Hypotheses

Turning to operationalization, we focus on formal international organizations for the empirical analysis. This is for several reasons. First, the universe of international organizations is well defined, and we can conduct a systematic analysis of all organizations in a way not possible for the universe of international institutions. Second, while the universe of institutions includes unwritten norms and principles for which the boundaries of participation are ambiguous, all international organizations are associated with at least some published language and well-defined memberships. It is thus possible to examine the presence of both racist language and patterns of racial bias in membership among a comparable set of institutions. In the robustness checks, we consider a wider set of institutions including informal institutions, which provides some leverage over causal mechanisms.

One premise of institutional racism is that overt expressions of racism have become less common over time. Although subject to contestation, a taboo against outright racism has become firmly entrenched across societies since the end of World War II, albeit to different degrees and not necessarily at the same time. The normative shift against open racist expression should affect language use by international organizations. Member states are unlikely to advocate for racist language that is delegitimized or illegal within their own societies. States are also more likely to support the inclusion of antiracist language if such principles pose no conflict with their own laws and domestic practices. Furthermore, at the international level, racial diversity regimes have evolved over time from outright racist discrimination to transformative efforts reflecting decolonization and civil rights movements.Footnote ³⁵ These normative shifts should make it more likely for international organizations to eschew racism in favor of antiracist expressions in their use of language. This leads to our first hypothesis:

For the purposes of operationalizing nonovert, institutional racism, we focus on patterns of membership according to white-majority status. Olson describes “whiteness” as both “an interest in and an expectation of favored treatment within a color-blind society.”Footnote ³⁶ In Bhopal’s study of white privilege, she refers to “white spaces” as places where “whiteness and white Western practices are the norm and those which do not comply with these are seen as outsiders and others.”Footnote ³⁷ Institutional racism is associated with segregation through seemingly race-neutral policies and rules that carve out or reinforce such “white spaces.”

In a similar vein, we examine the degree of domination of international organizations by white-majority states to the relative exclusion of others. Our focus on membership is motivated by both substantive importance and empirical tractability. Segregation and disparate access to key societal institutions—for example, legislatures, courts, banks, civic associations, educational institutions—is one of the most basic manifestations of institutionalized racism.Footnote ³⁸ Much like these domestic institutions, international organizations play an important role in setting policy, marking status, and mediating access to resources in the international system. Furthermore, unlike other manifestations of institutional racism—such as racially biased policy implementation or internal practices—membership is readily observable and subject to direct comparison across a diverse range of international organizations.

We evaluate two observable implications of institutional racism on membership patterns. First, if institutional racism remains persistent, this should manifest in the overrepresentation of white-majority states in international organizations. Exclusion from structures of power and influence is the most visible and among the most pernicious manifestations of institutional racism. Thus we predict:

Second, we predict a pattern of segregation in which membership expansion for white-dominated international organizations is biased in favor of white-majority countries. This reflects two plausible mechanisms. First, membership requests from white states may be prioritized or be subject to less scrutiny at the accession stage by current white members. This is the international analogue of how white-dominated workplaces and career networks advantage white job seekers at the expense of minority applicants.Footnote ³⁹ Second, nonwhite countries may eschew organizations dominated by the prerogatives and normative priorities of white-majority countries. In the domestic context, perceptions of hostility and the absence of role models can lead racialized minorities to forgo participation in white-dominated groups and organizations.Footnote ⁴⁰ Analogously, institutional membership comes with both benefits and obligations. If countries foresee being marginalized or mistreated upon joining a white-dominated organization, the expected utility and thus incentive to seek membership will diminish. We thus predict:

We test an analogous proposition for the cessation of membership. Unlike accession, exit decisions involve states that are existing members of an organization. Hence, exits are more likely to stem from conflicts over operations and implementation of organizational rules. We predict that white-dominated organizations are relatively more welcoming and accommodating of white-majority country interests, reducing the likelihood of irreconcilable differences and conflicts. Conversely, nonwhite countries are more likely to become dissatisfied and disgruntled with the status quo, which can lead to membership termination. Thus:

Finally, we consider causal mechanisms. As noted before, there are two plausible mechanisms that may contribute to institutional racism. First, institutionalization may contribute to path dependence, locking in and perpetuating racist patterns of interaction from earlier time periods. Once established, formal features and rules of international organizations, such as those that govern membership criteria, can be sticky. This mechanism implies a genuine decline in underlying racist attitudes and beliefs. Despite being less racist compared to their predecessors, policymakers are effectively trapped into patterns of racist behavior by institutions established in an earlier era. Hence:

Second, institutional racism may be attributable to persistent racial bias. A decline in open expressions of racism does not necessarily imply a decline in racist beliefs. It may be that policymakers of white-majority states feel normative pressures to adopt antiracist language in public pronouncements but behave similarly to their counterparts from earlier eras. Racist attitudes may persist as implicit biases that subconsciously shape evaluations of racialized countries or officials.Footnote ⁴¹ Institutions in this case are more properly understood as vehicles of power and exclusion rather than sources of path dependence. While the first mechanism suggests a gap between preferences and outcomes (that is, antiracist beliefs cannot be realized due to persistent institutional constraints), the second mechanism suggests a gap between racism professed publicly and as reflected in actual behavior. We expect that persistent racial bias will affect membership from both the supply and demand side, reducing the relative incentives of nonwhite countries to join white-dominated institutions while also diminishing their relative chances of admission subject to expressions of interest. Hence:

We will test H3 against H4 in the empirical section by leveraging several observable implications of each mechanism.

Operationalizing Race

To operationalize race for the purpose of analyzing membership, it is necessary to code race at the country level. We code race dichotomously as “white” and “nonwhite” following scholarship that identifies white supremacy as central to the historical and contemporary role of racism in international relations, along with associated perceptions of nonwhite countries as inherently deficient.Footnote ⁴⁷ We also consider an alternative, nondichotomous coding of race in the empirical analysis later.

We code “white” countries as follows. First, the origins of scientific racism and the distinction between “white” and other races have been traced to Eurocentric ideas that emerged along with Western Imperialism and the international slave trade.Footnote ⁴⁸ We thus code as “white” countries located within the European region according to the UN geoscheme classification. Second, to account for countries outside of Europe that acquired white identities through settlement, we draw on the Ethnic Power Relations (EPR) data set to identify countries characterized by a white-majority demographic and white-dominant access to power throughout the period from 1946 to 2021.Footnote ⁴⁹ Because the EPR data is unavailable prior to 1946, we further conducted a survey of primary and secondary literature on the demographics of these countries. Based on this information, we only code as “white” countries that had white demographic majorities and white political control throughout the entire period covered by our data set, 1816–2014, or since the country’s inception if later than 1816. In addition to Australia, Canada, New Zealand, and the United States, this results in ten Latin American countries being coded as “white.” Full details of the coding are available in Appendix A3.

Our coding emphasizes not only white demographics but also white-dominant access to power. We require white-dominant access to power as international relations is conducted by government elites, and the status of those elites is likely to play a significant role in how countries are racialized. The requirement that white demographics and political control be maintained during 1816–2014 is introduced to avoid sharp disjunctures in the racialized status of countries that are unlikely to match perceptions in reality. A potential disadvantage of this coding scheme is that we may be classifying some countries perceived as white during some periods as nonwhite (for example, South Africa during apartheid). In the empirical analysis, we will conduct a variety of robustness checks to confirm that the results are not sensitive to specific coding decisions.

Raw Data

To evaluate H2a, we first examine the membership composition of IGOs during their year of founding. Inception members in IGOs play a critical role in rule making and institutional design. They often create and lock in outsized privileges for themselves, such as persistent disparities in formal and informal leadership roles and decision-making authority.Footnote ⁵⁰ It is thus informative to examine the relative weight of white countries among inception members in IGOs, and to what extent this has changed over time.

Figure 2 depicts the share of white countries among inception members in all IGOs in the Intergovernmental Organizations Version 3.0 data set.Footnote ⁵¹ Each dot represents an IGO plotted at the founding year (x-axis) with the white share of founding members (y-axis), for example the League of Nations in 1919 (0.70), International Monetary Fund in 1945 (0.61), NATO in 1949 (1.0), United Nations in 1945 (0.61), WTO in 1995 (0.36). The solid line is a loess curve, and the dotted line is the share of white countries in the international system.

FIGURE 2.

White share of member states in intergovernmental organizations at founding

Notes: The figure shows that the share of white countries among inception members in IGOs has consistently exceeded the share of white countries in the international system. Although the shares almost converged in the 1970s, the gap has widened in recent years. The dots indicate the year of founding (x-axis) and white share of members at founding (y-axis) for each intergovernmental organization. The solid line is a loess curve of the same data. The dotted line depicts the white share of countries in the international system for each year.

As the figure shows, most IGOs prior to World War II overrepresented white members among their ranks at inception, even accounting for the large share of white states in the international system. This seems intuitive during a period of imperialism and racist ideologies espoused by leading Western states. During the postwar period, a large number of IGOs were created with nonwhite members, reflecting inclusive principles of United Nations organizations and greater assertiveness of nonwhite countries as decolonization progressed. However, the gap has subsequently widened again. The mean share of white countries among inception members in IGOs founded in the 1970s was 0.33 against a 0.29 share of white countries in the international system, while the same shares in the 2000s were 0.44 against 0.29.

Figure 3 depicts the average annual white share of member states in IGOs. This figure is produced by separately calculating the white share for each IGO in each year and then computing the average annual value across all IGOs. The figure thus accounts for both inception membership and subsequent membership changes. As the figure shows, membership of IGOs has consistently skewed toward white countries. The gap between white share of IGO membership and white share of the international system has persisted during the entire time period depicted.

FIGURE 3.

Average annual white share of member states in intergovernmental organizations, inclusive of membership changes

Notes: This figure depicts the average white share of current membership in IGOs on an annual basis, accounting for both inception members and subsequent membership changes. IGOs have consistently overrepresented white countries throughout the past two centuries.

Membership Changes

To consider H2b and H2c, we examine the evolution of IGO membership after inception. In the domestic political context, institutional approaches to the study of racism have highlighted systematic exclusion of racialized groups from important civic and political institutions.Footnote ⁵² Do IGOs dominated by white states analogously exhibit a bias against potential nonwhite member states? Alternatively, do they exhibit a preference toward remedying their lack of diversity by courting and prioritizing the admission of nonwhite members?

We compute a measure of white overrepresentation by dividing the share of white countries among an IGO’s founding members by the share of white countries in the international system during the founding year, f. The equation is computed as follows:

The measure effectively takes the vertical location of each data point plotted on Figure 2 and divides by the location of the dotted line. The measure will equal 1 if an IGO’s membership is representative of the racial composition of all states in the international system during the year of founding, for example, if all countries are included as members. The measure will be greater than 1 if white countries are overrepresented in the IGO relative to the international system, while it will be less than 1 if white countries are underrepresented.

We then compute an analogous measure of white overrepresentation among all postinception members added to the IGO. For this measure, we first calculate the white share of countries added to the IGO after the IGO’s inception through the end of its life or 2014, the last year of available data, $g$ . This is then divided by the white share of “available” states during the same time period, that is, states that existed in the international system but that were not inception members of the IGO. The equation is thus computed as follows:

Figure 4 plots our measure of white overrepresentation among IGO founding members against white overrepresentation among new members.Footnote ⁵³ There is a remarkably stark association between the two measures: IGOs that overrepresent white countries exhibit a strong tendency to admit new members that are also white countries. Combined with the relatively large number of IGOs that overrepresent white countries at their founding (Figure 2), this is suggestive of considerable structural bias against nonwhite countries in the international system. Rather than diversifying their membership, white-dominated IGOs tend to disproportionately expand their ranks among other white countries.

FIGURE 4.

White-overrepresented IGOs tend to add new white members

Notes: The figure shows that IGOs founded with more white member countries relative to the international system tend to overrepresent white countries among their new members as well. Overrepresentation of white countries among founding members (x-axis) is computed by calculating the share of white countries among IGO founding members (white inception IGO members/total inception IGO members) and then dividing this number by the white share of countries in the international system during the same year (white countries/total countries). Analogously, overrepresentation of white countries among new members is computed by calculating the share of white countries among new IGO members for all years subsequent to founding (white new IGO members/total new IGO members) and then dividing this number by the white share of all potential new members in the years after inception (white potential new members/total potential new members). Values exceeding one indicate overrepresentation of white members, while values below one indicate underrepresentation. If IGOs added new member states at random from the pool of nonmember states in the international system, we would expect the data points to fall on the depicted dotted horizontal line. The dots depict actual values, and the solid line is a loess curve.

Analysis

There are plausible alternative explanations for the apparent affinity of white-dominated IGOs toward new white members. Existing statistical analyses suggest that factors like alliance relationships,Footnote ⁵⁴ democratization,Footnote ⁵⁵ development, trade dependence, regional proximity,Footnote ⁵⁶ regional strategic rivalry,Footnote ⁵⁷ and risk of militarized conflictFootnote ⁵⁸ are associated with joint membership in IGOs. These can be considered potential confounders. For example, white countries are clustered regionally, and many are wealthy, democratic countries that share some degree of geopolitical alignment. They thus plausibly share common interests that are not necessarily related to race.

This race-neutral framing is the conventional wisdom in the study of IGOs. None of the existing studies of IGO membership cited before mentions race as a potential explanatory variable. It may be that any purported association between IGO membership and race is spurious. However, in light of the broader neglect of race in the study of international relations, it is also plausible that variation related to racism has been attributed to closely correlated race-neutral variables. In the domestic context, the hallmark of institutional racism is the justification or rationalization of racially disparate outcomes based on seemingly universalistic, race-neutral principles. Furthermore, it is plausible that some of the potential confounders are themselves endogenous to racism, such as the delineation of regions or alliance relationships according to racial lines. It is nonetheless useful to consider whether the association between white-dominated IGOs and subsequent membership changes favoring white members is attributable entirely to race-neutral variables identified in the existing literature.

To evaluate H2b and H2c, we draw on Davis and Pratt,Footnote ⁵⁹ who assembled a data set drawing on COW IGO Version 3.0 to analyze the determinants of IGO membership, expansion, and exit.Footnote ⁶⁰ All three are binary measures that respectively measure at the state-IGO-year level whether a state in the international system is an existing member, whether a nonmember state joined the IGO, and whether an existing member state exited the IGO. We build on the logistic regression model from Davis and Pratt by adding our time-invariant coding of “white” countries and the Inception Ratio (white overrepresentation in the IGO at the point of founding) to estimate IGO expansion and exit for state i in IGO j and year t:

In the robustness check section, we analogously model the current membership of state i in IGO j and year t as the dependent variable, though current membership is partially attributable to inception membership and should be interpreted with appropriate caution.

We begin by running the model in the postwar period (1949–2014), for which values on all relevant control variables are available. X _ijt-1 controls for other potential determinants of membership outcomes in IGOs that may be correlated with the white status of a country or white overrepresentation in an IGO.Footnote ⁶¹ Several variables are measured at the state level: logged GDP and GDP per capita to account for economic size and development and Trade Openness, measured as the trade-to-GDP ratio;Footnote ⁶² IGO membership, which records the total the number of IGOs that a state is a member of; and Polity, which captures the country’s level of democracy.Footnote ⁶³ At the IGO level, control variables include a dummy Regional IGO that codes IGOs that are regionally defined in their charter or title;Footnote ⁶⁴ a count of the number of states in the IGO;Footnote ⁶⁵ and a dummy Stringent Accession for IGOs that require at least a supermajority vote by existing members to permit new membership.Footnote ⁶⁶ To consider the relationship between each state and the existing IGO members, we control for Average Alliances, operationalized as the proportion of members in IGO j with which state i has an alliance;Footnote ⁶⁷ the number of Fatal Militarized Disputes between state i and members of IGO j;Footnote ⁶⁸ Members from Region, which indicates the number of states in the same geographic region as state i who are members of IGO j;Footnote ⁶⁹ Trade with Members measuring average volume of bilateral trade state i has with members in IGO j;Footnote ⁷⁰ potential effects of colonial history using a variable for Shared Colonial History that measures proportion of members with colonial history under the same colonizer as state i as well as Former Colonizer in IGO that measures the proportion of members who were former colonizers or colonies of state i; and the logged Average Geographic Distance between state i and members of IGO j.Footnote ⁷¹ Last, we consider the effects of time by including a cubic polynomial t for time dependence;Footnote ⁷² and a dummy variable Cold War that codes the period 1947 to 1991 to account for polarized membership of IGOs during this period. Standard errors are adjusted for clustering at the state level to account for correlation in the error term.

The statistical results are presented in Table 1. Each column depicts the coefficients and standard errors respectively for the dependent variables, expansion and exit. Columns 1–2 exclude the control variables, which are included in columns 3–4.Footnote ⁷³ As the results indicate, inclusion of the control variables for the most part does not meaningfully alter the statistical association between the independent variables of interest and the dependent variables, with the exception of the interaction term for the exit variable, which falls short of statistical significance in the model without control variables.

TABLE 1.

Effect on IGO expansion and exit

Note: ${{\rm{\;}}^{\rm{*}}}$ p $ \lt $ .05.

To interpret the results, we calculate the change in predicted probabilities of the dependent variables when increasing the inception ratio by one standard deviation (that is, when an IGO is more white-dominated at inception), based on a simulation of counterfactuals using observed values. Figure 5 depicts the predicted probabilities for the models in column 3–4, that is, including the full set of controls. As the figure shows, there is a statistically meaningful difference between the predicted probability of a white state and nonwhite state joining an IGO according to the share of white members in the IGO at inception. Substantively, a one standard deviation increase in the inception ratio is associated with about a 55 percent increase in the probability of a white state joining the IGO and a corresponding 20 percent decrease in nonwhite states joining the IGO.Footnote ⁷⁴ This is consistent with H2b. In terms of exit, a one standard deviation increase in the inception ratio suppresses the probability of exit by a white state by 33 percent, while the point estimate for nonwhite states is positive but indistinguishable from zero.Footnote ⁷⁵ Combined with the statistically weak results for the baseline model in column 2, support for H2c is suggestive but unstable. This is likely attributable to the relative rarity of exits from IGOs.Footnote ⁷⁶

FIGURE 5.

Predicted probabilities: white-overrepresented IGOs tend to disproportionately add and retain white members

Notes: Even in models that control for a wide variety of other determinants of IGO entry and exit, a one standard deviation increase in the white share of IGO membership at inception is associated with a higher likelihood of expansion involving new white members and lower likelihood of expansion involving new nonwhite members. White members are also less likely to exit from IGOs with a high inception ratio of white states. On the other hand, the inception ratio is not meaningfully associated with changes in likelihood of exit by nonwhite states.

Robustness Checks

We conducted a variety of robustness checks. We reran the models using current membership status in IGOs as the dependent variable in place of expansion and exit. Membership at any given time includes inception members and is thus a less precise measure of bias, but the results are consistent with our main analysis: white countries are relatively more likely to be members of IGOs with high inception white shares (Table 2). We also ran an IGO-level analysis that strips out time and regresses the inception ratio against the ratio of white states among all subsequent new members for each IGO along with IGO-level control variables. Consistent with the main analysis, the results show a strong association between the inception ratio and the white ratio of subsequent members (Appendix B2).

TABLE 2.

Effect on IGO membership

Note: p $ \lt $ .05.

Due to data availability, the main model including the full set of controls can only be run on the post-World War II period. To consider a longer time horizon, we reran the models while limiting control variables to those available for the entire period of available data (1817–2014; Appendix B3). The results are substantively similar to those reported for the postwar period for entry and membership, while the results for exit are sensitive to specification. As a further robustness check, we also reran models using a logistic regression with bias correction to account for the rare occurrence of accession and exit in the data (Appendix B4, B5) and found substantively similar results.Footnote ⁷⁷

To consider the sensitivity of our results to how we coded white states, we ran the analysis with different coding of ambiguous cases as well as an alternative measure of race from the existing literature. First, we excluded Latin American states, such that only European countries and Australia, Canada, New Zealand, and the United States are coded as white. Second, to consider countries characterized by white-minority rule, such as apartheid South Africa, we also ran a separate analysis recoding as “white” any states characterized by at least one instance of white dominant access to power.Footnote ⁷⁸ The substantive results were similar (Appendix B6, B7). We also reran the analysis using an independently coded country-level categorization of race based on visible minority status classifications in Canada.Footnote ⁷⁹ This yields similar results (Appendix B8). This data also allows for the disaggregation of nonwhite countries into specific racial categories. Rerunning the models on this disaggregated data, we find that higher inception ratios are only associated with higher rates of IGO accession for white states. Among nonwhite states, higher inception ratios are associated with particularly low probabilities of accession by Black, Central Asian, South Asian, and Southeast Asian states (Appendix B9).

It is possible that our results are biased by patterns of regionalism, such as the concentration of white countries and regional institution building in postwar Europe. We would note that boundaries of regions are themselves artificial and malleable constructs, which may be shaped by race. Boundary-breaking membership patterns among regional organizations are themselves informative about what kinds of states receive exceptional treatment.Footnote ⁸⁰ However, it is useful to consider whether our results depend on the inclusion of IGOs with geographically targeted membership. First, we considered whether regional IGO acts as a moderator by running a three-way interactive model between the regional IGO dummy, inception ratio, and white country status. The results suggest that the substantive patterns of expansion and exit are similar for regional and nonregional institutions (Appendix B10). We then reran the analyses while excluding regional IGOs, European IGOs, and commodity IGOs.Footnote ⁸¹ We also reran the models while constraining the potential pool of new members vis-à-vis regional institutions to: (1) states sharing the same region as founding members, (2) states sharing the same region or located in a neighboring region of the founding members. These results were similar with one exception: the coefficient for nonwhite expansion for nonregional IGOs was statistically indistinguishable from zero. However, there was still a clear and large difference in the relative rates of expansion for white and nonwhite states (Appendix B11, B12, B13, B14).

The COW data set includes a variety of IGOs ranging from universalistic organizations with large portfolios like the United Nations to relatively inactive and obscure organizations with small memberships. We therefore reran the analysis by limiting the sample to IGOs identified as important in Hooghe and colleagues’ Measuring International Authority data set.Footnote ⁸² There are seventy-six such IGOs, of which seventy-four are present in the COW data set. Limiting the data to this set of IGOs shows a consistent pattern of IGOs with high inception ratios disproportionately expanding membership among white countries. On the other hand, the estimates for exit do not exhibit a meaningful difference between white and nonwhite countries (Appendix B15).

Despite these robustness checks, the impossibility of randomizing the assignment of race raises intractable challenges of causal identification. We thus supplement our analyses with Imai and Lo’s nonparametric sensitivity analysis, a method originally proposed to assess the effect of democracy on peace, which raises similar problems of nonrandomization.Footnote ⁸³ This approach offers an indication of the strength of any potential confounder required to reject the hypothesized relationship and how proposed confounders perform against this standard. We ran the analysis on combinations of Inception Ratio and eleven confounders with statistically meaningful effects in the regression models (Appendix B16). The analysis calls for dichotomized variables. However, unlike democracy, which is conventionally defined as a polity score greater than +6, there is no accepted cutoff for the Inception Ratio. We thus ran dichotomized Inception Ratio separately at its 50th-, 70th-, and 90th-percentile values. The 50th percentile value includes many IGOs with near-equal representation, while the higher percentile values only include IGOs with high degrees of white overrepresentation. Based on the results, there were only two instances that met the threshold for a confounder: Members from Region and Shared Colonial History when the Inception Ratio was dichotomized at the 50th percentile for expansion. When the Inception Ratio was dichotomized to include only relative high degrees of white overrepresentation, none of the variables met the threshold for either expansion or exit. Notably, for an Inception Ratio in the 90th percentile, the relative risk of expansion to a new white member is about 27, which exceeds the relative risk of smoking to lung cancer.Footnote ⁸⁴

Overall, the robustness checks indicate a pattern similar to the main analysis: IGOs that overrepresent white countries at inception tend to attract new white members at disproportionate rates, even after accounting for a variety of potential confounders. The results for exit are suggestive but more sensitive to specification.

Formal and Informal IGOs

H3 posits that path dependence stemming from formal features of IGOs is responsible for the membership patterns we observe. If so, informal IGOs—which inherently lack formal rules—should be less subject to bias. To evaluate this, we reran our analysis on Roger and Rowan’s data set of 260 informal organizations in lieu of formal IGOs.Footnote ⁸⁵ We calculate the change in predicted probabilities of white and nonwhite states joining informal IGOs when the inception ratio increases one standard deviation from the mean, based on a simulation of counterfactuals using observed values (Appendix B17). Holding other things constant, a one standard deviation increase in the inception ratio is associated with an increase in the probability of a white state joining of about 83 percent, whereas the same probability for a nonwhite state decreases by 28 percent. The findings are largely in line with our main results that examine formal IGOs. This suggests that, at a minimum, formal rules are not a necessary condition for the biased patterns of membership we observe in the main analysis.

Stringent Accession Rules

The preceding analysis demonstrated that patterns of membership bias appear to persist even among institutions with no formal rules. We next examine whether the substance of formal rules has an impact on bias. To do so, we use Davis and Pratt’s coding of Stringent Accession, which considers “organizations that require a supermajority or unanimous consent of existing member states to admit new members, according to the founding charter.”Footnote ⁸⁶ Decision making about membership expansion is more likely to exhibit path dependence among IGOs with stringent accession rules. In such IGOs, a relatively small number of countries can block accession, allowing a minority with extreme preferences to determine outcomes. This is akin to the use of supermajority filibuster rules by Southern senators in the United States to veto racial equality legislation prior to the Civil Rights Act of 1964. On the other hand, if patterns of bias are invariant to accession rules, it would be suggestive of broader bias among a large share of IGO members.

We reran the main analysis separately for IGOs with stringent and nonstringent accession rules (Appendix B18). The changes in predicted probabilities associated with a one standard deviation increase in the inception ratio are greater in the case of IGOs with stringent rules for accession: there is a 71 percent increase in the predicted probability of white states joining IGOs with stringent rules, compared to a 36 percent increase for IGOs with nonstringent rules. For nonwhite states, the same increase in the inception ratio is associated with a 24 percent decrease in joining an IGO with stringent rules and 14 percent decrease for an IGO with nonstringent rules.

These results suggest that stringent accession rules are associated with stronger patterns of bias, which is supportive of H3. However, IGOs with nonstringent rules also exhibit similar—albeit weaker—bias. This is supportive of H4: even in IGOs where membership expansion cannot be vetoed by a small group of present members, white states are disproportionately preferred by white-dominated IGOs. The evidence from accession rules is thus consistent with both mechanisms.Footnote ⁸⁷

Over-Time Variation

If underlying racism has genuinely diminished—an important premise of H3—we would expect racial bias in membership to be less pronounced for IGOs founded in recent years. Members of IGOs created prior to World War II may be constrained by rules or practices carried over from a more racist era, but IGOs founded in recent years should not constrain their members in the same way. On the other hand, H4 is predicated on the persistence of racist attitudes and behavior, and we would expect less variation based on founding year.

We thus examine membership patterns for IGOs founded in different time periods. We separated IGOs into six groups according to their founding years and ran an IGO-level analysis (Appendix B19).Footnote ⁸⁸ If underlying racism has diminished over time consistent with H3, we would expect the relationship between the inception ratio and new member ratio to be largest in the first period and diminish for IGOs founded more recently. However, our analysis shows that the relationship remains similar for IGOs established across time periods. This is more consistent with H4 rather than H3: new IGOs do not exhibit less biased patterns of membership expansion despite being created in a “less racist” era.

IGO Language and Membership

As Figure 1 illustrates, many IGOs now include antiracist language in their founding documents. Are such organizations less prone to bias in their membership patterns? H3 posits that policymakers today are genuinely less racist, but their behavior is constrained by institutional rules. If so, institutions that explicitly adopt antiracist language may exhibit less racist behavior for several reasons. First, antiracist language in founding documents may serve as a formal constraint against discriminatory practices. For example, the rules could be invoked by aspiring members or current members that support nondiscrimination to facilitate a more diverse membership. Second, there may be a selection effect, that is, antiracist language is more likely written into IGO charters by inception members that have a genuine preference to avoid discrimination on account of race. On the other hand, H4 posits a gap between language and action—despite normative taboos against racist language, attitudes and behavior remain largely unchanged. If so, antiracist language may be better interpreted as performative cheap talk that bears little relationship to behavior and outcomes.

To evaluate the relationship between language, race, and membership, we ran a modified version of our main model with an additional variable based on our coding of racist language as described earlier, which takes on three possible values: -1 (antiracist); 0 (neutral); 1 (racist). This variable is interacted with the time-invariant coding of white countries and the inception ratio. All subinteractions and control variables from the main analysis are included in the model. If antiracist language is associated with less discriminatory membership outcomes, we would expect the point estimates for the predicted probabilities for antiracist language to be closer to zero compared to those for racist and neutral language. The findings indicate that this is not the case: there is no meaningful difference in the membership patterns of IGOs using antiracist language compared to IGOs using racist language (Appendix B20). We also ran alternative specifications that use separate models for each IGO language category in lieu of interactions. The results are substantively similar (Appendix B21).Footnote ⁸⁹

Discussion

Overall, these results are more consistent with H4 than H3. If membership bias is primarily attributable to the constraints of path-dependent formal rules, we should observe less bias where such rules are less binding. In most of our tests, this is not the case: informal IGOs, formal IGOs established more recently, and formal IGOs that profess antiracism do not diverge meaningfully from the patterns identified in the main analysis. H3 receives some support based on our analysis of stringent accession rules: bias appears to be stronger for IGOs where accession requires a supermajority or consensus support. However, IGOs with nonstringent IGOs also exhibit bias, which is consistent with H4. These findings are summarized in Table 3.

TABLE 3.

Mechanism tests

Note: H4 finds consistent support, while H3 is only supported in the case of stringent accession rules. See the appendix for details.

Gains from Cooperation

First, the overrepresentation of white countries among the ranks of IGO members implies asymmetries in access to the gains from cooperation. This is perhaps the most obvious and intuitive mechanism that connects racial bias in membership to policy outcomes. Foundational theories of neoliberal institutionalism posit that institutions facilitate cooperation by allowing states to overcome barriers like the absence of rules, transaction costs, and information asymmetries.Footnote ¹¹¹ By implication, the relative disparity in membership among white and nonwhite states implies that the latter have enjoyed fewer opportunities to secure gains from cooperation.

More practically speaking, many international organizations restrict access to benefits according to membership status. Security institutions generally limit guarantees of mutual assistance to member states.Footnote ¹¹² International financial institutions like UN aid agencies and the World Bank tie lending to membership.Footnote ¹¹³ Changes in membership status, such as through accession or expulsion, have been shown to significantly alter state behavior in a variety of issue areas.Footnote ¹¹⁴ Aside from access to direct benefits, institutional membership can confer voting power and representation that states can leverage as bargaining chips to secure quid pro quo deals.Footnote ¹¹⁵ Ceteris paribus, the underrepresentation of nonwhite states among IGO members implies relatively less access to such benefits.

Segregation and the Quality of Cooperation

Second, outcomes may be further distorted through membership segregation and variation in institutional quality. After the enactment of the Fourteenth Amendment in the United States, racial segregation was defended based on the “separate but equal” doctrine. An analogous rationalization could be offered for racial bias in IGO membership: even if IGOs dominated by white states tend to disproportionately attract other white states as new members, nonwhite states may not be at a great disadvantage if they form or join equally effective IGOs. On the other hand, membership disproportionality is more problematic if IGOs are “separate and unequal,” with more resources and activities concentrated in white-dominated institutions. If disparities in membership patterns are also associated with disparities in access to effective IGOs, it will tend to reinforce racial hierarchy.

To consider this question, we examine the relationship between racial disproportionality in IGO founding membership and the vitality of organizations.Footnote ¹¹⁷ Gray’s data set codes the vitality of seventy economic international organizations, where she defines vitality as “regular convocation and a basic level of output consistent with their goals.”Footnote ¹¹⁸ Of the seventy organizations, thirty-eight correspond to IGOs in the COW data set. A cross-tabulation of the inception ratio and vitality of these thirty-eight IGOs can be found in Table 4. As the table shows, a large majority of white-overrepresented IGOs fall into the “Life” category, while white-underrepresented IGOs tend to be “Zombie” organizations or “Dead.” Although this data covers only a subset of institutions, it is more consistent with a “separate and unequal” configuration, in which white-dominated institutions are more active and effective compared to nonwhite-dominated institutions. By implication, nonwhite states tend to suffer not only from relatively fewer IGO memberships, but also from lower quality of cooperation in the IGOs that they do join.

TABLE 4.

Inception ratios and IGO vitality

Notes: Average vitality is calculated from Gray’s binary coding of “life,” coded based on the occurrence of at least one meeting a year and higher than expected trade output predicted by a standard gravity trade model.Footnote ¹¹⁶ We collapse Gray’s two categories of low vitality, “zombie” and “death,” into a single category due to paucity of data in the “death” category.

IGO Decision Making

Finally, distortions in membership can affect policy outcomes by shaping the decision making of IGOs. We would caution that attributing specific policy outcomes to institutional racism is challenging for the same reason it is difficult to attribute individual hiring decisions or police shootings to racism. By definition, institutional racism is nonovert and occurs in the absence of clear statements of racist intent. It is also extremely difficult to rule out confounding factors without examining broader patterns in the aggregated data as we did in the analysis earlier. Here, we provide illustrative examples in which white-dominated IGOs have been criticized for biased decision making and policies, drawing on public statements and analyses by scholars, public officials, and nongovernmental organizations.

Some of the most white-dominated IGOs within our data set play long-standing roles in global rulemaking and agenda setting, which have wide-ranging effects that extend beyond member countries. This has led to accusations of racial bias and neocolonialism. Institutions like the International Institute for the Unification of Private Law (UNIDROIT) and Hague Conference on Private International Law (HCCH) have predominantly white memberships but nonetheless play pivotal roles in promulgating legal principles with global scope. This has led to recurrent criticism that the institutions privilege Eurocentric biases while overseeing rulemaking over issues with deep-seated racial and colonial legacies, such as international adoptions and stolen cultural artifacts.Footnote ¹¹⁹ For example, the UNIDROIT 1995 Convention on Stolen or Illegally Exported Cultural Objects was written to be explicitly nonretroactive and has been criticized for the “‘white-washing’ of looted or illegitimately acquired colonial objects through auction houses.”Footnote ¹²⁰ The Global Health Security Initiative (GHSI), an institution comprised of 75 percent white-majority states, has similarly been criticized for conceptualizing health security narrowly in terms of “the protection of developed Western states from exogenous disease threats”Footnote ¹²¹ and allocating resources away from Global South countries most severely affected by health crises.

The Organisation for Economic Co-operation and Development (OECD) has also been frequently criticized for seeking to impose the prerogatives of its membership—which is composed of about 85 percent white countries—on global rules and standards. The organization’s Base Erosion and Profit Shifting (BEPS) initiative, which seeks to combat tax avoidance, has been formally criticized by UN experts—including the Special Rapporteur on Contemporary Forms of Racism, Racial Discrimination, Xenophobia and Related Intolerance—who stated in a public letter to the OECD that, “In reifying patterns of economic extraction with historical origins in systems of colonialism and slavery, the deal has the potential to prejudice the predominantly nonwhite nations of the Global South.”Footnote ¹²² Nongovernmental organizations like the Tax Justice Network and Center for Economic and Social Rights point out that the OECD has turned a blind eye toward tax haven arrangements among its own members like Ireland, Switzerland, and the United Kingdom while explicitly blacklisting and targeting similar practices among countries in the Global South.Footnote ¹²³ Other OECD initiatives, such as the Programme for International Student Assessment (PISA) and the Learning Compass 2030, have also been accused of “educational colonialism” and seeking to impose Western norms through one-size-fits-all approaches that neglect local culture and practices.Footnote ¹²⁴

Another organization that has been frequently criticized for its white-dominated membership and biased priorities is the Union for the Protection of New Varieties of Plants (UPOV). UPOV was founded by the United Kingdom, the Netherlands, and Germany in 1961. UPOV subsequently expanded its membership but notoriously included no members from the Global South except apartheid South Africa when it negotiated its most recent convention in 1991. As of 27 February 2025, UPOV membership consisted of seventy-eight states and two regional organizations, with a 65.2 percent share of white countries.Footnote ¹²⁵ UPOV is part of the regime complex for plant genetic resources,Footnote ¹²⁶ and its decision making is widely perceived as skewing in favor of white-majority states, which make up the lion’s share of its membership. The organization’s prioritization of breeders’ rights over farmers’ rights has been the subject of intense criticism as a form of “seed colonialism.”Footnote ¹²⁷ The UPOV convention is widely seen as privileging Western multinational firms at the expense of smallholder and Indigenous farmers in the Global South.Footnote ¹²⁸ Despite these concerns of bias, Global South countries have often been compelled to adopt the UPOV Convention due to the inclusion of intellectual property terms in trade agreements with European countries and the United States.Footnote ¹²⁹