative interactions. When examining these positive interactions more closely in columns 3 and 4, we find that leadership-driven engagement significantly expands the professional staff from leaders’ home countries within these organizations. These findings, especially those in columns 1 and 2, are consistent with H3, showing that once leadership positions are secured, they help foster productive interactions between the IO and the home country. These interactions, in turn, contribute to greater staff representation for the member state. Our hypotheses regarding the mechanisms do not take a position on whether leadership has a direct effect beyond the channels of voluntary contributions and IO–state interaction. Columns 3 and 4 suggest that, in addition to these channels, leadership itself appears to have a positive effect, though this effect is not statistically significant in the reduced-form regression.

Table 5.

Leadership impact through inducing interactions with home countries

*** Standard errors (in parentheses) are clustered at the dyadic (i.e., country-IO) level: ***p < 0.01, **p < 0.05, *p < 0.1. All the explanatory variables are lagged by 1 year.

4.6. Heterogeneity

To provide additional evidence, we examine the potentially heterogeneous impacts of leadership across different subsamples. Funding mechanisms significantly shape how influence is exercised within IOs, with research by Bayram and Graham (Reference Bayram and Graham2017) and Graham (2017) demonstrating that wealthy donors can substantially impact decision-making and program implementation. Therefore, the stability and diversity of funding sources are crucial for organizational independence and operational effectiveness.

We expect that IOs with more stable funding sources are more likely to maintain institutional independence. To test this, we divide the IOs into two subsamples based on the share of assessed contributions in their total budget (i.e., the sum of assessed and voluntary contributions). About half of the organizations in our sample have a share above 40%, so we use this threshold to split the sample into those with a higher (above 40%) and lower (40% or below) reliance on assessed contributions. Organizations with a relatively lower share of assessed contributions rely more heavily on voluntary contributions, making staff recruitment more vulnerable to external influence. In contrast, those with a higher share of assessed contributions are likely to exercise greater autonomy in hiring. Table 6 presents the regression results for the two subsamples. The findings indicate that leadership has a stronger marginal effect on professional staff growth in IOs with less stable (more voluntary) funding, although the difference between the two coefficients is not statistically significant.

Table 6.

Heterogeneity of organizations’ funding sources

*** Standard errors (in parentheses) are clustered at the dyadic (i.e., country-IO) level: *** p < 0.01, **p < 0.05, *p < 0.1. The variable leadership and other control variables are lagged by 1 year.

The second type of heterogeneity we explore concerns the differential roles of developed and developing countries in IOs. Historically, developed countries have been disproportionately represented in both rank-and-file and senior positions within IOs, a pattern criticized for being detrimental to such organizations’ legitimacy and functional effectiveness (Kleine, Reference Kleine2013; Stone, Reference Stone2013). In recent years, there has been growing advocacy for equitable representation to enhance IO legitimacy and protect the low-income countries’ rights.Footnote ¹⁶ The UN exemplifies this shift through its initiatives to attract talent from underrepresented states as well as targeted recruitment efforts in regions like Sub-Saharan Africa (Gui, Reference Gui2024). These efforts have yielded results, with low-income countries achieving substantially improved representation in IOs (Parízek, Reference Parízek2017; Parízek and Stephen, Reference Parízek and Stephen2021a).

To explore how leadership affects staff representation across different country groups, we divide our sample into developed and developing countries. Following World Bank classifications,Footnote ¹⁷ we categorize high-income countries as developed and all others into developing. To better understand temporal variations in these leadership dynamics, we analyze two distinct periods: 1996–2010 and post-2010.

The heterogeneity results are shown in Table 7. Columns 1 and 4 present the results for the entire period (1996–2022) for developed and developing countries, respectively. The findings reveal divergent patterns between country groups. For developed countries (Columns 1–3), leaders have historically succeeded in expanding professional staff from their home countries, though this effect weakens and becomes less significant after 2010. In contrast, developing countries demonstrate an increasing trend in leadership impact on representation. Notably, the marginal impact of leadership becomes statistically significant in the post-2010 period, highlighting the growing influence of developing countries in shaping global governance.Footnote ¹⁸

Table 7.

Heterogeneity of developed and developing countries’ influence

*** Standard errors (in parentheses) are clustered at the dyadic (i.e., country-IO) level: ***p < 0.01, **p < 0.05, *p < 0.1. The variable leadership and other control variables are lagged by 1 year.

4.7. Spillover effect of leadership

Amid growing calls for more equitable representation reforms, the traditional influence of great powers has been challenged by rising developing countries. As shown in Table 7, there appears to be a decreasing trend in the role of developed countries. Yet concerns remain that some donors continue to shape staffing outcomes through more indirect means. For instance, as discussed earlier, donor countries have broadened staff sponsorship programs to include candidates from partner countries in the Global South—a move that enhances legitimacy on the surface but may also serve to consolidate influence through transnational patronage. In this section, we empirically examine these practices and test the fourth hypothesis: a country tends to gain greater professional staff representation in IOs when its connected countries hold leadership positions.

We first measure diplomatic connections based on UNGA voting records (Voeten et al., Reference Voeten, Strezhnev and Bailey2009). Specifically, we construct a variable leadership_ally, which capture whether a country’s diplomatically connected partners have nationals serving as leaders in IOs. We aim to investigate whether the leadership of such partners enhances a country’s staff representation, even when the country itself lacks nationals in leadership roles. Specifically, we define the variables as follows. For a country-IO dyad at period t, $leadership\_connectio{n_{c,i,t}}$ is defined as

(2)\begin{equation}\begin{array}{*{20}{c}} {leadership\_connectio{n_{c,i,t}} = \mathop \sum \limits_n connectio{n_{c,n}} \times leadershi{p_{n,i,t}},\ } \end{array}\end{equation}

where $connectio{n_{c,n}}$ captures the diplomatic closeness between country c and country n based on UNGA voting data. We use the Affinity Score in 2010 from the World Economics and Politics Dataverse (B. A. T. Graham and Tucker, Reference Graham and Tucker2019)Footnote ¹⁹ as a measure of diplomatic closeness between two countries. Since the score ranges from −1 to 1, we apply a monotonic linear transformation by subtracting the minimum value and dividing by the range, so that the rescaled values fall between 0 and 1 is the key explanatory variable in the benchmark analysis, a binary variable indicating whether country n has at least one leader in IO i at time t. Therefore, $leadership\_connectio{n_{c,i,t}}$ captures the implicit stake a country has in an organization through its diplomatically connected partners.

The first column of Table 8 shows a positive association between a country’s professional staff representation and the presence of its connected states in IO leadership. Column 2 further indicates that this relationship is particularly evident among developing countries.Footnote ²⁰ That is, when an diplomatically connected partner of a developing country holds an influential position within an IO, the developing country tends to have greater professional staff representation. This pattern is consistent with our theoretical expectation that IO leadership can generate spillover effects that reflect broader interstate competition within IOs.

Table 8.

spillover effect of connected countries’ leadership

*** Standard errors (in parentheses) are clustered at the dyadic (i.e., country-IO) level: ***p < 0.01, **p < 0.05, *p < 0.1. All the explanatory variables are lagged by 1 year.

An alternative explanation, however, is that IO leaders may favor individuals who share their language or cultural background, facilitating closer and more efficient working relationships. While we do not dismiss this possibility, we maintain that our core argument regarding strategic behavior among allied states remains valid regardless of whether cultural or linguistic affinity plays a role. To empirically examine this contention, we construct a new variable, $leadership\_connectio{n_{c,i,t}}\left( {different\_language} \right)$, which captures the degree of leadership held by a country’s non–same-language connections, as measured by the intensity of their relational ties. Specifically, language similarity is measured using data from the CEPII database (Conte et al., Reference Conte, Cotterlaz and Mayer2022), where two countries are coded as having a common language if at least 9% of the population in both countries speak the same language. Accordingly, we modify the original leadership variable $leadershi{p_{n,i,t}}$ in Equation (2) such that it equals 1 only when country n holds a leadership position in IO i at time t, and country c and country n do not share a common language. The underlying intuition behind using this new explanatory variable is straightforward: if IO leaders recruit staff from allied countries speaking different languages, the behavior is unlikely to be driven purely by convenience or cultural affinity. Columns 3 and 4 of Table 8 support this argument and provide additional evidence for our hypothesis.

Notice that geographic proximity likely plays an important role in shaping connections among countries. As an alternative measure of country connections, we use membership in regional organizations. Specifically, we consider five regional organizations: the European Union (EU), the African Union (AU), the Organization of American States (OAS), the Commonwealth of Independent States (CIS), and the Association of Southeast Asian Nations (ASEAN). Countries belonging to the same regional organization are treated as connected. We exclude the United Kingdom from the EU and Cuba, Nicaragua, and Venezuela from the OAS, as these countries have either withdrawn or abandoned their membership. Countries not belonging to any of the five organizations are excluded from the analysis. Using a method similar to Equation (2), we construct a new variable, which captures the extent to which a country’s regional partners have nationals serving as leaders in IOs.

The regression results are reported in columns 5 and 6 of Table 8. They show that when regional partners hold leadership positions in IOs, countries in the same regional organization experience greater staff representation, even if they themselves have no nationals in leadership roles. This effect is statistically significant for both the full sample of countries across the five regional organizations and the subsample of developing countries.

These results suggest an evolution in how influence is exercised in IOs. Rather than maintaining direct control through their own nationals, leadership-holding countries appeared to be adopting more nuanced strategies by leveraging influence through their networks of relational ties. This pattern indicates a potential shift from overt to more subtle mechanisms of institutional influence in global governance structures. Therefore, the increased staff representation of some developing countries, facilitated by their connected partner countries, should not be interpreted as evidence of a broader rise in influence across all the developing countries. In other words, our findings do not rule out the possibility that many still face institutional barriers and path-dependent processes that favor established powers (Parízek and Stephen, Reference Parízek and Stephen2021b; Lam and Fung, Reference Lam and Fung2024).

4.8. Impact of the national distribution of IO staff on organizational performance

What are the normative implications of IO staff politicization? In this section, we offer a preliminary analysis of how the national distribution of IO staff affects organizational performance. We draw on data from the Performance of International Institutions Project (PIIP), which compiles performance assessments of 54 major IOs conducted by five national governments (Australia, Denmark, the Netherlands, Sweden, and the United States) and the Multilateral Organization Performance Assessment Network (MOPAN). Lall (Reference Lall2023) constructs composite scores of organizational performance for a set of IOs using these sources. We adopt performance index of Lall (Reference Lall2023) as our primary measure of organizational performance, which we denote as IO_performance.

Our key explanatory variable is the Herfindahl index of staff nationality concentration, calculated separately for P-level and G-level employees in each organization within our working sample. The Herfindahl index is defined as the sum of squared shares of staff by nationality; a higher value indicates a more homogeneous distribution of staff nationalities. We match these concentration indices with the PIIP performance data. We successfully match performance indicators for 14 IOs in our working sample. These include: FAO, IFAD, ILO, IOM, ITC, UNDP, UNESCO, UNFPA, UNHCR, UNICEF, UNIDO, UNRWA, WFP, and WHO.

In our regression analysis, we include a set of control variables as specified in the PIIP codebook: (1) Ln(IO_income): the natural logarithm of organizational income (Logged income in millions of US dollars); (2) Ln(IO_expenditures): the natural logarithm of organizational expenditures (Logged expenditures in millions of US dollars); (3) Ln(IO_age): the natural logarithm of organizational age; (4) Policy_autonomy: an index measuring the organization’s capacity for agenda-setting, avoidance of state vetoes, and access to non-governmental funding. This variable is based on survey data from senior IO officials compiled by Lall (Reference Lall2023) (De facto policy autonomy); (5) Member_discrepancy: the standard deviation of member states’ foreign policy ideal points, reflecting internal heterogeneity (Variance of member states’ foreign policy “ideal points”). As is shown in Table 9, the staff concentration is negatively associated with the performance of IO. Given data limitations, the number of observations in the regressions reported in Table 9 is scarce. We therefore regard this result as only suggestive.

Table 9.

Staff concentration and IO performance

*** Standard errors (in parentheses) are clustered at the IO level: ***p < 0.01, **p < 0.05, *p < 0.1. All the explanatory variables are lagged by 1 year.